TR Forums

I'm not going to apologize about calling the GTX970 gimped, it was compared to the GTX980.

After reading a couple reviews I am a bit let down, I was hoping for better, mostly on the power and temperature front. Gaming performance is a touch less than what I was hoping for. Over time I think the drivers will get it where I was hoping it would be.

The slow ram in the 4gb card is a bit disappointing and the mostly $239 for 8gb instead of $229 as well.

What I took from the reviews I read is that the RX 480 is almost a very good 1440p card. It misses at that resolution a bit, I'm not sure if I would buy it if I had a fancy 1440p monitor. At 1080p, it is a very very good card, but also a bit hot. That leaves me waiting for RX 470 and hoping to see some more in depth reviews that illustrate the memory clock and core clock scaling for the Polaris architecture.

My biggest disappointment is I went to Fry's Electronics to try and buy a card. They had never heard of it. That's really strange because AMD headquarters is maybe 1,000' from the computer department at Fry's Sunnyvale where I was asking.

Major fail Fry's.

I will try again during lunch at the small shop that came through with an i7-6700k when I needed one.

blahsaysblah wrote:

chuckula wrote:

It's not a bad card, but I think Frugal should make a few apologies for hurling around terms like "gimped" in describing the two year old GTX-970, especially when we look at TR's 99th percentile results and the fact that the Rx 480 is only marginally more power efficient.

Agree, I'm sorry but if your read the Tom's Hardware's power review page for the RX 480. Its really bad. Read about their power consumption setup first.

[...]

edit: not doomed, partners can fix this by using 8pin power instead of 6pin. but bad reference design pushed to limit. Probably afraid to put 8pin on it due to PR nightmare.

Possible, but still AMD.

So unacceptable (if this result can be reproduced), pulling 155 W through the PCIe slot's 12V rail is a ridiculous 13 A, more like 14 A if the voltage drops but stays within the 8% voltage spec. The spec is 5.5 A, how can 14 A (=235% !!) be justified? ... there's clearly no control from the card, it doesn't know when to stop.

I'd love to see other investigations of this issue, including with other AMD and NVIDIA cards.

Frugal wrote:

I'm not going to apologize about calling the GTX970 gimped, it was compared to the GTX980.

That was a joke.

Why are you getting a reference card. No matter how you look at the power numbers, all sites are agreeing on around 165 W during typical high end gaming. That means AMD is knowingly selling a card that is not PCI-E certified. PCI-E is 75 W for bus and 75 W for 6pin. I would wait for a partner card with 8pin power(75 W + 150 W). Just regular heavy gaming is over amping your PC without your permission.

Maybe AMD will give a driver/firmware update to make the card not go above 150W during normal operation.

Also GTX 1060 will be announced on July 7. Not saying go green. But wait for 8pin at least.

All the overclocking results I've seen so far have been very mediocre. Hopefully that's due to the cheapo heatsink and budget power delivery design.

Topinio wrote:

Possible, but still AMD.

So unacceptable (if this result can be reproduced), pulling 155 W through the PCIe slot's 12V rail is a ridiculous 13 A, more like 14 A if the voltage drops but stays within the 8% voltage spec. The spec is 5.5 A, how can 14 A (=235% !!) be justified? ... there's clearly no control from the card, it doesn't know when to stop.

I'd love to see other investigations of this issue, including with other AMD and NVIDIA cards.

The card has a 6 pin power connector for 75 Watts of power at 12V. 6.25A + 5.5 is 11.75 Amps for 141 Watts (plus 10 Watts available at 3.3V) for PCI-E 150 Watts available for a 6 pin card.

I'm going to marvel at all those who considered the 500MB of slow GTX 970 RAM (no real world impact) an utter travesty yet completely overlook an overload of the PCIe power spec (potentially ruinous).

blahsaysblah wrote:

Topinio wrote:

Possible, but still AMD.

So unacceptable (if this result can be reproduced), pulling 155 W through the PCIe slot's 12V rail is a ridiculous 13 A, more like 14 A if the voltage drops but stays within the 8% voltage spec. The spec is 5.5 A, how can 14 A (=235% !!) be justified? ... there's clearly no control from the card, it doesn't know when to stop.

I'd love to see other investigations of this issue, including with other AMD and NVIDIA cards.

The card has a 6 pin power connector for 75 Watts of power at 12V. 6.25A + 5.5 is 11.75 Amps for 141 Watts (plus 10 Watts available at 3.3V) for PCI-E 150 Watts available for a 6 pin card.

