Radeon RX 480 performance revisited with AMD’s 16.7.1 driver

It’s been an exciting couple of weeks in the short life of the Radeon RX 480. AMD’s new graphics card is the first to use the next-generation Polaris 10 GPU, hot off the fabs. Unfortunately, not all of the excitement has been positive. Shortly after launch, the good people over at Tom’s Hardware and PC Perspective did some deep-diving on the new card’s power draw. The sites found that the card exceeded the specification for amperage drawn from the PCIe slot. This news ended up being a big concern for some potential RX 480 buyers.

In response, AMD released a statement saying that it didn’t think the 480’s excessive power draw was a threat to motherboards or other components in a system. Nevertheless, the company pushed out a new driver—Radeon Software 16.7.1—that re-balanced the board’s power draw, allowing it to pull more power from the PCIe 6-pin connector rather than the PCIe slot. The company also included a new option in the updated driver that would prevent the card from drawing quite as much juice in total for users who were particularly concerned about their systems. This “Compatibility Mode” can be toggled on and off in the Radeon Settings software, and it simply lowers the board’s power target. AMD is shipping the driver with this feature turned off by default.

Of course, making such low-level tweaks to the way the Radeon RX 480 works raised questions about how the card will perform post-update. Now that we’ve had some time with the new drivers and the new compatibility mode option we can give a clearer picture of the effect of this driver update on the games we tested. In the interest of time, we didn’t run our whole suite of benchmarks again. For the games that we did retest, we used the same set of settings and procedures that we used in our primary RX 480 review. We chose two games—Grand Theft Auto V and Crysis 3—to put back through the wringer, since they were both fairly playable at our test settings. 

Crysis 3 revisited

We’ll start with Crysis 3 today, since it’s both a performance test and the benchmark we use to generate our power-use numbers. Crysis 3 is also a potentially interesting test because it’s an older game that AMD isn’t likely to have released any specific optimizations for. As we noted, we used the same graphics settings, the same benchmark run, and the same processing methods as we did in our full RX 480 review. We compared the results from our initial test with the new 16.7.1 driver in its default state, and with the “compatibility mode” switch thrown.  Let’s see how much of a difference the new software makes, if any. 

Our average FPS numbers look good. No news here. Most importantly, there’s barely any change in performance between the release driver for the card and the Radeon Software 16.7.1 update. While the new driver does appear to increase performance slightly, these are averages of hand-run benchmarks, so the results are probably well within any margin of error. As always, we don’t put all that much stock in pure-FPS measures since they don’t show the whole picture of graphics performance.

Our frame-time graphs are a bit crowded, but they basically tell us the tale we want to see. Performance between the release driver and the 16.7.1 update is largely the same. Turning on Compatibility Mode does cause our card to display a couple large spikes that could affect perceived smoothness, but that’s unsurprising since we’re probably reducing its power budget—and, hence, performance—somewhat.


These “time spent beyond X” graphs are measures of “badness”—the amount of time that a card was displaying animation that may have been less than fluid, or at least less than perfect, during our test. If frames are taking longer than 33.3ms to render, the frame rate will drop below 30 FPS. In those cases, we expect to have some judder and other ugliness with vsync enabled on a 60-Hz monitor.

The new driver performs admirably here, although we still see a small increase in the amount of time spent beyond both 16.7 and 8.3ms. The compatibility driver fares a bit worse, with an almost doubling of the time spent beyond the key 16.7ms. This isn’t surprising, given the lower power use, and it also does a good job of demonstrating exactly how sensitive these frame-time measurements are.

 

Grand Theft Auto V revisited

We also re-ran our Grand Theft Auto V tests with the same settings, the same benchmark run, and the same processing as our initial RX 480 review. The charts below are once again made up data from the new 16.7.1 driver, both with compatibility on and off, as well as the RX 480 performance from the release review. Grand Theft Auto V received specific attention from AMD in the new driver to reduce “minor stuttering,” so we’ll be looking to see whether the card delivers on that point.

The frame time chart itself is crowded, but that’s probably a good thing. The Radeon RX 480 already did well by this measure with GTA V, and the fact that the three data plots mostly overlap one another indicates that there aren’t any major changes in smoothness from our initial review.

