GPU Utilization Investigations

One of the issues we experienced with the initial drivers on the Radeon HD 7970M was lower than expected performance, and many users/owners of the 7970M Clevo notebooks have pointed towards GPU utilization as being the culprit. That seemed a bit odd when I first heard about the topic, as low GPU utilization should be a symptom of lower than expected performance rather than the cause—basically, if the GPU is capable of running at 150 FPS at certain settings but it’s only running at 75 FPS, you’d expect the GPU use to be around 50%. With the initial P170EM review, I did some quick checking on GPU utilization and found that, yes, it seemed low. With the Hotfix driver (which still hasn’t reached the public, though we’re hearing “this week”—meaning, some time during October 15-19), we also noted that GPU utilization did improve. Nevertheless, GPU utilization isn’t “always 90% or more”, which is what some people have been asking for.

To investigate this issue more thoroughly, I tested GPU utilization across our three benchmark settings in six of the games. For these tests, I’m not using built-in benchmarks or even FRAPS runs; instead, I played the same segment of each game in a pattern that was as close to the same as possible, and I did this over a longer segment of the games: typically at least two minutes of gameplay, with a decent amount of “action” as appropriate.

After testing the P170EM with both HD 7970M and GTX 680M, I noticed that GPU use was relatively consistent but still occasionally low. At that point it seemed that Enduro and Optimus might be the cause of lower than expected GPU utilization. The other possibility of course is that the CPU and other areas were the bottleneck and that the GPU simply wasn’t able to get enough data to work on quickly enough. To test whether Optimus/Enduro might be causing issues, I added my own gaming desktop system into the mix; it has an i7-965X clocked at 3.65GHz, 12GB (6x2GB) DDR3-1333 RAM, and a GeForce GTX 580 GPU. I ran the same gaming tests on all three systems, and here are the results:

Value GPU Utilization

Mainstream GPU Utilization

Enthusiast GPU Utilization

Not surprisingly, GPU use is lower by far at the Value settings—1366x768 and ~Medium detail isn’t likely to tax the GPUs all that much, and the desktop system (with a faster GPU but potentially slower CPU) ends up with the lowest overall GPU utilization. The Mainstream testing pushes the GPUs quite a bit more, while at our Enthusiast settings most of the games are hitting 95% or higher GPU use. Looking at the games individually, we also notice that games that are typically more dependent on CPU performance (e.g. Skyrim), or at least not as demanding graphically (e.g. Portal 2), have lower GPU utilization.

I didn’t run a full set of tests using the earlier AMD 7970M drivers, but I can tell you that the Value and Mainstream results would be much lower, and even the Enthusiast settings are probably in the low 90s or high 80s on average. With the Hotfix driver installed, however, AMD’s 7970M ends up being the “best” overall at achieving high GPU utilization. I’m not sure that really matters all that much, as the real story is gaming performance. Whether low GPU utilization causes low frame rates or the low frame rates reduce GPU utilization, either way the fix is generally going to involve reworking the graphics drivers.

GTX 680M vs. HD 7970M – the Big Picture Clevo P170EM Battery Life Revisited
Comments Locked

58 Comments

View All Comments

  • Wolfpup - Tuesday, October 16, 2012 - link

    Yeah, I agree with everything there. It's...just disgusting to be insulting the author like that, and on top of that it's the commentor's "logic" that's iffy, not the authors.
  • krumme - Wednesday, October 17, 2012 - link

    "If this article compared two smartphones with the same numbers, would you make your same trollish complaint?"

    Yes. If a gamers phone was sold for 1150 usd compared to 1000 usd for exactly the same phone except gpu power.

    The total cost should be compared to the total benefit for the consumer. Even for a gamer, not everything is fps. There is a lot more to it when buying a machine. Therefore the argument is stupid.

    Jarred completely missed the total benefits, and only looked at the fps side. Thats okey, but then dont compare to the total cost. There is no consistency.
  • Wolfpup - Tuesday, October 16, 2012 - link

    No, his reasoning makes complete sense. It's the same reason you don't buy one of these notebooks and then opt for a GTX 660. The GTX 660 isn't BAD, but if you're already spending that much, and getting this notebook, it makes sense to get the best, particularly since you can't upgrade.

