AMD responds to Radeon RX 480 power draw controversy

AMD's Radeon RX 480 offers compelling performance for its price, but some reviewers have discovered a potentially troubling behavior from the card's power-delivery subsystem. Controversy arose when Tom's Hardware and PC Perspective discovered that the reference RX 480 is drawing more power than might be considered reasonable from the PCI Express slot's 12V rail.

The fine folks over at PC Perspective are especially insightful on this point. The site notes that the PCI Express specification limits connected boards to drawing no more than 5.5 amps over the 12V connection from the slot itself. The reference version of that card pulls 7A of current from the slot on average, while overclocking increases that current draw to 8.3A.

PC Perspective contacted some internal sources at motherboard manufacturers to determine whether this behavior—especially that of the overclocked card—would be an issue. The site learned that while the traces on the board likely wouldn't be affected, the pins and connectors on lower-cost motherboards might be harmed over time by sustained operation under the highest loads from the overclocked card.

AMD has directed concerned users and the press to the following statement on Anandtech (also provided to PC Perspective):

As you know, we continuously tune our GPUs in order to maximize their performance within their given power envelopes and the speed of the memory interface, which in this case is an unprecedented 8Gbps for GDDR5. Recently, we identified select scenarios where the tuning of some RX 480 boards was not optimal. Fortunately, we can adjust the GPU's tuning via software in order to resolve this issue. We are already testing a driver that implements a fix, and we will provide an update to the community on our progress on Tuesday (July 5, 2016).

Since we don't have the kind of in-depth power measurement hardware that PC Perspective does, we can't independently confirm this issue or its effects. The site does note that it hasn't experienced any negative side effects from running a single RX 480 in its test rigs, though, and we didn't note any weirdness in our test systems, either. 

Our original conclusion to our RX 480 review was to wait for custom versions of that card to begin shipping because of noise and temperature concerns, and we think that advice is even more relevant now for folks who might be concerned about the RX 480's power draw. PCPer suggests this issue could be avoided entirely with a single eight-pin connector or a pair of six-pin connectors, and we'd expect most custom boards to feature that kind of power-delivery setup. For now, we'll see what Tuesday brings.

Comments closed
    • smilingcrow
    • 3 years ago

    [url<]http://www.anandtech.com/show/10469/amds-tuesday-radeon-rx-480-update-new-driver-by-late-thursday[/url<]

      • tipoo
      • 3 years ago

      Seems like the fix and lower power toggle all but confirms a last minute overclock. They also say the 3% boost in games the new driver provides should substantially offset any impact from the lower energy setting, so it seems pretty sure to me they got worried and pushed it past what they found was it’s peak efficiency point, or at least the better tradeoff of efficiency to performance (since power use scales more than linearly with voltage+frequency, while performance scales less than linearly with clock), in favor of boosting it a bit more.

      All that debacle didn’t seem worth it if it’s 3-5% they squeezed out.

    • Klimax
    • 3 years ago

    Reminder: There were rumors about AMD having trouble with hitting clock target with Polaris. They may have been true…

    • Pancake
    • 3 years ago

    The speculation is that GTX1060 will be released tomorrow or the day after. That’s going to suck up all the media attention as it rightly should. My guess is that AMD will defer any resolution to powergate for a few weeks as it may be detrimental to card performance right at the very time it will be held up for comparison against GTX1060. You’d be mad to buy RX480 today.

    • tipoo
    • 3 years ago

    So, when along this fine Tuesday July 5th are we expecting this?

      • Leader952
      • 3 years ago

      Maybe their fix still has problems?

        • tipoo
        • 3 years ago

        But all today was supposed to be was a statement on the fix, how they would do it, even if it wasn’t going out today. If they’re still stalling on *that*…

      • Pancake
      • 3 years ago

      *crickets*

      • smilingcrow
      • 3 years ago

      It’s only meant to be an update on the progress not a fix so if they can’t even manage that…

      • tipoo
      • 3 years ago

      The silence is making me picture people running around in a panic in a lab with the counter on fire and notes flying everywhere

    • Leader952
    • 3 years ago

    AMD Radeon RX 480 Power Measurements Repeated And Clarified

    [url<]http://www.tomshardware.com/reviews/amd-radeon-rx-480-power-measurements,4622.html[/url<] [quote<]Even if the power consumption of the memory module and other electronics was reduced below realistic levels, there’s just no way to equally distribute the load with three phases! In the end, we're not sure if it is really a physical 4:2 split, or only done with the firmware to change the balance in the direction of the PCIe slot, but the result is exactly the same. It’s just not possible to achieve 75-80W for the PCIe slot with two or three phases, no matter how many people have stated this over the last few days. Two phases would result in 40-41W, and three phases would result in 60-62W. [/quote<]

    • flip-mode
    • 3 years ago

    Seriously, though, once 3rd party cards come out with none of these power draw issues, the RX 480 is back to being an awesome card for the money.

      • rudimentary_lathe
      • 3 years ago

      Unless those AIB cards are priced closer to $300 than $239.

        • flip-mode
        • 3 years ago

        And other obvious things like the GTX 1060 launching at an even better price/performance.

      • Leader952
      • 3 years ago

      Unless those 3rd party cards still follow AMD’s reference with 4 phases on the 12V PCI-e connector and still pull more than 66 watts from it.

      AMD Radeon RX 480 Power Measurements Repeated And Clarified

      [url<]http://www.tomshardware.com/reviews/amd-radeon-rx-480-power-measurements,4622.html[/url<]

        • flip-mode
        • 3 years ago

        I hope none of them are that daft.

    • Farting Bob
    • 3 years ago

    Ive RMA’d my 480X, i may get a custom board when they come out if they have 8 pins, but at that point the 1060 might be around and if its better i may just go green this time.

      • Anonymous Coward
      • 3 years ago

      You’re a busy guy, getting one of those ordered at the first instant, and then returning it at the first instant.

    • maxxcool
    • 3 years ago

    Hey Jeff, I know it is a bit of a ask… but can you buy a fresh in-box 480 card in about 2 weeks and try both the drivers and bios that shipped with your ‘test’ card and the ‘retail bios and adjusted drivers’ that will fix this ?

    if there is a significant performance difference then we have a much better story πŸ™‚

    I am a HUGE Nvidia guy, but i remember 750 cards having issues with the older process and OC oem boards. A fair shake should be given.

    • deruberhanyok
    • 3 years ago

    This may point us toward their solution:

    [url<]http://wccftech.com/article/radeon-rx-480-reducing-voltage-increasing-efficiency/[/url<] Short version: undervolting without sacrificing performance can be done, and at least in one case it seems to show a pretty decent drop in power usage.

      • Firestarter
      • 3 years ago

      that’s true for just about every GPU and CPU and it works the same way overclocking does: you reduce the stability margin to increase performance. If AMD could have done this with all RX480 cards they shipped without introducing instability, they would have done so already.

      If you want to know what happens when companies try to do this anyway, look no further than what happened when Intel first released the 1.13GHz Pentium III in July 2000, it proved to be unstable and Intel had to recall it

        • deruberhanyok
        • 3 years ago

        Normally I’d agree, but isn’t that basically what AMD is saying they’re going to do? Fix it for all of the cards that have already been shipped without introducing instability?

          • Firestarter
          • 3 years ago

          yes, which is why I’m skeptical as to their ability to pull this off

      • tipoo
      • 3 years ago

      I think the solution will be in the programmable VRMs. The VRMs in the 480 are ridiculously overbuilt if you look at the specs, they could supply hundreds of amps or something like that, its was overkill for the Fury X already and they brought it down to a 200 dollar part.

      So they could disable two phases and run the ones from the 6 pin at a higher power, it would still violate the 6 pin spec but those are also overbuilt, better to do that than violate PCI-E through a motherboard.

      [url<]https://forum.beyond3d.com/threads/amd-speculation-rumors-and-discussion.56719/page-196#post-1928122[/url<]

        • Waco
        • 3 years ago

        150 amps at 1 volt is only 150 watts, remember that. πŸ™‚

      • Mat3
      • 3 years ago

      I don’t think undervolting will be their solution to this problem, but it sounds like they’ve volted their cards too high to get more 480s out there. They could have made their power consumption look a lot better if they had lowered voltage to begin with and put more of the crappy but still functional chips into the lower clocked 470 version. I’ve read plenty lately to suggest that their previous cards could have looked much better if they had done this as well.

    • ronch
    • 3 years ago

    Some folks say people are blowing this issue out of proportion. To those folks, I say, grab one of these cards and plug them into your system and see if you’re OK with that.

    Some folks say this is real. So if you’ve been wanting to get Polaris I think it’s good advise to just wait for the 8-pin versions to come out. Or maybe see what GTX 1060 brings to the table first.

      • Anonymous Coward
      • 3 years ago

      Meh, lots of regular people will be plugging these into their systems, no doubt they’ll sell out.