It’s not really a surprise that the RX 480 with the new driver outperformed the RX 480 with the old driver in raw FPS, since the game received specific optimizations in this update. It’s also nice to see that that the power changes of compatibility mode balance out with the improvements in the new driver. With Compatibility Mode off, the RX 480 gets a 7% boost from this update.

The 99th-percentile frame time of the RX 480 also improves in this game. The card is just a hair off delivering 99% of its frames within 16.7 ms, so gamers can expect a mostly smooth experience from this update. That’s not to say the RX 480 performed badly with GTA V in the first place, but installing the 16.7.1 update makes it just that little bit better.


In keeping with our measured 99th-percentile frame time result, the 16.7.1 drivers shave the time spent beyond 16.7 ms way down compared to the release driver. Whether in Compatibility Mode or in regular operation, the RX 480 spends little time beyond that critical threshold, and it even beats the GeForce GTX 970 (although it’s worth noting that card has also had a couple driver updates since our initial tests). That’s a result AMD’s driver team should be proud of.

What about that power issue?

While it’s good to see performance remain mostly unchanged with this driver release, the major question is whether AMD actually fixed the bus power draw issue that prompted so much outrage to begin with. The purpose of this driver update wasn’t to make the card draw less power overall, but to improve the balance of power use between the PCIe 6-pin connector and the PCIe slot. Unfortunately, with our current power-testing setup we don’t have the granularity required to detect a power use change at the PCIe bus, and measuring system power draw doesn’t tell the whole story.

Tom’s Hardware and PC Perspective have both released updated power numbers with the new driver, and both sites report that the new update lets the card draw less power from the PCIe slot and more from the beefier six-pin PCIe connector direct from the PSU.  With the new driver installed, and without compatibility mode enabled, PC Perspective found that the RX 480 draws about 6% more power from the PCIe slot than the standard would allow, as opposed to about 20% more with the release driver. While that number is still out of spec, it’s likely well within the capacity of any well-built motherboard to tolerate.

With the tools at hand, we discovered that turning on compatibility mode resulted in a roughly 5W drop in the power use on our test rig. That might not sound like a whole lot in the context of the whole system, but it’s a fairly big deal from a board-power perspective. Considering the fairly minor performance hit, it might be worth it to users who are particularly worried about whether their power supplies or motherboards are up to the task of running an RX 480.

All told, AMD’s update looks like it mostly resolves what could have been a sticky situation for the Radeon RX 480 reference card.  The company appears to have successfully rebalanced the Radeon RX 480’s power draw from the PCIe slot to the six-pin auxiliary power connector without negatively impacting performance. The “compatibility mode” switch should offer an extra margin of safety for owners of systems who are concerned about the card’s impact on their systems, at the cost of a tiny bit of performance. If you’re OK with the reference blower design, the RX 480 once again looks like the $200-$240 card to beat for now.

Comments closed
    • tootercomputer
    • 3 years ago

    Has Damage played a role in the development of this card/drigers? Anyone know?

      • sweatshopking
      • 3 years ago

      I think he was was helping with drigers a lot.

      • tipoo
      • 3 years ago

      I recall he was hired in part for the exact type of frame pacing testing TR pioneered, and sure enough it does pretty well there.

    • ronch
    • 3 years ago

    That’s all well and good, but I think, if I’m buying am RX 480 I’d just get one with an 8–pin aux power connector.

      • willmore
      • 3 years ago

      And a non-blower fan.

    • Klimax
    • 3 years ago

    AMD is major inventor in ways to shoot yourself in the foot. They could have not only avoided the mess and violation of PCI-E specs but could have apparently even improve some performance metrics.

    Good job AMD for another stellar lunch…

      • ronch
      • 3 years ago

      I think you meant ‘launch’.

        • maxxcool
        • 3 years ago

        Na he meant Doritos …

    • kuttan
    • 3 years ago

    TR could have used the latest 16.7.2 driver which had more performance optimizations.

      • JustAnEngineer
      • 3 years ago

      It’s a moving target. AMD’s driver team has been providing updates faster that TR’s review team can keep up.