The Tom's Hardware test results are:

W	Min.	Ave.	Max.
PCIe 6-pin	24	79	142
Slot 12 V	23	82	155
Slot 3.3 V	2	4	7

which translate to:

A	Min.	Ave.	Max.	Limit	% reached
PCIe 6-pin	2.0	6.6	11.8	6.25	189%
Slot 12 V	1.9	6.8	12.9	5.5	235%
Slot 3.3 V	0.6	1.2	2.1	3.0	70%

which is uncool if valid test.

FWIW having read all the other reviews, I'm not convinced the Tom's Hardware test hasn't got it wrong, but its method looks good so...

Topinio wrote:

blahsaysblah wrote:

Topinio wrote:

Possible, but still AMD.

So unacceptable (if this result can be reproduced), pulling 155 W through the PCIe slot's 12V rail is a ridiculous 13 A, more like 14 A if the voltage drops but stays within the 8% voltage spec. The spec is 5.5 A, how can 14 A (=235% !!) be justified? ... there's clearly no control from the card, it doesn't know when to stop.

I'd love to see other investigations of this issue, including with other AMD and NVIDIA cards.

The card has a 6 pin power connector for 75 Watts of power at 12V. 6.25A + 5.5 is 11.75 Amps for 141 Watts (plus 10 Watts available at 3.3V) for PCI-E 150 Watts available for a 6 pin card.

The Tom's Hardware test results are:

W	Min.	Ave.	Max.
PCIe 6-pin	24	79	142
Slot 12 V	23	82	155
Slot 3.3 V	2	4	7

which translate to:

A	Min.	Ave.	Max.	Limit	% reached
PCIe 6-pin	2.0	6.6	11.8	6.25	189%
Slot 12 V	1.9	6.8	12.9	5.5	235%
Slot 3.3 V	0.6	1.2	2.1	3.0	70%

which is uncool if valid test.

FWIW having read all the other reviews, I'm not convinced the Tom's Hardware test hasn't got it wrong, but its method looks good so...

The "Max" measurements could be transient spikes that aren't necessarily accurate of true power draw.
However, sustained power levels that are near or above 150 watts are a little more worrisome and are probably close to accurate.

Hopefully it turns out to be just like GTX 970 3.5GB issue. Not big deal. Still, the GTX 1080 review shows they targeted a much smaller draw against motherboard and they did not use 3.3V at all.

PC-Per says Raj said their launch 4GB cards actually have 8GB with half disabled... Was cheaper to only validate one card. That there is literally one PCB design for all 470 and 480 variants is likely. The cheap/simple 50/50 power split makes sense on those cards. It might have even made sense for the 480, until they decided to give everyone the maximum stable OC at last minute.

That leaves room for more savings for 4GB and more growth for custom 8GB boards with safer power design.

If only they pulled all the power via the 6pin. I wouldnt care.

Mr Bill wrote:

morphine wrote:

Frugal wrote:

So with three cards and 12gb total you were running out of memory at 4k?

VRAM isn't pooled in SLI/XFire.

Is that still true with DX12 sharing without SLI/Xfire links?

Generally, you still cannot add up the RAM from each card.

Implicit mode is basically SLI/Crossfire with a few improvements, which means AMD/nVidia are handling most of the work behind the scenes.

But even in explicit mode, developers will have to duplicate data in VRAM between the GPUs. And the framebuffer cannot be split between GPUs either.

Basically, you will never be able to get 2x 4 GB graphics cards and have them perform as if you had a superior 8 GB card. It fundamentally cannot work that way.

Developers could, in theory, get clever with explicit mode---but the underlying issue of duplicating texture, geometry, and frame data will never go away completely.

I also don't expect many developers to push explicit mode anyway, as the target market is miniscule.

Frugal wrote:

I have seen that AMD Poland has confirmed the MSRP of the 8gb RX 480 as $229.

If you look at other video cards, that suggests that we should have a decent selection of non-reference 8gb cards at $260 or less.

Most Nvidia fanboys have been estimating it at $249 for 8gb then applying an exchange rate (or two) and claiming it's overpriced.

http://www.newegg.com/Product/Product.a ... 6814129322

You were saying?

From what I have seen, only one review sees more than 75 watts drawn from the slot. That could be one card, it's s little too soon to tell, other reviewers should check the power independently, swap cards, find any trends and then pass judgement.

My strategy at this point is to get an RX 470, with some cores disabled it should draw less power, run cooler and be good enough for 1080p which is where I'm at right now.

I want a reference board because my horizontal HTPC style case will work better with a blower card than open air. It's pretty simple.