    • TheRazorsEdge
    • 3 years ago

    Lower the clocks and voltage, and make up the performance with driver improvements.

    As long as they pull that off, I don’t see much room for complaints.

    Any drop in performance with this fix, though, and I’m calling shenanigans.

    • Sahrin
    • 3 years ago

    Looks like they fired one engineer too many.

      • torquer
      • 3 years ago

      Looks like you made one post too many.

    • Sahrin
    • 3 years ago

    Looks like they fired one engineer too many.

      • arbiter9605
      • 3 years ago

      i don’t think 8gb one was last minute thing, more likely the clocks were last minute change after pascal performance hit them in the stomach

      • tipoo
      • 3 years ago

      “So what do you do?”
      “Power delivery”
      “Just connect wires to other wires? We can do that! You’re fired!”
      -Some MBA, probably

      • HisDivineOrder
      • 3 years ago

      “Hey Bob, come on in. I’m glad you came so quickly.”

      “Sir, I got your email. I’m here. I needed to tell you something important about the PCIe power draw of our card. It’s–”

      “Ahem. Shhhh, Bob. That can wait. First, I want to tell you that you’ve done a LOT of great work here at AMD and I have really appreciated it. I mean, truly, it’s been great.”

      “Wow. Thank you, sir. I thought you didn’t notice. I mean, I know I was way down the hall past all the empty cubicles and offices, past the other set of empty cubicles and offices, way past the third set of empty cubicles and offices, down next to the restroom. I thought you’d totally forgotten about me.”

      “No, Bob. No, uh, I couldn’t forget about you! So, Bob.”

      “Oh, wow, sir. I thought you’d called me here to fire me. Everyone in my whole team that had been working the PCIe power delivery spec was fired over the last few years. I figured my number was up! Whew, thanks for just being a swell boss.”

      “Yeah. About that. I just want you to know how great it’s been having you here.”

      “Wait? What?” He turns. “Why are there so many security guards here?”

      “I’m afraid that while I did not forget you–I really didn’t, I promise–I do have to finish cleaning out the floor. You were the last holdout. I couldn’t FIND you. You were hidden back near the restroom and I had a devil of a time figuring out where you were. You understand.”

      “…yes.”

      “These fine gentlemen already have your stuff all packed up in a few tidy boxes. They’re ready to walk you out.”

      “…okay.”

      “So, Bob? Before you go? Didn’t you have something to tell me about the PCIe power delivery of Polaris 10?”

      “Yes, sir. It’s perfectly on spec. No problems at all.”

      “Great! Gentlemen! Oh, and gentleladies. Show him the door. Thanks for all the hard work, Bob!”

        • bill94el
        • 3 years ago

        If they took his red Swingline stapler that building may be on fire soon.

    • Sam125
    • 3 years ago

    Well obviously the RX 480 was designed to be a 4GB video card, duh.

    • tipoo
    • 3 years ago

    On the upswing of things, some of the 4GB cards actually have 8GB, just limited down in BIOS. Interesting. I wonder if this’ll be like the tri-cores turned quad cores on the AMD X3s, and if the power fix improves things, could represent quite the value.

    [url<]http://www.neogaf.com/forum/showpost.php?p=209038156&postcount=1612[/url<]

      • anotherengineer
      • 3 years ago

      Wow so if that is indeed true, AMD is eating the cost of 4GB of ram then? Also if you can unlock with a BIOS update that means you can get an 8GB card for $200 bucks!!

        • tipoo
        • 3 years ago

        It could be that some modules on the 4GB ones are defective? Or maybe just very few 8GB were converted to make up for 4GB demand. We’ll have to see more sample size

        [url<]https://www.techpowerup.com/223913/amd-retail-radeon-rx-480-4gb-to-8gb-memory-unlock-mod-works-we-benchmarked[/url<]

      • cygnus1
      • 3 years ago

      That was clarified that only review boards were sent with a firmware that disabled half the ram. They said all retail units have either 4GB or 8GB and its all 100% enables.

        • tipoo
        • 3 years ago

        No, this is about retail boards, TPU tested and benchmarked:

        [url<]https://www.techpowerup.com/223913/amd-retail-radeon-rx-480-4gb-to-8gb-memory-unlock-mod-works-we-benchmarked[/url<]

    • DPete27
    • 3 years ago

    Could TR use one of those Corsair AXi PSUs to monitor power draw more accurately? Seems like it might be a “cheap” improvement.

    • DPete27
    • 3 years ago

    [quote<]an unprecedented 8Gbps for GDDR5[/quote<] Careful AMD, "unprecedented" means you're the first to do it, however the GTX 1070 has 8Gbps GDDR5 and it launched before the RX480

      • anotherengineer
      • 3 years ago

      Maybe they meant “unprecedented” for a ~$230 card?

      • arbiter9605
      • 3 years ago

      really want to get technical the 1080 has 10ghz effective clocked memory so.

        • DancinJack
        • 3 years ago

        except it’s GDDR5X…

        If we’re going to get technical, let’s at least be correct.

      • tipoo
      • 3 years ago

      Unprecedentedly PCI-E overdrawing

      • travbrad
      • 3 years ago

      I’m actually pretty impressed. That is almost an Apple level RDF response to a problem. “Our cards draw too much power over pcie because they are so fast!”

      I’m surprised they didn’t mention how thin it is too.

        • tipoo
        • 3 years ago

        “It’ll almost certainly be smaller than Polaris 10’s 232-mmΒ² die, and AMD further notes that this chip is its thinnest GPU ever.”

        [url<]https://techreport.com/review/30328/amd-radeon-rx-480-graphics-card-reviewed[/url<]

        • BurntMyBacon
        • 3 years ago

        Interesting perspective. Is this what Apple would look like if its bank account was empty?

          • tipoo
          • 3 years ago

          They’ve been there or lower in 1997, not sure about who works out to be on top after inflation calculations

    • UberGerbil
    • 3 years ago

    It’s the first NPU — Nietzsche Processing Unit. [i<]That which does not kill your motherboard makes it stronger.[/i<]

    • Asif1924
    • 3 years ago

    RX480 is Fermi all over again (remember the power draw of the GTX 480). Bad naming AMD; Fermigate haunts the RX480 now!

    In any case, who cares. There are better mid-range cards than the RX480 in the GeForce stack. And nVidia has better drivers and a better ecosystem anyway.

      • Krogoth
      • 3 years ago

      RX 480 issues have nothing in common with Fermi (GF100).

      It is more like reference R9 290X debacle. The reference design cards weren’t up to the task for handling the hardware they were driving. In the 290X it was thermal issues with VRMs because stock HSF didn’t cut it. In the RX 480’s case it is that power circuitry. The reference boards was meant to for a 150W TDP not 225W TDP that is [b<]needed[/b<] for RX 480's current clockspeed and memory capacity.

        • blahsaysblah
        • 3 years ago

        Except it has way more than enough actual hardware on the board to handle way more than 225W? WTF??

      • JumpingJack
      • 3 years ago

      I would not go so far as to say it is Fermi revisted — they do not draw that much power overall– AMD has priced these cards to move and they are hitting GTX 970 performance at 2/3’s the cost. The only thing a buyer must be wary of is pairing this card with an el cheapo motherboard that skimps on Cu in the traces and cannot handle even a few milliamps over the PCIe 3.0 spec.

    • jokinin
    • 3 years ago

    There is already a video on youtube showing a custom cooled Sapphire RX480 with a 8 pin pci-e power connector instead of the reference 6 pin. I guess, I will follow your advice (which is always good, by the way), and wait for custom cooled, custom designed RX480’s to become available before getting one.

    • chuckula
    • 3 years ago

    Man, I step away from smartphone territory for one day and look what happens!

    The initial post in this forum thread lists some more links to review sites that did direct power measurement and observed the out of spec behavior: [url<]https://techreport.com/forums/viewtopic.php?f=3&t=118144[/url<] As for the driver fix, I'm 100% certain that AMD can adjust the behavior of the card to operate within spec. What I'm not 100% certain of is whether AMD can adjust the behavior of the card to operate within spec and not take a performance hit in the process.

    • tootercomputer
    • 3 years ago

    Has anyone compared a variety of video cards X a variety of motherboards X a variety of metrics to see if the 480 findings are indeed unique? I was impressed (as always) by PC Per’s Ryan Shroud, but my non-engineer mind was (a) not always able to follow what he and his colleague presented on that YouTube video and (b) not clear about whether their testint methods sufficient to the questions a hand.

    Hmm, must be a 3 day weekend, as I had the time and chose to watch the PC Per video a second time and looked over their article. They make a pretty complling case that this is a legitimate flaw that could lead to problems , at the very least, automoatic motherboard safety power shutdowns. That is a very interesting hypothesis that now can be tested: as these cards are purchased and used, will such shutdowns be reported?