      The drivers covered in this review addressed the concerns over PCIe slot power draw. TR’s testing showed that performance wasn’t crippled to achieve this feat. Further performance optimizations are sure to come as the drivers for this new GPU generation mature and as more DirectX 12 games arrive.

        • willmore
        • 3 years ago

        Did you just say that AMD was issuing fast driver updates? Let that soak in for a bit. Maybe old dogs can learn new tricks.

    • chuckula
    • 3 years ago

    Good news everyone!

    Newegg’s got a combo:
    [url<]http://www.newegg.com/Product/ComboBundleDetails.aspx?ItemList=Combo.2990853&nm_mc=AFC-C8Junction&cm_mmc=AFC-C8Junction-Veeralava%20LLC-_-na-_-na-_-na&cm_sp=&AID=10446076&PID=6202798&SID=NIS_926[/url<]

      • MOSFET
      • 3 years ago

      It’s the magic combination of elite gaming hardware from the last 5 years, together in one convenient bundle. Good thing we have those refreshed 990FX boards to house this stuff. And we all know that the staying pure with an AMD CPU, GPU, northbridge, and southbridge provides a smoothness to the experience which cannot be matched by competitors. Throw in the Radeon SSD and RAM, and the old AMD (Thermaltake) dragon (oops, scorpion) case and [i<]damn![/i<] [url<]http://www.newegg.com/Product/Product.aspx?Item=N82E16811133200&ignorebbr=1&cm_re=thermaltake_amd-_-11-133-200-_-Product[/url<]

    • Shobai
    • 3 years ago

    Thanks for tidying up the frame number plots, Robert, they’re looking good. I note that you’ve left the GTX 970 results off them. At first I thought “fair enough: the chart would be too busy with the 970 plot, and I could just go back to the previous review to see how it fared against the original RX 480 plot”, but then you say at the end of the article that

    [quote<] although it's worth noting that card (i.e. The GTX 970) has also had a couple driver updates since our initial tests[/quote<] This leads me to think that the GTX 970 results perhaps should have been retested, rather than harvested from the previous review. Or am I off base there? If GTX 970 results are recycled, I would suggest calling that out (along with the note I've quoted above) before presenting the first results, so that it's clear what we're seeing (or not seeing, in the case of the Frame Number chart). If they are fresh, perhaps a note as to why they aren't on the chart would also be appropriate. Thanks again for your work

    • TheMonkeyKing
    • 3 years ago

    Hey, with that fix, are you going to buy one of the reference cards?

    My answer: [url<]https://pbs.twimg.com/media/B-JwCV6CAAAJ1Rl.png[/url<] (I don't care, over spec is still over spec. And that's not even overclocking it.)

    • Chrispy_
    • 3 years ago

    [quote<]Our frame-time graphs are a bit crowded[/quote<] They wouldn't be if the graph started at 95% instead of 50%. 😉 Everyone knows it's a pretty straight line up until the last few percent anyway.

      • Shobai
      • 3 years ago

      The capability for having tabs on the frame number graphs could be extended to the frame time graphs, so that you could see the 50+ sweep for the overall view and then focus in on the tight end for the details.

      How about it, TR?

        • jensend
        • 3 years ago

        Instead, the axis should simply not be linear. When TR first introduced the continuous quantile plots in 2012 [url=https://techreport.com/news/22666/etc?post=623948<]I suggested[/url<] that they switch to using a semilog plot. The bottom axis would most naturally have even markings along the lines of e.g. 50%, 75%, 87.5%, 93.8%, 96.9%, 98.4%, 99.2%, 99.6% rather than 50%, 60%, 70%, 80%, 90%, 100%.

          • JustAnEngineer
          • 3 years ago

          Many folks don’t understand fractions. Asking the readers to grok exponentiation may be too much.

            • jensend
            • 3 years ago

            Huh? You don’t need to understand exponentiation or logarithms to understand the gist of what’s presented on a semilog plot. Only if you want to draw your own by hand.