About the price, AMD Poland screwed up and then AMD underclocked the 4gb cards over $10. It's a crummy thing to do but at least it wasn't $100.

Frugal wrote:

techguy wrote:

GTX 960 came 4 months after 980 & 970 and was in fact using a separate GPU. The last time NV required a 3rd tier salvage SKU was way back in the GTX 200 (i.e. Fermi) days, and that was a MASSIVE GPU. GP104 is already tiny (relatively) at only 314mm^2, and has a quite gimped salvage part already in the 1070. 1060 is highly unlikely to be GP104-based. 1060 Ti on the other hand... Could be say 85% of a 1070 or so and probably still beat RX 480 at maybe a $30-$50 premium (since it's NV). RX 480 isn't going to approach 1070 level performance, you should let go of this idea (when you say 1070 will "probably beat RX 480" you don't sound fully convinced). We'll know in 5 days, but again, it doesn't make sense to sell a competitive part for half what your competitor charges. It's not like NV doesn't have the margins to drop the prices on 1070 if needed, it's a 314mm^2 GPU using regular old GDDR5 for crying out loud.

I say probably only because the embargo is still on and we don't know what the release day drivers will do.

There is a small chance that the RX 480 is somewhat competitive with the GTX1070. I don't think it will be quite that good but I don't know.

If it's 80% of the performance at half the cost and actually available, wow.

I think there is a chance of that because of that 640 core advantage.

There is a rumor that the release day driver enables a discard accelerator or something that greatly helps with tessellation and brings it up to Nvidia level. That might be the secret sauce that gets it competitive and allows stuff like hairworks to run just as well on AMD.

It's a rumor that sounds very nice, maybe too good to be true but since AMD has consistently been behind in tessellation, it sounds plausible that they would try to fix it.

As far as margins go, the Polaris die is 232 mm^2 and GF is probably cheaper than TSMC.

Nvidia GTX10X0 cards have so far been a trickle, slow enough that I have to assume some sort of yield problem so the price per die may be huge. It could be something else like Apple taking up all of TSMC's fab capacity so Nvidia can't get wafers which is almost as bad.

The latest word on the RX 480 launch is that some stores will receive 25x the number of cards for launch day as the GTX10X0 launch. That sounds like AMD can pump them out. It is unlikely that we will see a lot of price gouging.

What were you saying? 80% of a 1070, ROFLMAO. RX 480 = GTX 970 (wins some, loses some) WITHOUT the overclocking. Overclocked 970 can be 15-20% faster than stock. OC'd 1070 is 50% FASTER than OC'D 970. We're not even in the same ballpark at that point.

Look, I can appreciate optimism, especially when it comes to AMD, I used to be one of the biggest AMD fanboys on the planet (username = shaidarharan @ Rage3d). But they haven't been competitive at the high-end for 10+ years now unless you count Crossfire-on-a-stick solutions. Maybe Vega and Zen will be competitive, maybe they'll be more of the same. After 10 years of disappointment, I'm not going to get my hopes up.

chuckula wrote:

The "Max" measurements could be transient spikes that aren't necessarily accurate of true power draw.

I'm not concerned by those in a vacuum.

If it exceeds published specs for transient PCIe spikes, then I might be worried. Or maybe a comparison to previous/competing products would be enlightening.

However, sustained power levels that are near or above 150 watts are a little more worrisome and are probably close to accurate.

They shouldn't ship a product out of spec like that at all.

Drawing too much power from the motherboard can cause additional wear or even failure.

If they need to play loose with power specs, they should at least do it on the 6-pin line. That way, they are only straining the PSU---which most likely can cope with it since the 12V line is usually shared anyway.

I really wanted to buy Vega when it launched, but this is making me reconsider.

At this point I'd be all over the 480 but those power figures definitely give me pause. I mean, look, stress tests will easily put cards over the "limit" of what the PCI-E cables/PCI-E bus are supposedly capable of. And this isn't the first card to have gone over even the spec's maximum wattage either, for example: https://www.techpowerup.com/reviews/MSI ... ng/28.html (based on that cards connectors it should never go above 300W)

But this is a reference card... Heck AMD gave two 6 pin connectors for their reference 7870 and that used even less power than the RX 480. Not sure what they were thinking giving a 150W card only 1 6 pin, or even developing it to derive that much power over the bus to begin with.

Has anyone reviewed the 470/460 yet? I thought those cards had the same release date.

Frugal wrote:

My biggest disappointment is I went to Fry's Electronics to try and buy a card. They had never heard of it. That's really strange because AMD headquarters is maybe 1,000' from the computer department at Fry's Sunnyvale where I was asking.