    • Chrispy_
    • 3 years ago

    If it’s a firmware fix, and the fix is 100% verified as working before the vendor cards start to appear, then this is a non-issue.

    Cheap-ass cooler and probable last-minute overclock (I reckon Polaris got a 66MHz overclock and overvolt at the last minute).

    A mediocre cooler on a $200-240 card is probably still better than Nvidia charging $100 for the cooler alone.

      • smilingcrow
      • 3 years ago

      This has nothing to do with the cooler.

        • Chrispy_
        • 3 years ago

        Indirectly, it does; Better cooling = lower temperatures = higher efficiency = lower power draw.

        Ignoring transient spikes, the reports are saying that it uses 95W from the motherboard slot, and that could easily be enough that a slower-clocked, lower-voltage, better cooled card would not exceed 75W.

        AMD officially released a statement saying that the “The RX 480 has passed PCIe compliance testing with PCI-SIG” so the increase in power draw from the slot could easily be a result of the last minute clock changes affecting voltage, temperature and power use. Those three things are intrinsically linked and the cooler plays a non-trivial role.

          • tootercomputer
          • 3 years ago

          Your logic is a off here.

            • Waco
            • 3 years ago

            Not really. Lower temperatures actually reduce power draw substantially.

            I dropped 100+ watts on average(no, not a typo) off of my 4870X2 Quadfire rig by watercooling them. Zero other changes.

            • Chrispy_
            • 3 years ago

            Please enlighten me.

            • tootercomputer
            • 3 years ago

            “I dropped 100+ watts on average(no, not a typo) off of my 4870X2 Quadfire rig by watercooling them. Zero other changes.”

            Hmm, okay.

            • Firestarter
            • 3 years ago

            You can try it for yourself if you like, GPU-Z makes it easy to monitor the sensors on your graphics card and yours probably has one that tells you how much power it’s using. Just start it up, run a benchmark and keep an eye on that number and the GPU temperature, it might not be very obvious in your case but my single GPU has a clearly increased power consumption at 80 degrees C compared to 50C, about 24 watts in my test. If a beefy watercooling setup can keep Waco’s GPUs at 50C instead of the high temperatures typical of stock coolers, I don’t doubt that the total power consumption of his 4 GPUs could have dropped by 100 watts

            • tootercomputer
            • 3 years ago

            I find this very interesting, which is actually what I meant when I say hmm, okday. I never thought of temp and power as being a two-way relationship.

            • Firestarter
            • 3 years ago

            [quote<]which is actually what I meant[/quote<] - if you say so

            • Waco
            • 3 years ago

            If you don’t understand, it’s probably best to not accuse others of being illogical.

            Leakage current scales with temperature. VRMs and GPUs are more efficient at lower temperatures due to this.

            Why do you think the Fury/Fury X/Fury Nano are water cooled? It keeps leakage down, which decreases power usage.

            • Firestarter
            • 3 years ago

            power consumption does vary with temperature, the biggest factor as far as I know is leak current which increases a lot at high temperatures

            you could probably find an answer in this paper: [url<]http://www.springer.com/cda/content/document/cda_downloaddocument/9781461407478-c1.pdf?SGWID=0-0-45-1268751-p174130080[/url<] I've personally seen this effect pretty clearly with my GPU, the power draw is measurably lower (as indicated in MSI Afterburner and similar tools) when the GPU is cool versus when the GPU and heatsink have heated up after a few minutes at full load

            • Firestarter
            • 3 years ago

            Here, I measured my GPU current while heating it up with the dreaded Kombustor power virus:

            [code<] C | A | 12v A 47 | 109.5 | 10.8 47 | 111.3 | 11.0 48 | 111.3 | 11.2 48 | 110.3 | 11.1 48 | 110.0 | 11.0 49 | 111.0 | 11.0 49 | 111.0 | 11.2 49 | 111.5 | 11.1 49 | 111.3 | 11.1 50 | 111.3 | 11.2 50 | 110.5 | 11.1 50 | 111.5 | 11.2 50 | 111.5 | 11.3 51 | 111.8 | 11.2 51 | 112.5 | 11.1 51 | 112.3 | 11.4 51 | 111.0 | 11.2 52 | 112.8 | 11.3 52 | 112.3 | 11.3 52 | 112.3 | 11.3 52 | 112.3 | 11.3 53 | 112.5 | 11.3 53 | 112.3 | 11.3 53 | 112.3 | 11.3 53 | 112.8 | 11.3 53 | 112.5 | 11.4 54 | 113.3 | 11.5 54 | 113.3 | 11.3 54 | 113.0 | 11.3 54 | 111.5 | 11.3 55 | 113.0 | 11.5 55 | 113.0 | 11.3 55 | 113.0 | 11.3 55 | 113.5 | 11.4 55 | 113.5 | 11.4 55 | 113.0 | 11.3 55 | 113.3 | 11.3 56 | 114.3 | 11.5 56 | 113.8 | 11.4 56 | 113.3 | 11.5 56 | 114.5 | 11.5 56 | 113.3 | 11.5 56 | 112.8 | 11.3 57 | 113.8 | 11.4 57 | 113.5 | 11.1 57 | 114.5 | 11.5 57 | 114.0 | 11.5 57 | 113.8 | 11.4 57 | 113.8 | 11.4 57 | 114.5 | 11.6 58 | 114.0 | 11.6 58 | 114.3 | 11.5 58 | 115.3 | 11.8 58 | 114.8 | 11.6 58 | 114.5 | 11.5 58 | 114.8 | 11.5 59 | 115.0 | 11.5 59 | 114.5 | 11.5 59 | 115.5 | 11.6 59 | 115.3 | 11.6 59 | 114.8 | 11.7 59 | 115.0 | 11.6 59 | 115.3 | 11.7 60 | 114.5 | 11.6 60 | 116.5 | 11.7 60 | 115.3 | 11.8 60 | 114.5 | 11.8 60 | 114.5 | 11.8 60 | 115.0 | 11.7 61 | 115.0 | 11.5 61 | 115.3 | 11.7 61 | 115.5 | 11.6 61 | 115.0 | 11.7 61 | 114.8 | 11.7 61 | 115.8 | 11.9 61 | 116.0 | 11.8 61 | 116.3 | 11.8 62 | 116.3 | 11.8 62 | 116.5 | 12.0 62 | 115.8 | 11.7 62 | 116.3 | 11.8 62 | 116.8 | 12.0 62 | 115.8 | 11.8 62 | 116.3 | 11.8 63 | 116.8 | 11.8 62 | 115.8 | 11.8 62 | 116.8 | 11.8 63 | 116.5 | 11.8 63 | 116.0 | 11.8 63 | 116.5 | 11.8 63 | 116.8 | 12.0 63 | 116.0 | 11.7 63 | 115.8 | 11.8 63 | 117.0 | 12.0 63 | 116.5 | 11.8 64 | 117.0 | 11.8 64 | 117.3 | 11.9 64 | 117.0 | 12.1 64 | 115.8 | 11.8 64 | 117.3 | 11.9 64 | 117.5 | 12.0 64 | 117.8 | 12.1 64 | 118.0 | 11.9 64 | 118.0 | 12.0 65 | 116.8 | 11.8 65 | 117.3 | 12.1 65 | 117.8 | 12.0 65 | 118.5 | 12.0 65 | 118.0 | 11.9 65 | 115.8 | 11.8 65 | 117.3 | 11.9 65 | 117.8 | 12.0 65 | 118.3 | 12.0 66 | 117.8 | 11.9 66 | 117.8 | 11.9 66 | 118.5 | 12.0 66 | 117.8 | 12.0 66 | 117.3 | 11.9 66 | 118.0 | 12.0 66 | 117.5 | 12.1 66 | 118.3 | 12.0 66 | 118.0 | 12.1 66 | 118.8 | 12.1 67 | 117.0 | 11.9 67 | 118.3 | 12.0 67 | 118.5 | 12.0 67 | 118.5 | 12.0 67 | 119.5 | 12.1 67 | 119.5 | 12.1 67 | 118.3 | 12.2 67 | 118.8 | 12.2 67 | 119.0 | 12.1 68 | 119.3 | 12.1 68 | 118.8 | 12.0 68 | 119.5 | 12.3 68 | 118.5 | 12.0 68 | 118.3 | 12.2 68 | 119.0 | 12.1 68 | 118.3 | 12.0 68 | 118.8 | 12.2 68 | 120.3 | 12.3 68 | 120.0 | 12.3 68 | 118.0 | 12.0 68 | 119.5 | 12.2 69 | 118.8 | 12.0 69 | 119.3 | 12.1 69 | 119.3 | 12.1 69 | 119.3 | 12.2 69 | 118.0 | 12.1 69 | 119.3 | 12.3 69 | 119.8 | 12.4 70 | 119.5 | 12.4 70 | 120.0 | 12.4 70 | 120.0 | 12.2 70 | 119.3 | 12.4 70 | 119.0 | 12.2 70 | 119.5 | 12.4 70 | 119.3 | 12.2 70 | 120.3 | 12.3 71 | 120.5 | 12.3 71 | 119.8 | 12.3 71 | 119.3 | 12.2 71 | 120.0 | 12.3 71 | 120.3 | 12.3 71 | 120.0 | 12.3 71 | 120.8 | 12.3 71 | 120.0 | 12.3 71 | 119.0 | 12.3 71 | 120.3 | 12.3 72 | 120.5 | 12.3 72 | 120.5 | 12.3 72 | 121.0 | 12.4 72 | 121.0 | 12.4 72 | 120.3 | 12.3 72 | 120.8 | 12.4 72 | 121.0 | 12.4 72 | 121.0 | 12.4 72 | 121.0 | 12.4 72 | 121.3 | 12.6 72 | 120.3 | 12.5 73 | 120.8 | 12.4 73 | 121.0 | 12.4 73 | 121.0 | 12.4 73 | 121.3 | 12.4 73 | 120.5 | 12.5 73 | 121.3 | 12.5 73 | 120.3 | 12.3 73 | 120.3 | 12.4 73 | 120.3 | 12.3 73 | 120.5 | 12.4 73 | 120.3 | 12.4 74 | 121.3 | 12.5 74 | 120.3 | 12.3 74 | 115.3 | 12.1 74 | 120.8 | 12.4 74 | 105.8 | 12.0 74 | 122.3 | 12.7 74 | 122.3 | 10.2 74 | 121.5 | 12.7 74 | 121.3 | 12.6 74 | 122.0 | 12.5 74 | 121.8 | 12.5 74 | 122.0 | 12.5 75 | 122.8 | 12.7 75 | 121.3 | 12.5 75 | 120.5 | 12.4 75 | 122.3 | 12.7 75 | 122.0 | 12.7 75 | 122.3 | 12.8 75 | 122.8 | 12.7 75 | 122.3 | 12.7 75 | 121.0 | 12.6 75 | 122.0 | 12.6 75 | 122.5 | 12.7 75 | 122.3 | 12.5 75 | 123.0 | 12.8 75 | 122.8 | 12.9 75 | 122.5 | 12.6 76 | 122.8 | 12.6 76 | 122.5 | 12.7 76 | 122.5 | 12.6 76 | 123.3 | 12.8 76 | 123.3 | 12.7 76 | 122.0 | 12.7 76 | 122.3 | 12.5 76 | 123.3 | 12.6 76 | 122.5 | 12.7 76 | 123.0 | 12.7 76 | 124.3 | 12.7 76 | 123.0 | 12.8 76 | 122.0 | 12.7 76 | 123.0 | 12.7 76 | 122.8 | 12.9 76 | 123.3 | 12.9 77 | 123.5 | 12.9 77 | 123.0 | 12.8 77 | 122.0 | 12.8 77 | 123.0 | 12.8 77 | 123.5 | 12.8 77 | 123.0 | 12.9 77 | 123.8 | 12.8 77 | 124.0 | 13.0 77 | 123.0 | 12.8 77 | 123.3 | 12.8 77 | 123.5 | 12.8 77 | 122.5 | 12.7 77 | 123.5 | 12.8 77 | 123.8 | 12.8 77 | 122.8 | 12.8 77 | 122.3 | 12.8 77 | 123.8 | 12.8 78 | 122.8 | 12.8 78 | 123.3 | 12.8 78 | 124.0 | 13.0 78 | 123.3 | 12.9 78 | 122.3 | 12.7 78 | 123.5 | 12.8 78 | 123.5 | 12.8 78 | 123.8 | 12.8 78 | 124.0 | 12.8 78 | 123.8 | 12.8 78 | 122.8 | 12.7 78 | 123.5 | 12.8 78 | 123.5 | 12.8 78 | 123.0 | 12.7 78 | 123.8 | 12.8 78 | 124.0 | 12.8 78 | 123.0 | 12.8 78 | 123.0 | 12.8 78 | 123.5 | 12.8 78 | 123.5 | 12.8 78 | 123.8 | 12.8 78 | 123.8 | 13.0 79 | 123.5 | 12.8 79 | 122.5 | 12.7 79 | 124.0 | 12.8 79 | 123.8 | 12.8 79 | 123.8 | 12.8 79 | 124.0 | 12.8 79 | 123.8 | 12.8 79 | 122.8 | 12.8 79 | 124.0 | 12.9 79 | 124.0 | 13.0 79 | 123.5 | 13.0 79 | 124.3 | 12.9 79 | 124.5 | 13.0 79 | 123.8 | 12.8 79 | 123.8 | 12.8 79 | 124.3 | 13.0 79 | 124.0 | 13.0 79 | 124.0 | 12.8 79 | 124.5 | 12.9 79 | 123.8 | 12.8 79 | 123.5 | 12.8 79 | 124.3 | 12.9 79 | 123.8 | 12.8 79 | 124.3 | 12.8 79 | 124.5 | 12.9 79 | 124.5 | 12.9 79 | 122.8 | 12.8 79 | 124.3 | 13.0 79 | 124.5 | 12.9 79 | 124.3 | 12.9 80 | 124.3 | 12.9 80 | 124.3 | 13.0 80 | 123.8 | 12.8 80 | 124.5 | 12.9 80 | 123.8 | 13.0 80 | 124.0 | 13.0 80 | 123.3 | 12.9 80 | 108.0 | 12.4 80 | 124.0 | 12.8 80 | 123.5 | 12.9 80 | 105.3 | 12.2 80 | 106.8 | 12.3 80 | 124.0 | 12.8 80 | 124.3 | 12.9 80 | 124.3 | 12.9 80 | 123.3 | 12.8 80 | 124.5 | 12.9 80 | 124.5 | 13.0 80 | 124.8 | 13.1 80 | 124.8 | 13.2 80 | 124.5 | 13.1 80 | 123.0 | 12.8 80 | 124.5 | 12.9 80 | 124.5 | 12.9 80 | 124.0 | 12.8 80 | 124.5 | 12.9 80 | 125.0 | 13.1 80 | 124.0 | 13.0 80 | 123.8 | 13.1 80 | 124.3 | 12.9 80 | 124.3 | 12.9 80 | 124.0 | 13.0 80 | 124.3 | 13.0 80 | 123.8 | 13.2 80 | 123.5 | 13.0 80 | 124.8 | 13.0 80 | 124.3 | 12.9 80 | 124.8 | 13.0 80 | 125.0 | 13.1 80 | 124.0 | 13.2 80 | 123.3 | 12.8 80 | 124.5 | 13.0 80 | 124.3 | 13.2 80 | 124.3 | 13.1 80 | 124.5 | 13.2 80 | 124.8 | 13.0 80 | 123.8 | 13.0 80 | 124.3 | 13.0 80 | 124.5 | 13.0 80 | 124.3 | 12.9 80 | 124.8 | 13.0 80 | 124.5 | 13.0 80 | 123.3 | 12.9 80 | 123.3 | 12.8 80 | 124.3 | 13.2 [/code<] It's my HD7950 at stock clocks with the cooler at minimum speed. First column is GPU temperature in C, second is VDDC current and third is VDDC current input (12v). As you can see the card pulls 2 ampere or 24 watt more when it's hot compared to when it's cold.

          • smilingcrow
          • 3 years ago

          “a slower-clocked, lower-voltage, better cooled card would not exceed 75W.”

          That’s all then. πŸ™‚

          • smilingcrow
          • 3 years ago

          Check out this link which closely at undervolting the RX 480:
          [url<]http://semiaccurate.com/2016/07/01/investigating-thermal-throttling-undervolting-amds-rx-480/[/url<]

            • Firestarter
            • 3 years ago

            each one of those power consumption graphs neatly shows that the card consumes more power at the end of the benchmark after heating up than at the start. The cooler might be completely adequate but that doesn’t mean that a more effective heatsink/fan can’t reduce power consumption

            • smilingcrow
            • 3 years ago

            The graphs spike up and down as you would expect as they are playing a game so it’s not a fixed load on the GPU.
            But even if you want to treat it as a fixed load if you compare the starting watts versus the average it doesn’t vary much and certainly not enough to bring the card into line with specs.