    This is about as cut and dried a choice as there's ever been-not an ad. The GTX 680 is just plain the fastest, AND it remains a reality that Nvidia is a safer choice even if it was slower, because they have more than a decade trackrecord with solid drivers, while AMD has...well, I'm not sure they're at 1 month yet, they keep screwing up, and then promising it'll be different.

    I *am* still very concerned about Optimus/Enduro though, and wish you could get these systems WITHOUT them at least as an option, without having to spend $400 extra on the "3D" screen.

    The M17x-R4 would actually be an easy choice for me *if* it didn't have Optimus. When you tack on the extra $400 for hte "3D" screen I'd be getting solely to get rid of Optimus...well, the price gets harder to stomach.
  • JarredWalton - Wednesday, October 17, 2012 - link

    You can disable Optimus/Enduro on the M17x R4 in the BIOS I believe -- I know at least Enduro can be turned off, so I assume it's both. They have physical muxes on the motherboard so that all the display outputs can be routed to the dGPU, which is the major thing preventing Clevo from supporting non-Enduro/Optimus. But personally, since I only run Windows, I have no problem with Optimus. It works well for all the stuff I've done; at worst I occasionally have to tweak a game with a custom profile.
  • transphasic - Saturday, October 20, 2012 - link

    I agree with your comments and rationale on this Jarred. Since I am an owner of the 7970m, I can speak to this issue with my own experiences and buyer's remorse at foolishly choosing the 7970m in the first place.
    It's really worth it to pay a little more for better quality, better drivers, and better support in choosing what to do in deciding what GPU to put in a gaming laptop.
    Why quibble over an extra $250 dollars or so, when you are already spending $2000 anyways?
    This strikes me as penny-wise, pound foolishness to try to scrimp and save $250-$300 dollars on choosing a flawed AMD 7970m product, that almost 4 months later, we STILL have not yet gotten proper driver support from AMD.
    (what's worse, is that AMD just laid off about 3,000 of their engineers who were probably working this so-called "hotfix" driver, so we might have to wait for a lot longer time now to get it, if we ever do. AMD is on shaky ground now, and that makes me even more nervous about their present and future).

    As for me, and the testing work that you did, Jarred, I am impressed with all the time and energies that you put into it, so thank you very much for your work on this, and after seeing that the "hotfix" still doesn't close the gap by that much, it leads me to the conclusion that I (and others as well) are better off going to Nvidia from now on. In fact, I am now going to take the next step in this, by swapping out my 7970m, and switching to the 680m, and be done with AMD.
  • bennyg - Monday, October 15, 2012 - link

    Conclusion mentions cooling is better in Clevo (vs MSI/AW). Where's the results of that test? AT reviews seem to be getting more focused on overanalysis and pennypinching comparisons compared with taking more measures of the actual notebook. Like idle/load/surface temps!

    Real enthusiasts also don't really care much about options other than CPU as we know most of the time you end up better off (especially with Dell/AW) buying the parts yourself and installing.

    Re the actual review, I don't understand why Clevo take a bad nonstandard keyboard, and find a way to make it even more annoyingly 'custom' and worse...
  • JarredWalton - Monday, October 15, 2012 - link

    The stress testing of the GTX 680M wasn't especially different from the HD 7970M:
    http://www.anandtech.com/show/6343/avadirect-clevo...
  • Freakie - Monday, October 15, 2012 - link

    If I'm reading those graphs right, GPU utilization is actually higher on the 7970 than the 680 on a number of games that the 680 still beats it at. Seems to me like even if GPU utilization were "fixed" to be even greater on both systems, then the 680 would still beat it out in most games and therefor the Utilization argument is kind of a weak one. The only games that it seems it would help with is a couple of games at Mainstream/Value settings which as you already said in the article, most people wont be bothering with. But of course to test GPU utilization THAT thoroughly would take an incredible about of time xP So it shall remain a mystery I suppose.
  • jigglywiggly - Monday, October 15, 2012 - link

    why are you hating on the look? I love the way clevos look
  • Brojo - Tuesday, October 16, 2012 - link

    I pretty much have that Clevo system except 16GB of RAM with the 7970. I knew I should of went with the 680 =p and kicking myself in the ass after seeing more and more comparisons. I will be optimistic and hope for better driver release but...if i want to swap cards It shouldnt be too difficult right?

Log in

Don't have an account? Sign up now