Major fail Fry's.

I will try again during lunch at the small shop that came through with an i7-6700k when I needed one.

Central Computers? Given my experiences with local Frys', the odds of them knowing anything is very low unless you're at the Fremont one.

I would be less concerned with audio noise when your motherboard is slowly dying from too much power going through the traces. To be fair, how many people here actually wanted to buy AMD's stock cooler? **Crickets** I'll be waiting for Sapphire + Asus + Gigabyte to release their boards.

so.

the card is here. it seemingly violates the PCIe spec. it seems it is possible it could damage a low-end motherboard (not that this card would ever find its way into a build that cut corners on a motherboard to fit inside a budget). and it is basically a 970 without DX12, which is a teensy bit of a stretch of the "somewhere between a 970 and 980" scuttlebutt. granted, DX12 will be all that's relevant going forward, but it remains that, today, in today's context, still a teensy bit of a stretch.

thoughts? I think it best to wait for non-reference 8pin boards yes?

Was trying to figure out limits for my PC. I recently built my own power cables and bought tin clad stranded copper 18AWG 16/30 wire to match what came with my Corsair SF450 PS. They say they are rated for 7A.

Anyway, assuming the minifit pins are the higher Amp ones and the traces inside PS can handle it:
2 x 12V x 7A = 168W for 6 pin (which is possibly why you see some PS using single 6pin wires with daisy chain at end to provide 2x6)
3 x 12V x 7A = 252W for 8 pin (im not 100% sure ive seen this versus just daisy chain adapter available to do it)

Tom's Hardware old forum post explaining PCI-E 6+8 pin cables/pins and power capacity

Definitely an 8 pin with a real cooler to bring down power usage would make this card look better. Simply PCB miss-step, nothing to do with the chip. That's really great.

I just really feel sorry for any kids starting summer break, waiting for this card, putting into old computer and possibly frying their MB after a couple of weeks of hard gaming. Because who's not also going to OC it a bit even if hot summer days are coming. Worse, they shorten life of computer unknowingly.

I never got re-reimbursed when i replaced the Nvidia GPU in my laptop for overheating/failing. Never new about the class action until it was long over. I was pretty mad about paying $200+ for low end ATI replacement(only thing i could find, nothing against ATI).

fhohj wrote:

so.

the card is here. it seemingly violates the PCIe spec. it seems it is possible it could damage a low-end motherboard (not that this card would ever find its way into a build that cut corners on a motherboard to fit inside a budget). and it is basically a 970 without DX12, which is a teensy bit of a stretch of the "somewhere between a 970 and 980" scuttlebutt. granted, DX12 will be all that's relevant going forward, but it remains that, today, in today's context, still a teensy bit of a stretch.

thoughts? I think it best to wait for non-reference 8pin boards yes?

$200 price range is exactly a card to go into budget machine.
970's Maxwell has slight DX12 disadvantage. Team red's GCN has had DX12 advantage as plus for a while.
No, non-reference 8pin is not entire answer. You have to wait for a review to show they also limited the power draw on the PCI-E side. The GTX 1080 for example uses around 40W average to leave a safe peak of 60-ish watts per Toms Hardware review.

chuckula wrote:

The "Max" measurements could be transient spikes that aren't necessarily accurate of true power draw.
However, sustained power levels that are near or above 150 watts are a little more worrisome and are probably close to accurate.

Did you click through and look at the presented results? It's not spikes, the spikes are 300 W (full card) which is at 142 W from the slot plus 155 W from the 6-pin -- at the same time. In game.

The sustained slot power is way over the spec (in the Metro: Last Light UHD test), the average load on the slot's 12V supply is 126% of the limit per spec.

If this test is accurate, a reasonable hypothesis might be that this card was designed to run at ~1.0 GHz and would have stayed within spec, but AMD needed a win so turned it all the way up to 12, ignoring that (at least on the lower ASIC quality chips) it will pull more current than potentially it and the customers' motherboards and PSUs can necessarily deal with.

The only way to know, though, is if someone tries to independently reproduce the test with a decent sample size of cards bought at retail. Let's hope.

Edit:

It's also shown at Golem.de, see the bar charts, click right for images 66-69 with averages for "Mainboard 3,3 Volt", "Mainboard 12 Volt", and "Card 6-Pol 12 Volt" in each game. Results are 4 W for slot 3.3 V, and range 78-83 W and 79-84 W for slot 12 V and 6-pin respectively. On average. Which is 6.5-6.9 A average from the slot 12 V supply, assuming voltage is close to 12 V.