            • Firestarter
            • 3 years ago

            you’re right, it’s only about 10 watts at most. Still that’s almost 10 watts gained already, with a slightly lower voltage or clock speed the combined effect would probably be enough to bring the card into spec

            edit: to be fair a better normal cooler could hardly be expected to keep the GPU at ~50C, so 10 watts power reduction is probably completely unattainable. At most, a good HSF might shave a few watts from the consumption of a relatively small GPU like this

      • BurntMyBacon
      • 3 years ago

      [quote<]If it's a firmware fix, and the fix is 100% verified as working before the vendor cards start to appear, then this is a non-issue.[/quote<] [url<]https://techreport.com/news/30346/sapphire-nitro-rx-480-gives-polaris-a-beefier-body[/url<] Well ... [url<]http://www.anandtech.com/show/10457/asus-reveals-their-new-rx-480-gpu-design[/url<] Umm ... [url<]http://hothardware.com/news/asus-and-msi-bust-out-custom-amd-radeon-rx-480-cards[/url<] Err ... Good luck with that.

        • Chrispy_
        • 3 years ago

        Power delivery to the GPU is a board-vendor design choice. Number of VRM phases and chokes are optional, PCB routing and delivery circuits are at the vendor’s discretion. I don’t have any more information than you, but board vendors have always abandoned the reference design ASAP after a product launch. The first wave of non-reference designs have historically been different coolers on a reference PCB. Given the different power connectors, that may not happen with these RX 480 vendor designs but we won’t know for sure until they launch and review NDA’s lift.

        If the reference board is at fault in a way that cannot be corrected in firmware, I would be disappointed, but nothing more – since the reference board and cooler will soon be superseded by OEM designs. If the future OEM designs have the problem, [i<]then[/i<] I will be worried for AMD. Seriously though, the GPU is not in charge of its power delivery. That's handled indirectly. If firmware can't solve it, a board revision can without a doubt. I'm not the one who needs luck (he says, typing from a 4x Nvidia powered household).

          • BurntMyBacon
          • 3 years ago

          I was just suggesting that the 100% validated fix will have a hard time beating the custom designs to market given that there are several ready to go (with 8-pin power connectors so you know the power delivery has been changed if only slightly). One of these (Asus IIRC) is already selling in UK.

          [quote<] If the future OEM designs have the problem, then I will be worried for AMD. [/quote<] I wouldn't be too worried. This isn't as big a problem as the media are making it out as. Does AMD deserve to be raked over the coals for producing an out of spec product and claiming it is spec compliant? Emphatically yes. Will a normally functioning RX480 turn an otherwise good system into molten metal and glowing cinders? Humorous, but no. Looking at the per pin rating of the standard connectors for the PCIe slot, PCIe power connector, and motherboard power connector as well as typical current carrying capability of the motherboard traces, I don't think most people would have any clue that this was an issue if not told. The systems to worry about are OEM budget boxes that cut too many corners trying to save cost while barely meeting spec (and sometimes missing) and DIY systems with PSUs that have no business powering a modern computer. I wouldn't recommend using the 2xHDD power to 6-pin PCIe adapter for this one.

      • Klimax
      • 3 years ago

      Unless it drops performance. (So far nobody apart of AMD is sure what all options are available beside dropping clock and voltage)

    • lilbuddhaman
    • 3 years ago

    Nvidia is celebrating this weekend.

      • danny e.
      • 3 years ago

      So is AMD…. with fireworks!

        • BurntMyBacon
        • 3 years ago

        AMD and nVidia in collaboration (0_0)

    • Unknown-Error
    • 3 years ago

    Only from AMD. How this company manages to survive so many xxxx-ups is just amazing. I think they’ll be a lot more profitable if they started publishing books like “How to survive xxxx-up – The AMD method” .

      • dragontamer5788
      • 3 years ago

      Although in practice, it doesn’t really affect me.

      Don’t buy reference cards: NVidia ones are overpriced (“Founders Edition”) while AMD seems to make bugs. Things seem to be fixed by the time MSI / Gigabyte / whatever start making cards.

        • NovusBogus
        • 3 years ago

        Reference cards made a lot of sense back in the day when none of the OEMs had any clue how to build a stable graphics card. These days, not so much.

          • dragontamer5788
          • 3 years ago

          [quote<]Reference cards made a lot of sense back in the day when none of the OEMs had any clue how to build a stable graphics card. These days, not so much.[/quote<] We've gone full circle now, haven't we?

      • Puiucs
      • 3 years ago

      Pretty sure it was Nvidia that had more problems than AMD when it comes to hardware failure and drivers that damage your video card.
      Even the 1080 has problems.

      • tipoo
      • 3 years ago

      Other companies have plenty of debacles, but unlike AMD they don’t make us worry for their existence.

    • tipoo
    • 3 years ago

    Skip 30 minutes, and then skip to 53 minutes, but this guy explains whats going on with the power circuitry

    [url<]https://www.twitch.tv/buildzoid/v/75850933[/url<] That's what I feared, it's physically wired that way, such that the board is essentially seeing no difference between PCI-E power and 6 pin power, so I wonder what any software or even video bios fix could possibly "fix". Why, AMD, why...Anyways, the gist of all this is...Just get a third party board with a better power system (and cooler while you're at it) We need a roundup of which third party boards do what with the power, and if any follow AMD off the cliff

      • Krogoth
      • 3 years ago

      Somebody dun goofed with reference card layout at AMD after first samples were done. AMD was forced to stick with it along with bumping the clockspeed and memory capacity.

      This is the first-generation reference R9 290Xs debacle all over again

        • James296
        • 3 years ago

        [quote<]This is the first-generation reference R9 290Xs all over again[/quote<] who knows, could be the very same person who designed both boards

      • puppetworx
      • 3 years ago

      Watt(the f***)Man?

      • willyolio
      • 3 years ago

      huh… he says that shorting out a few connectors should “fix” the problem. I want to see him do that to his card and see what actually happens.

        • Kougar
        • 3 years ago

        Shorting the voltage planes so they draw from the PSU connector is a simple fix and the card is not going to overdaw the connector without serious overclocking. But as the guy mentions multiple times doing so still violates a bunch of PCIe design specs.

    • brucethemoose
    • 3 years ago

    I still think of Fermi whenever I hear “480”. Maybe that number is cursed in the world of GPUs.

    Also, it’s too bad AMD pushed reference designs so hard. Some of AMD’s past reference designs were horrible (like the 7950), some were great (like the 6850), but it didn’t matter because you could hardly find them anywhere.

    • tipoo
    • 3 years ago

    This video was some of the best explanation of it. Now, I wonder if the “fix” will be able to get the PCI-E power draw lower than what they draw from the 6 pin, because it seemed like they always ran neck and neck like the GPU saw no difference between them. I wonder if that’s a physical circuitry thing, as Toms also noted the phases were oddly set up.

    It also seemed the spikes in total power were only for the nanoseconds during a switch, so that at least may be an easier fix without sacrificing performance. But for the ones with only a 6-pin, the question will be if the card can separate how much it draws from the two sources.

    But the short of all of this is….Just get a third party one with an 8 pin and an actual good cooler.ο»Ώ

    [url<]https://www.youtube.com/watch?v=ZjAlrGzHAkI[/url<]

    • TwoEars
    • 3 years ago

    “PCPer suggests this issue could be avoided entirely with a single eight-pin connector or a pair of six-pin connectors, and we’d expect most custom boards to feature that kind of power-delivery setup.”

    +1. Should be a non-issue.

      • arbiter9605
      • 3 years ago

      Would be when those come out but as of NOW the cards on sale it is a big issues since only cards you can buy are effected by it. AIB makers that just used Ref card design probably just had to scrap them or modify the bios(if possible) to fix this issue.

      As much as you say non-issue, it is a very real issue as its only card out on the market and all of them have it atm.

    • CScottG
    • 3 years ago

    Just go and watch this video, it explains everything:

    [url<]https://www.youtube.com/watch?v=ZjAlrGzHAkI[/url<]

      • Mr Bill
      • 3 years ago

      +++ good video! Its all about that bass; no treble.

        • CScottG
        • 3 years ago

        LOL, I also laughed at that little impromptu reference.

        Digression: good song too, and I can’t say that for much of what I’ve heard from top 40 these days.

          • Mr Bill
          • 3 years ago

          I remember back around 1975, The local Harmon Kardon rep was bragging about how their amp could take a straight square wave signal and amplify it with almost no bounce (audible as harmonic ringing). They had this challenge that if you brought in an amp that was better, they would give a cash prize. I took in my Yamaha CR600 and had it measured but did not win.

        • tipoo
        • 3 years ago

        And I love how after the technical difficulties cut, it cuts back to them all downing their glasses of bourbon and the bottle is empty, lol.

      • tipoo
      • 3 years ago

      30 minutes in, Techreport namedrop

      • danny e.
      • 3 years ago

      Awesome opportunity for a rick roll missed.

    • tipoo
    • 3 years ago

    Interesting, I hadn’t known it only impacts the 8GB and therefore 8GBps cards before. Did people test the 4GB and find there was no issue? That’s probably why there was so much confusion over whether this was regular or not.

    And any changes to performance after the patch will definitely be interesting. Even if it goes fine though, what a PR mess.