            My example frametime quantile semilog plot I linked from that page has disappeared from the web and it’d take a little bit of digging for me to find the stuff again, but it’s no harder to comprehend than the present quantile plot. Semilog plots like [url=https://upload.wikimedia.org/wikipedia/commons/0/00/Transistor_Count_and_Moore%27s_Law_-_2011.svg<]this[/url<] aren't uncommon or hard to understand either.

          • Mr Bill
          • 3 years ago

          +++ I made a similar suggestion a while ago. Log linear plots of cumulative frequency versus concentration are used to plot things like the distribution of gold in a population of ore samples or the distribution of wealth in a population of people. In the case of gold, having two linear regions is good. It suggests there is a mineralized population. In the case of distribution of wealth it can be inferred that the mechanism of acquisition of wealth is different and indeed one can clearly see the break in slope between wage earners and “other” means of aquisition. So, the visually linear stretches can be interpreted as similar populations. Its possible that in a video card rendering frames, that there are similar population statistics going on and so it would be interesting to compare the slopes and linear regions for a group of cards as they change in response to different games.

          • Shobai
          • 3 years ago

          Why not both? [mexican dancing] [/mexican dancing]

          Linear scales are easier for lay people to read. TR could keep the existing plot and add a second tab, possibly employing a semilog style for clarity.

      • anotherengineer
      • 3 years ago

      I will leave this here

      [url<]http://edwardtufte.com.s3.amazonaws.com/colorblind_palette.png[/url<] edit - full article [url<]http://jfly.iam.u-tokyo.ac.jp/color/[/url<]

        • Chrispy_
        • 3 years ago

        +9000.
        I can concur this works not just for deuteranomolies (1-in-7 men) but also for deuternaopia (1-in-200 men).

        Trust me – as a deuteranopic this is both rare and satisfying, so I’ve bookmarked that article to share among my dev friends.

    • DPete27
    • 3 years ago

    Did I miss the part in this review where you explained what Compatibility Mode actually changes to achieve the lower power draw (clockspeeds)?

      • Jeff Kampman
      • 3 years ago

      Whoops, I forgot a sentence. It lowers the board’s power target, which is doubtless one of the inputs the card uses to determine its DVFS curve.

    • Waco
    • 3 years ago

    I’m still waiting for results of power draw when overclocking, as it sounds like the card is just barely in spec with compatibility mode enabled, and still slightly violates it when it’s not. With some decent overclocking you’re still looking at nearly 100 watts drawn from the slot…

      • flip-mode
      • 3 years ago

      Wait for cards from partner OEMs. They will have better coolers than the reference design and I expect a lot of them will have custom design PCBs. I’ve heard of one already that will come with an 8-pin PEG plug instead of 6-pin.

        • Waco
        • 3 years ago

        I’m just surprised AMD managed to screw this design up. I’m not in the market, but I do build systems for others often enough that I worry about this kind of thing.

      • DarkUltra
      • 3 years ago

      I’m also looking for this. How much current do “proper” 3D cards draw from PCIe slot at default and how much when ovetclocked.

        • Waco
        • 3 years ago

        In general, not much at all. I’ve only seen a handful of measurements, but even the 295X2 was essentially capped at 30 watts.

        • Voldenuit
        • 3 years ago

        [quote<]I'm also looking for this. How much current do "proper" 3D cards draw from PCIe slot at default and how much when ovetclocked.[/quote<] PCPer has been testing this. They tested the 480, 960 and 1060. Both the nvidia cards tested stayed well below spec (remember, 75W is for 12V and 3.3V combined, so cards only have 66W from the 12V line to stay in spec).

    • sweatshopking
    • 3 years ago

    Heres the problem: These things are going to end up 350$ish+ Canadian. A 970 can be found for 275. This card isn’t worth the price increase. If if was 40$ cheaper, sure. Without a doubt. But it isn’t, and i don’t think we will see it fall that far for a while.

      • tay
      • 3 years ago

      You could get a RX 480 4GB for the same price as a 970.