Hardware.fr has it measured at 6.92 A in Battlefield 4 at defaults, and when its limits are raised (in WattMan?) it's at 7.10 A in Battlefield 4 and 7.79 A in The Witcher 3. The article maybe seems to suggest that the cause might be that the AMD tweaking controls only the limits of the GPU chip itself, and that the power drawn by the other on-card components, particularly the RAM, might be the problem. I could be misreading this, though, as my French is more than a bit weak.

Topinio wrote:

Did you click through and look at the presented results? It's not spikes, the spikes are 300 W (full card) which is at 142 W from the slot plus 155 W from the 6-pin -- at the same time. In game.

It certainly does average way too high, however, a short spike like that is not terribly concerning. The average load, however, is.

Are there any other reviews out there featuring multiple DX 12 titles besides PC Perspective? It's interesting because you have titles like Rise of the Tomb Raider which show significantly different performance results with it enabled vs disabled when comparing different reviews.

TR review: http://techreport.com/review/30328/amd- ... reviewed/8
PCPer review: http://www.pcper.com/reviews/Graphics-C ... omb-Raider

In the TR review the GTX 970 and RX 480 are neck and neck, that's in DX11 mode. In the PCPer review where its in DX12 mode the RX 480 is significantly faster than the GTX 970.

Both reviews feature the latest Hitman game too, again TR using DX11 and PCPer using DX12. Both results show the RX 480 significantly head but the lead is even greater in PCPer's DX12 benchmark.

Topinio wrote:

chuckula wrote:

Power related things....

More power related things....

So, apparently, the 75W through the slot is just a starting number. Motherboard manufacturers who use better components, bigger traces, what have you, can provide more power through the slot. The adapter must request more power, and the PCIe controller must respond with the new limit. Whatever the PCIe controller says the new limit is, the adapter may not exceed it.

I'm not claiming to be an expert, but people much more knowledgeable in this matter are here:

https://www.reddit.com/r/Amd/comments/4 ... _solution/

That makes a lot of sense. AMD designing a card out of spec and risking enormous litigation and reputational damage does not.

That revelation and PCPer's claim that 8GB and 4GB cards are identical cards is really making the card look more appealing.

If that Redit thread is the same one I read it seems to suggest that the motherboard used in those test was possibly set for a higher power delivery which would make this a motherboard issue.

This should be easy to prove or disprove by just swapping out the motherboard and re-running the test.

I'm just waiting for someone to plug one of these into a barely-spec Dell or HP, or even just your odd ECS motherboard, and catch stuff on fire.

That sole 6-pin connector is deceiving, and if the RX480 winds up being the high-volume product that its performance deserves it might cause quite a bit of trouble if these power-consumption concerns are well-founded.


(that said, one could hook it up with a min-spec PSU and motherboard and it might well fall within tolerances...)

Airmantharp wrote:

I'm just waiting for someone to plug one of these into a barely-spec Dell or HP, or even just your odd ECS motherboard, and catch stuff on fire.

That sole 6-pin connector is deceiving, and if the RX480 winds up being the high-volume product that its performance deserves it might cause quite a bit of trouble if these power-consumption concerns are well-founded.


(that said, one could hook it up with a min-spec PSU and motherboard and it might well fall within tolerances...)

If the news about the PCIe slot handshake is true then the card should either run in limp mode or not work at all. I sincerely doubt people need to be worried about their PCs setting on fire or frankly even failed PCIe slots. I think a larger concern should be side effects related to the extra draw. Such as noise in on-board audio solutions.

titan wrote:

Topinio wrote:

chuckula wrote:

Power related things....

More power related things....

So, apparently, the 75W through the slot is just a starting number. Motherboard manufacturers who use better components, bigger traces, what have you, can provide more power through the slot. The adapter must request more power, and the PCIe controller must respond with the new limit. Whatever the PCIe controller says the new limit is, the adapter may not exceed it.

I'm not claiming to be an expert, but people much more knowledgeable in this matter are here:

https://www.reddit.com/r/Amd/comments/4 ... _solution/

Whomever wrote that up doesn't understand the spec, unfortunately.

The limit for *any* slot is 75 watts total power (with current slot design). Exceeding that violates the specification. There's room in the specification to allow for higher power slots, but they don't exist in the current physical form factor.

Does anyone on TR have access to the PCIe 3.0 spec and can look up the power specifications? I would look this myself, but .
And possibly (I'm pretty sure this one is only for PCIe 2.0 though.