      • Krogoth
      • 3 years ago

      I don’t think a patch is going to fix the reference cards unless they “force” the reference 480 to scale back into “470” territory.

        • tipoo
        • 3 years ago

        That’s what we’ll have to see about, but it depends what the issue is, as this PCper video nicely explains it’s in very very minute time slices, so it could be a bug rather than needed for the clock speed.

        [url<]https://www.youtube.com/watch?v=ZjAlrGzHAkI[/url<]

      • arbiter9605
      • 3 years ago

      reviews i don’t think have a 4gb card. What Ryan said @Pcper they had a reviewer card that could switch between 8 and 4gb how ever that works and is possible you got me.

      4gb card runs at a slower 7ghz on memory where as the 8gb card runs at 8ghz. Issue could still happen on 4gb card if people overclock it.

    • DrDominodog51
    • 3 years ago

    Jeff, go get yourself a beer and relax. TR’s long time policy of no articles on the weekend exists for a reason.

      • tipoo
      • 3 years ago

      Ya know, I never realized that was a thing until this comment, and probably checked the site dozens or hundreds of times on the weekends over the years like an idiot.

        • Mr Bill
        • 3 years ago

        I had not noticed either; but I still check to read the comments.

        • djayjp
        • 3 years ago

        Nah there are shortbreads on the weekend sometimes and news posts if I’m not mistaken

        • Chrispy_
        • 3 years ago

        The forums still change on weekends, and TR can still get ad revenue from your visits on weekends πŸ˜‰

      • anotherengineer
      • 3 years ago

      Especially the CANADA DAY long weekend. Go have a case of beer for us, and a burger drowning in CND bacon!! πŸ˜‰

      • ImSpartacus
      • 3 years ago

      The man wrote less than 400 words. He’s fine.

      Shoot, I figure anyone that does this kind of job is the kind of person that enjoys it. I’ve written longer forum posts over the weekend (hell, I might’ve done that today, alone) and I’m not some super human.

      Obviously I don’t think the fella should feel an obligation, but he shouldn’t feel an obligation to write for TR at all, weekend or not.

        • tipoo
        • 3 years ago

        “I’ve written longer forum posts over the weekend”

        Makes me wonder how much free content any of us have blathered across the internet over a lifetime. Most regulars would probably have filled bookshelves I’m sure.

    • Tristan
    • 3 years ago

    this GPU can reach decent perf and OC level only with 2x8pin and watercoling

      • Ninjitsu
      • 3 years ago

      no no no it needs a 24 pin connector and an open test bed in the antarctic

    • Flapdrol
    • 3 years ago

    I think it’s a good thing, keeps the miners away. More cards for gamers.

      • Meadows
      • 3 years ago

      You do know it’s 2016, right?

        • Flapdrol
        • 3 years ago

        I hear people are still mining, not bitcoin but ethereum or something. And other people seem to be buying it hoping to sell it for more.

        • Firestarter
        • 3 years ago

        people mine ethereum with them now, for example: [url<]https://www.reddit.com/r/EtherMining/comments/4r231k/ether_here_i_come/[/url<]

          • NeelyCam
          • 3 years ago

          Lol. Miners never learn. Instead of wasting power on zero sum game, maybe they should use them for something remotely useful, like folding.

            • HisDivineOrder
            • 3 years ago

            Or gaming.

    • WaltC
    • 3 years ago

    Much ado about absolutely nothing…ask nVidia why its GPUs throttle, why don’t you…;) Somebody just trying to rain on AMD’s parade…lots of peripherals routinely draw more power than spec on occasion–it’s common. And it’s *always* true when overclocking beyond a negligible amount. But the people who dreamed up this little complaint evidently aren’t that familiar with the overheads built into motherboard voltage regulation and PSUs, etc. It would be a “serious” issue only if motherboards around the world suddenly started shorting out, etc., which isn’t happening and isn’t going to happen, either. None of the AIB partners have experienced this–obviously–and nobody at AMD has experienced damage to their electronics–so you pretty much have to be an idiot to say, “Oh, gee, I can’t buy this now…!”…;) I’ve heard it all, now…*shakes head and walks off*

      • Krogoth
      • 3 years ago

      It is just Nvidiots trying to hype-up the “negative” news.

      Nvidia vendors pulled off similar stuff and the reference design 1080s have throttling issues (if you have poor chassis cooling) and run pretty close to their power limits at load.

        • NovusBogus
        • 3 years ago

        Bias, eh? I seem to recall a mighty fleet of threadnoughts setting sail with NV played fast and loose with the 970’s specs. And that didn’t hold a candle to the V300 debacle. Let’s face it, PC enthusiasts are a tough crowd…

      • JumpingJack
      • 3 years ago

      “Much ado about absolutely nothing” … a little more serious than that.

      AMD’s position as the budget provider means most of the board makers are also going for the low cost angle which usually means component and quality on the order of hitting the minimum specs.

      AMD’s cards drawing anywhere from 27% (Stock) to 51% (Overclocked) could potentially create a fire hazard.

      This is not unlike AMD to stretch specs and push to the limit to be competitive, and it is not always a bad thing, but in this case it is something to warrant concern.

        • Jalebi
        • 3 years ago

        The power delivery system on the reference Rx 480 is much much better than the one on the 1080 FE.
        [url<]http://wccftech.com/amd-rx-480-pcie-power-issue-detailed-overclocking-investigated/[/url<]

          • JumpingJack
          • 3 years ago

          Not really keen on the wccftech site, but you can believe them if you want.

          • Waco
          • 3 years ago

          So you think a card that draws more power…and has a beefier power delivery system to go with it…is better? Because bigger numbers?

          :shrug:

          • maxxcool
          • 3 years ago

          So much better it will ruin the slot and plugs of motherboards.

      • Flying Fox
      • 3 years ago

      There are already reports of dead PCIe slots that may not be up to snuff in terms of protection and/or buffer overhead.

        • Krogoth
        • 3 years ago

        It is all from pre-PCIe 3.0 slots no less.

        Ironically, they are commonly found on AMD platforms. You have to go to Sandy Bridge and earlier to get PCIe 2.0 .

          • Airmantharp
          • 3 years ago

          Rolling a pair of GTX970’s on a Sandy with PCIe 2.0 (though the board is a Z77 that supports PCIe 3.0- replaced only because of funky ASRock BIOS corruption, the Z68 board still works more or less).

          Had I not upgraded to the GTX970’s, I’d actually be in the target market for these cards. And I’d be pissed if plopping in a pair fried my board.

            • Krogoth
            • 3 years ago

            Z77 needs a Ivy Bridge chip if you want to enable PCIe 3.0 mode though. πŸ˜‰

            • Airmantharp
            • 3 years ago

            Oh agreed- and I looked at nabbing one, but they’re selling for higher than retail, while the performance gains are simply not worth the cost- especially given that I’d rather upgrade to the 6000-series or the upcoming 7000-series (assumed) if I’m going to be putting money into the system.

      • jerubedo
      • 3 years ago

      The nVidia cards do overdraw, yes. But on average they stay within specification with some spikes above specification. The RX 480 is staying above specification on average and is spiking WAY higher in addition to staying above specification on average. That’s a huge unprecedented problem. There are also several verified owners claiming dead PCI-E slots on the AMD forum (pictures included). See this link:

      [url<]https://community.amd.com/thread/202410#[/url<] The guy's mobo is a little dirty, but nothing to cause concern and his temperatures and power draw are all at good levels.

      • djayjp
      • 3 years ago

      Maybe +/- 5% voltage but anything more is dangerous and/or compromises system/component stability or life.

      • NeelyCam
      • 3 years ago

      Why is this comment so short?

    • puppetworx
    • 3 years ago

    Amateur power.

    • XenicSpinout
    • 3 years ago

    Very interesting terminology used. They didn’t say specifically say bios or firmware update, they said “driver”. Pretty ambiguous if you ask me. Is this going to be a part of a Crimson driver update or a change that will require flashing the Bios of the card?

      • nanoflower
      • 3 years ago

      It’s unknown at this point. Given their wording it does seem like it will just be a driver update. Now maybe that driver just sets a few values and everything is good or maybe the driver uploads some microcode to fix the issue, but either way we should find out if it works and if there is any impact from the change next week. It’s entirely possible that the change will reduce power draw and yet won’t impact performance at all but we will have to wait till AMD releases it for people to test and report on the tech’s impact.

    • GatoRat
    • 3 years ago

    Occam’s Razor – the preliminary review units were massively overclocked since the performance was unspectacular and AMD didn’t want a DOA card.

      • TwistedKestrel
      • 3 years ago

      That would fool nobody as clock rates are usually disclosed on the box, word would spread even faster than the discussion about the power consumption, and/or open up AMD to the lawsuit that would end the company

      • Krogoth
      • 3 years ago

      480 units weren’t massive overclocked though. They were bump-up at the last minute to meet the 970-tier performance at 960-tier pricing goal. AMD didn’t have the time or resources to do complete card redesign.