      • BurntMyBacon
      • 3 years ago

      I don’t see any GTX970 model cheaper than 390CAD on Newegg.ca at the moment. The RX480 4G currently goes for 275CAD and the 8G model goes for 320CAD on Newegg.ca (in and out of stock). I dropped the lowest priced outliers for the RX480. Newegg.com charges 275USD, 200USD, and 240USD for the GTX970, RX480 4G, and RX480 8G respectively. Where are you getting your prices? Are you sure you aren’t comparing RX480 CAD prices to GTX970 USD prices?
      [quote<]Heres the problem: These things are going to end up 350$ish+ Canadian. A 970 can be found for 275.[/quote<]

        • sweatshopking
        • 3 years ago

        Whoa. Hadn’t seen the 4gb for sale in Canada! Much better looking prices!

      • End User
      • 3 years ago

      [url=http://www.canadacomputers.com/product_info.php?cPath=43_1200_557_559&item_id=077305<]Cheapest 4GB 970 I found goes for $369 CDN (mail in rebates).[/url<] [url=http://www.newegg.ca/Product/Product.aspx?Item=N82E16814131693&cm_re=rx_480_4gb-_-14-131-693-_-Product<]The RX 480 4GB can be found for $260 CDN.[/url<]

      • interbet
      • 3 years ago

      Unless you’re getting a 970 for a deep discount, it makes no sense to get one at this point. Newer titles and a reduced inclination from Nvidia to spend resources optimizing the drivers for Maxwell performance (especially with DX12) will probably make anyone buying one today depressed by end of year, unless they plan to never buy a new game title.

    • 1sh
    • 3 years ago

    Now only if the RX 480 was available for purchase… O_O

      • flip-mode
      • 3 years ago

      yeah you will never find one…
      [url<]http://www.newegg.com/Product/Product.aspx?Item=N82E16814202221[/url<]

        • 1sh
        • 3 years ago

        I can’t purchase something that is not in stock. Besides, I don’t want that reference cooler junk.

          • sweatshopking
          • 3 years ago

          I do! 8gb of ram for the price of 4!

          • JustAnEngineer
          • 3 years ago

          You can choose from [url=http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&IsNodeId=1&N=100007709%20601203818<]more than one[/url<] Radeon RX-480 model that's currently in stock, if you're willing to pay Newegg's early-adopter premium. Wait [url=https://techreport.com/news/30395/asus-rog-strix-radeon-rx-480-takes-flight-in-august<]a month[/url<] to see the best non-reference versions. I'm also curious to see where Radeon RX-470 will fit into the price/performance landscape.

    • willyolio
    • 3 years ago

    Is TR also going to do an Async compute benchmark in future reviews (like when the 1060 comes out)?

    With RotTR adding async support, and Doom adding Vulkan support, etc., I think it would be useful to see how these new cards stack up to the new APIs. This really looks like how the industry will be trending towards so it should have its own category of testing.

      • Jeff Kampman
      • 3 years ago

      We would like to test this. The problem is that there’s not a public build of any of these games with a toggle to enable or disable async compute support available right now. 3DMark Time Spy does expose such a toggle, though, so we may have to rely on it to determine whether cards are getting a boost from the feature for now.

        • odizzido
        • 3 years ago

        I’d be interested in seeing some doom CPU benchmarks with vulkan. Even if it’s just a quick one with a couple CPUs from both companies.

        • Shobai
        • 3 years ago

        I thought I had read that Doom had such a toggle, but I may have been mistaken. Even so, a single game doesn’t give you much to go with…

          • Jeff Kampman
          • 3 years ago

          Doom flips async on and off depending on the type of anti-aliasing you employ, which isn’t ideal for head-to-head testing.

            • Shobai
            • 3 years ago

            Yep, the “TSSAA or none” thing, as per the tweet? [edit: I could have sworn I’d read people talking about a menu option, but I don’t have the game to check. I’ll take your word for it, anyway]

            I guess the closest you could get would be to test contenders on OpenGL at TSSAA, none and another AA level, then retest under Vulkan at the same set – a lot of work for the lack of a toggle!

        • Philldoe
        • 3 years ago

        Ashes of the Singularity. DX12. Has an ini option to disable Async. Hi5!

        EDIT: If you need a game key or two to get the game and use it as a benchmark, I can likely get you some straight from the devs.