      They wanted to beat “1060” to the $199-249 punch.

    • Meadows
    • 3 years ago

    The real question: does the RX 480 really use a lot of power, or does it simply *draw* a lot of power due to a design issue?

    Probably the latter, otherwise they wouldn’t be able to just patch it out.

      • BoilerGamer
      • 3 years ago

      RX 480 draws more power(163W) than its power delivery system is rated to support(PCIE+6pin = 150W) so this issue won’t go away without AMD pushing out a driver to downclocks the card(in this case the 8Gbps 8GB memory on Reference 8GB 480 that is probably responsible for the overdraw).

        • Meadows
        • 3 years ago

        You haven’t answered the question.

          • Jigar
          • 3 years ago

          Not the design issue, but the bios configuration issue, you drop the voltage via watt man and things are back to normal. Infact if you want to avoid the watt man, re-upload the bios with tweak setting and call it a day.

          • flip-mode
          • 3 years ago

          Maybe read it again, it sounded like an excellent answer to your question. Your question was kind of vague though: does the RX 480 really use a lot of power…. well what do you mean by “a lot of power”? How much is “a lot”?

            • Voldenuit
            • 3 years ago

            [quote<]does the RX 480 really use a lot of power.... well what do you mean by "a lot of power"? How much is "a lot"?[/quote<] - More than PCIE spec for power delivery (80-90W vs 75 spec) - More than GTX 970, which is on 28 nm vs 14 nm - More than AMD's claimed TDP (164W vs 150W) I guess that much is a lot. More worrying is what happens if you put 2 reference 480s in crossfire, since with the unified power delivery on the card, you could draw much more power over the PCIE slots than spec (that 75W is combined, not per-slot). It probably won't be 180W over the motherboard, since the 6-pins will probably start supplying more power as the motherboard voltage drops, but it's probably not going to be healthy for the grade of motherboards that get paired with $200 cards.

            • Meadows
            • 3 years ago

            The key point of my original comment was not what “a lot of” stands for, but the distinction between the card actually taking that much power to operate and merely drawing that much power for no good reason.

            It’s received better answers in the responses below and so far it looks like the chip doesn’t need as much power as the board is getting.

            • Pancake
            • 3 years ago

            Your original question and follow up are stupid questions. What a device uses and what it draws are the same thing by definition. There isn’t some portion of power that magically vanishes into hyperspace.

            • Meadows
            • 3 years ago

            1st rule of thermodynamics: you do not talk about thermodynamics. Yes, yes, I know all that.

            However, it’s you who is wrong. It’s entirely possible for a badly designed board to draw more power than the chip calls for for its stable operation, e.g. through bad voltage management. Which is the kind of phenomenon I meant.

            Granted, that’s also the sort of issue that should have never slipped through testing and validation, but you never know.

            • NeoForever
            • 3 years ago

            “… only stupid answers.”

            I am not speaking for anyone else. But please don’t discourage people from asking questions.

        • Jalebi
        • 3 years ago

        Actaully, those affected by powergate had gpus that performed better. Interestingly even when thier max peak was at 115w the gpu still couldn’t oc. Trips were a bit lower as well. It performed as good as a 980ssc from evga which in line with the rumours and amds figures as well

        • raddude9
        • 3 years ago

        Just Downclocking! Are you trying to spread FUD or something, there are many other ways to reduce PCIe power.

          • raddude9
          • 3 years ago

          So many downvotes! I’m disappointed with the TR readership. At the very least you people should know about under-volting. There have already been a number of articles about very successful under-volting of the RX480.

            • DoomGuy64
            • 3 years ago

            Hawaii was the same. Don’t know what it is about AMD forcing their cards to use more power than they need.

            • NeelyCam
            • 3 years ago

            Is this not something that AMD should address [b<]before[/b<] shipping the product? I don't think it's appropriate to expect your customers to go to the Interwebs to find articles about modifying the settings of their new GPU just to bring it within specs? The folks here would be knowledgeable to do it, but it should not be expected from regular customers.

            • raddude9
            • 3 years ago

            Yes, it is something they should have addressed before shipping, I never said otherwise.

            But to say that down-clocking is the only fix is disingenuous (apart from under-clocking, I’ve seen reports that AMD can re-balance where the power is being drawn from through software).

            So there should be a number of options to fix the issue through a driver update, i.e. the average user will never have to know about it.

        • willyolio
        • 3 years ago

        the real question is if it actually NEEDS that high power draw to maintain those clocks, or if it was some driver/design error that makes it pull more power than it’s supposed to.

      • XenicSpinout
      • 3 years ago

      What do you mean by “due to a design issue”? Do you mean the pcb, architecture of the chip or something else?

      The 480 draws a lot of power probably because Global Foundries 14nm chips are less efficient than TSMC’s 16nm chips.

        • Meadows
        • 3 years ago

        Here’s the thing: if the card uses a lot of power because the chips are just that bad, then the software fix will inevitably entail downclocking and then not even these lukewarm reviews will be representative of the actual product anymore.

        Which is probably not a risk AMD would take so nonchalantly, given their position.

          • flip-mode
          • 3 years ago

          If I understand the situation, it is not a problem with the chip, it is a problem the design of the power delivery circuitry on the circuit board. I have heard that if the board would have been supplied with an 8-pin power connector then it would have been fine.

          What I am not sure of, though, is why the card overdraws on the PCIe slot instead of on the PCIe power connector. I do not know if that is a chip issue or a circuit board issue.

        • Jalebi
        • 3 years ago

        Thats not really the case. Idk why everyone thinks tsmc is the best. The samsung Finfet is slightly better then the one from tsmc

          • Airmantharp
          • 3 years ago

          The process may be, but we’re talking about AMD’s design and GloFo’s interpretation of it, not Samsung producing their own designs in their own fabs.

          Really, we should be happy that AMD has done as well as they have weening themselves off of TSMCs teat. This level of independence, and hopefully mastery of their fabrication process, will make them less affected by market conditions where they’d take a back seat to TSMCs higher-volume/profile customers (like Apple), while keeping their designs relatively easy to be producible in Samsung fabs should demand drastically exceed expectations.

            • dragontamer5788
            • 3 years ago

            [quote<]Really, we should be happy that AMD has done as well as they have weening themselves off of TSMCs teat. [/quote<] I get what you're trying to say and all... but considering that GloFo used to be part of AMD... I really don't think your metaphor here works at all.

            • Airmantharp
            • 3 years ago

            Part of AMD, sure, but the part that made CPUs- this is the first time they’ve made a GPU, and it’s quite the decent first effort.

          • NeelyCam
          • 3 years ago

          And Intel’s is better than either one of those. But it doesn’t really relate to the discussion we’re having now.

      • ImSpartacus
      • 3 years ago

      I keep wanting to believe AMD, but recent history makes it hard to trust them on something like this.

      While the 480 seems to perform well enough to expectations, this power consumption issue is not cool and I don’t have complete confidence that AMD will be able to remedy it.

      I wouldn’t be surprised if we see most non-reference 480s sporting 8-pin power connectors for both OC reasons as well as to potentially power the card on its own (or reduce load on PCI-e) if that’s possible.

        • BurntMyBacon
        • 3 years ago

        [quote<]I keep wanting to believe AMD, but recent history makes it hard to trust them on something like this. [/quote<] [b<]Then don't![/b<] Board manufacturers are not locked into the reference power delivery in their custom boards. MSI, Asus, and Sapphire have all already announced custom products. They are clearly in a race to capitalize on the reference board issues, so I'm not seeing a very long time to market. I recommended waiting on custom boards ever since I saw the diminutive stock cooler. Power delivery is just another reinforcing point, though I'm actually less concerned with that than the cooler. Don't get me wrong, I berate AMD for not meeting spec. However, the weak point for power delivery in most motherboards is the connectors. Most DIY motherboards have PCIe connectors rated for many times the current that PCISIG specifies as the maximum. The 24pin molex power connector for the motherboard might perhaps get over burdened if you use several overspec cards in a bargain basement motherboard that wasn't designed for it. However, I think the most danger comes from upgrading super budget OEM desktop systems that cut every corner possible, often sacrificing the ability to run [b<]in-spec[/b<] cards stably.

      • Krogoth
      • 3 years ago

      RX 480 at load pulls more current then the recommend amount for 3.0 PCIe and 6pin external PCIe power connector spec calls for.

      Some factory-overclocked 750Ti and 960 units also ran into similar problems, but they were factory-overclocked and vendors were doing at their own risk.

      This whole thing is purely a design issue on reference boards. I suspect it is because AMD was originally going to throw in only 4GiB of VRAM and use lower-clocked chips, but the first round samples were unable to match their performance target (970-tier). AMD threw more memory and push the clock speed higher to meet performance goal, but it threw the power consumption beyond 6-pin + PCIe 3.0 slot specs.

      This normally would force AMD to do a complete redesign on reference boards, but they couldn’t afford any further delays and R&D costs. The current reference 480 is the result.

        • arbiter9605
        • 3 years ago

        6ping PCI-e power can do more then 75watts and its fine, those wires can handle more. The traces in board for power are very small and if board is cheap they are as small as they can be made to meet the spec where as higher end boards would be able to handle it but still a question on if and what would happen long term 6+months or 1+years if it could degrade some mid tier boards.

          • Krogoth
          • 3 years ago

          The problem is that reference 480s are drawing too much current from PCIe slot which does cause issues with older pre-PCIe 3.0 motherboards (there are stories already) and possible long-term issues with sub-par boards.

          This can easily be addressed if it 480 draw more power from the external PCIe port and had used a 8-pin connector. AMD didn’t have the time for this in order to meet their launch goal.

            • Mr Bill
            • 3 years ago

            The guys at PCper seem to think the 6-pin connector could handle the load. So, maybe this can be fixed in the driver in the same way that AMD’s overclock (AOC) app could make all sorts of adjustments in the CPU and chipset. It may well be that the power logic can be instructed to get the power from the 6-pin connector rather than the PCIe bus.

            Edit: That of course still does not address that the TDP would still be 165-170W. The youtube video linked by CScottG covers it pretty well. See the discussion at 34 minutes for available power via the 6-pin. The spec says 75W for the 6-pin. But they say the power supply can easily deliver more over the 6-pin. Out of spec, but preferable to hitting the PCIe bus for the power.

            • Meadows
            • 3 years ago

            Sounds like a better explanation than “hurr durr GloFo”.

            • homerdog
            • 3 years ago

            It can’t be fixed by a driver or BIOS update, the problem is in the way the power delivery is wired on the PCB. You can apparently fix it with a soldering iron and some time and know-how though.

            • Voldenuit
            • 3 years ago

            Yeah, sounds like a board design problem; but it’s disingenous of AMD to imply that a fix is possible in software, when the only thing they can do in software is to downclock and downvolt the GPU/memory.

            A Rev 2 board is sorely needed to fix this problem, but I have a feeling that early adopters will be left in the cold, especially if the promised solution is a driver downclock, as I am suspecting.

            • BurntMyBacon
            • 3 years ago

            [quote=”homerdog”<]It can't be fixed by a driver or BIOS update, the problem is in the way the power delivery is wired on the PCB. [/quote<] [quote="Voldenuit"<]it's disingenous of AMD to imply that a fix is possible in software[/quote<] There are, in fact, voltage regulation schemes that can be controlled programmatically. In other words, if they use the appropriate chip instead of hard wiring the VRMs, then they may be able to do exactly as they say. I find it doubtful that they'd use such a part that would no doubt incur more cost when they were pushing so hard to hit the $200 mark (heatsink anybody). However, I don't know that they didn't use it either. It may have help with the power efficiency end of things.

          • BurntMyBacon
          • 3 years ago

          Flat traces can handle more current than round wires. The current flows through the skin of the conductor, so the more surface area, the better. Even on cheap motherboards, the traces are generally not the weakest link. Connectors that rely on force physical proximity are much more resistive than proper solder joints which are in turn much more resistive than traces and wires. Just think about how many burnt connectors you’ve seen where the connecting traces are fine. Have you ever seen it the other way around?

          Back in University, my professor and I made a fuse out of a PCB where we cut the trace thinner at a junction point. The junction point was very narrow (on the order of a milimeter or two) before it would blow at an approximate 100A load. The connector for the load was huge by comparison. Point is, I’d be more worried about degradation and corrosion of the PCIe connector itself than the PCB traces.

        • flip-mode
        • 3 years ago

        [quote<] I suspect it is because AMD was originally going to throw in only 4GiB of VRAM and use lower-clocked chips, but the first round samples were unable to match their performance target [/quote<] Man, you sure are comfortable with making wild speculations.

          • Krogoth
          • 3 years ago

          It is quite evident with the current reference 480 cards. The layout for first batch of cards was meant for 150W TDP at most. The Polaris chips that operated within that envelope couldn’t match AMD’s performance goal (970-tier). AMD was forced to bump clockspeeds up but the cards went beyond that threshold.

          AMD couldn’t afford a delay for a redesign for a higher TDP since 1060 was coming out soon and AMD want to steal Nvidia’s thunder in $199-249 market. Frankly, AMD’s GPU division is already on shaky ground with several rounds of product delays in the past two years. Just imagine the PR blacklash if 480 was delayed for a redesign or AMD was forced to downclock the 480 but it made it only somewhat faster than R9 380/960 that it intended to usurp?

          The result is the current reference 480 boards. AMD is hoping that third-party AIB vendors will fix the problems. They are gambling that PCIe power issues will only an minority of early-adopters. Funny, that problems are found mostly on pre-PCIe 3.0 boards and AMD platforms make up the bulk of that. I don’t think it will burn slots in aforementioned motherboard,s but rather it will cause stability issues (random CTD, BSOD and system powering off abruptly). Patches will force the card to tone down to the original 150W TDP envelope.

            • Meadows
            • 3 years ago

            Cite your sources.

            • Krogoth
            • 3 years ago

            The various videos of debacle and people who took a close look at the hardware itself.

            • Meadows
            • 3 years ago

            “The various videos” is not a citation.

            • smilingcrow
            • 3 years ago

            “Just imagine the PR blacklash if 480 was delayed for a redesign or AMD was forced to downclock the 480 but it made it only somewhat faster than R9 380/960 that it intended to usurp?”

            Not as much as releasing a reference board that is seemingly out of spec and potentially unstable I imagine.
            Disappointing performance is acceptable but out of spec hardware is a PR disaster.

            • Krogoth
            • 3 years ago

            AMD’s GPU division at the moment is between a hard place and a rock.

            Nvidia has been relentless since their comeback with Kepler after GF100 debacle. The last game changing chip from AMD was the Hawaii a.k.a R9 290 and its rebrands.

            • tipoo
            • 3 years ago

            The wafer agreement with Glofo can’t be helping things either. I really hope Glofo cuts them loose of it before the natural expiry of 2024.

            • smilingcrow
            • 3 years ago

            If this situation turns out to be near to or worst case then they as well have hit themselves over the head with a rock.
            If they damage the AMD brand even more then some will forget about waiting for Vega or Zen and buy elsewhere.
            This puts even more pressure on Zen to succeed and if it’s not on GloFo then I am confident.

            • tipoo
            • 3 years ago

            Zen and higher end GPUs are still TSMC, thankfully. The 480 as a mid range high volume part was probably to satisfy the wafer agreement.

            • Mat3
            • 3 years ago

            Is that official? I thought it was Glofo for everything now.

            • tipoo
            • 3 years ago

            [url<]http://wccftech.com/amd-contracts-tsmc-produce-zen-16nm-woes-14nm-process-troubles-globalfoundries/[/url<] Gods I hope it's not Glofo for everything

            • shank15217
            • 3 years ago

            The 970 has been out for a very long time, its not a target they would miss at the last moment.

    • TripQue
    • 3 years ago

    TL:DR: AMD – Still ready to set your computer aflame since 1969! πŸ˜€

      • ronch
      • 3 years ago

      Er, wat?

        • sycomonkey
        • 3 years ago

        He’s probably referring to before AMD had thermal throttling and taking the heatsink off would literally set the CPU on fire. This was before the first dual-cores, ~2002.

          • UberGerbil
          • 3 years ago

          Yeah, I believe Tom’s used to have a video of doing exactly that.

          • Krogoth
          • 3 years ago

          Actually any Intel CPU before Pentium 3 Coppermine didn’t have thermal protection. Pentium Pros and Pentium IIs would fry themselves if you were to remove their HSF or a suffer fan failure during a load.

          AMD didn’t start putting thermal protection on their chips until K8. K7s were depended on their motherboards for thermal protection (They started doing this around when first gen-Athlon XPs [Palominos] came out).

          I don’t believe Cyrix ever bother putting any thermal protection on their chips and Centauri was gone long before chips start requiring cooling.

            • jessterman21
            • 3 years ago

            yep, back in college, my poor hand-me-down Pentium II melted because the fan died suddenly… πŸ™

          • tipoo
          • 3 years ago

          [url<]https://www.youtube.com/watch?v=_Vrh3WoATzw[/url<] Good times

        • Mr Bill
        • 3 years ago

        [quote<]Er, watt?[/quote<] fixed it for ya!

        • curtisb
        • 3 years ago

      • Jigar
      • 3 years ago

      That is exactly the people Nvidia is looking for, GO NVIDIA… /s

Pin It on Pinterest

Share This