      • DPete27
      • 3 years ago

      Unfortunately not a TechReport article ( well technically [url=https://techreport.com/news/30394/3dmark-time-spy-benchmark-puts-directx-12-to-the-test<]it is here[/url<] but: [url<]https://www.pcper.com/reviews/Graphics-Cards/3DMark-Time-Spy-Looking-DX12-Asynchronous-Compute-Performance[/url<]

    • odizzido
    • 3 years ago

    480 has had some pretty solid improvements since release. By the time the 1060 comes out you guys are going to have to redo all of the 480 testing you’ve done. Sounds like fun times 😛

      • PrincipalSkinner
      • 3 years ago

      Not if you’re a reviewer under pressure.

      • DPete27
      • 3 years ago

      Yeah! Now the RX480 finally where it should have been when it first launched!!

    • DrCR
    • 3 years ago

    AMD handled this much better than I anticipated. With this last wave of results published, it’s basically non-news now, which as good as it gets for AMD (unless future drivers at some point cause the issue to reoccur).

    I’m still holding out a bit longer before the next by, but AMD is looking good enough to get my wallet vote, and that’s saying a lot considering my last ATI card was the 9800 Pro. (I actively use the Linux binary driver.)

      • stefem
      • 3 years ago

      The have already come out with a fix in the linux driver?

    • Mr Bill
    • 3 years ago

    The hospital I am visiting is blocking Tom’s hardware… SonicWALL claims its due to… Jscript.Dloader.G_17 (Trojan). Hmmm…

      • swaaye
      • 3 years ago

      I was on here a few days ago and the page popped to some bogus “Firefox update” page and offered a wonderful .js file. I haven’t seen it since though. Maybe a rogue advertisement.

      • DrCR
      • 3 years ago

      For geeks, modern web browsing entails NoScript.

        • Mr Bill
        • 3 years ago

        Rule 12

    • Sam125
    • 3 years ago

    The tech industry is so much drama these days, do not like.

    • OneShotOneKill
    • 3 years ago

    It is looking prettier. Now get that ugly cooler off of it and give me some 8 phase power delivery module so that we can go beast mode.

    Sapphire, Asus I am talking to you.

    • chµck
    • 3 years ago

    It took me a whole 5 seconds to figure out that (CM) was (Compatibility Mode).

      • BurntMyBacon
      • 3 years ago

      Was that [b<]whole[/b<] 5 seconds within the Spec tolerances?

        • chµck
        • 3 years ago

        No, I became increasingly frustrated after 2.5 seconds.

          • Redocbew
          • 3 years ago

          Sounds like someone needs to switch on your “compatibility”.

    • Freon
    • 3 years ago

    Neat that they actually rebalanced power draw from slot to PCIe 6 pin. I was worried they would not have the granular controls required to accomplish that.

      • willmore
      • 3 years ago

      PCperspective mentions it in a post a week ago. There are six phases to the power system. Three draw from the slot and three from the 6 pin connector. They basically reprogrammed the switching mode power controller to just use the three phases on the 6 pin more.

      Lots of comments along the lines of “why didn’t it ship that way?”

        • Mr Bill
        • 3 years ago

        I wonder if the board is meant to be a platform for several versions. Others have pointed out that those VRM’s are hugely over-specified for the load and just one or two of the 6 could run the card.

          • stefem
          • 3 years ago

          They wouldn’t have put the component on then, leaving the place to adapt to more powerfull cards

        • stefem
        • 3 years ago

        They have to reprogram at each boot so every driver must contain the fix, I’m wondering what’s the situation on linux.

        Edit:
        Lol, down thumb AMD not me, have not developed the card nor have chosen to fit it via driver. Would have you preferred to update the card’s BIOS?

          • OneShotOneKill
          • 3 years ago

          Translation: Current AMD drivers team fixed a problem that it will later forget to include in future upgrades.

          This will become a legacy neglected fix in a year or two.

          • Questar
          • 3 years ago

          I would think initial VRM programming is accomplished via VBIOS.

            • stefem
            • 3 years ago

            Depends on the controller but can be override via driver, AMD chosen to avoid a delicate BIOS update

Pin It on Pinterest

Share This