A request for reviews of graphics cards

[Oxford]

Hello,

I strongly hope that Techpowerup will be lead the pack by including minimum frame rates in important tests like Unigine, Metro, Crysis, and similarly demanding titles.

The following shots show why this may be especially important. Compare the frame rates of the 580/570 to the 480. So far, no site -- including the site the results are from -- has addressed this dramatic result. Instead, I've seen many sites stop using Unigine and practically none post minimum frame rates if they do.

http://techgage.com/reviews/nvidia/geforce_gtx_580/unigine_heaven_1680x1050.png
http://techgage.com/reviews/nvidia/geforce_gtx_580/unigine_heaven_1920x1080.png
http://techgage.com/reviews/nvidia/geforce_gtx_570/unigine_heaven_1680x1050.png
http://techgage.com/reviews/nvidia/geforce_gtx_570/unigine_heaven_1920x1080.png

Minimum frame rates can say a lot about playability, beyond what the average does.

The 480 has much better minimum frame rates in Unigine 2.1 than the 580/570 at resolutions below 2560x1600, something that is either related to the architectural change from GF100 to GF110 or drivers. If it's due to the architecture, then readers will likely want to know about it because Unigine may offer a useful perspective into how well a GPU will stand up to future games that leverage DirectX 11 more. Oddly enough, the 580/570 do slightly better than the 480 at 2560x1600 -- which suggests to me that this isn't a driver issue but an architectural one.

I would like to see a more in-depth examination of the current GPUs when it comes to Unigine. For instance, that site only had tessellation set to "normal". It didn't test extreme.

 

[Steevo]

With a card that achieves a average frame rate of 60 with some eye candy most users will turn down the eye candy or drop the resolution to play it.


I can attest to this as Intel video is still a decent percentage of users in the Steam hardware survey.



And the fact is that all your links aren't a game, but a benchmark program. So attaining 60FPS in is means nothing in the real world. Only highly unoptimized games like Metro will have the same results, and even the 580 means it is unplayable due to crappy coding at anything above 1990's display settings.

 

[copenhagen69]

www.techreaction.net has done this since the beginning of their reviews ... you are not looking hard enough

 

[Oxford]

www.techreaction.net has done this since the beginning of their reviews ... you are not looking hard enough

Never heard of them, although it doesn't really address the fact that my request was about this site. Thanks for the tip, though.

 

[Oxford]

With a card that achieves a average frame rate of 60 with some eye candy most users will turn down the eye candy or drop the resolution to play it.

Dropping the resolution leads to the 580 losing badly to the 480 in terms of minimum frame rate in Unigine. Plus, I'm not really following your point. What does 60 FPS have to do with this, other than not being even close to the minimum frame rate results posted?

30 FPS is often considered a basic threshold and the 480 is able to meet it while the 580/570 don't. That's interesting.

And the fact is that all your links aren't a game, but a benchmark program. So attaining 60FPS in is means nothing in the real world. Only highly unoptimized games like Metro will have the same results, and even the 580 means it is unplayable due to crappy coding at anything above 1990's display settings.

Unigine is supposed to be an engine that will be used in upcoming DirectX 11 titles. Also, there are games out now that give the 480 the lead in minimum frame rate (over the 580), although the results aren't as dramatic -- likely because the game, StarCraft, isn't as advanced as Unigine. In Crysis at 1920, Anandtech also found that two 480s in SLI beat the 580s in SLI in minimum frame rate. That site didn't bother to post minimums for anything else, though.

It also turns out that the 580 beats the 480 in "unoptimized" Metro.

 

[copenhagen69]

Never heard of them, although it doesn't really address the fact that my request was about this site. Thanks for the tip, though.

I know ... just saying

 

[Fourstaff]

Minimum framerates sometimes can be confusing and misleading, for example there are benchmarks which starts "slowly", ie the initial fps will be horrible single digit ones, and that can be extremely misleading if the reviewer is not careful. I am not saying that the reviewers are bad generally, but they are human too and they can be careless. Also, some benchmarks have "money shots", where the benchmark will show some nice (but excessively hard to render) vistas which can drop the framerate dramatically but will not affect general gameplay that much.

I do get your point but AMD's graphics cards generally provides a less stable framerate (ie it fluctuates more) than Nvidia's but they will perform slightly more consistently, but Nvidia's graphics card usually costs ever so slightly more than AMD's offerings.

 

[W1zzard]

Minimum framerates sometimes can be confusing and misleading, for example there are benchmarks which starts "slowly", ie the initial fps will be horrible single digit ones, and that can be extremely misleading if the reviewer is not careful

that. if you have 1000 frames at 100 fps and a single frame drops down due to a hdd access your result is completely wrong.

need a more mathematical approach to that. and actually capture the frametime for each single frame, then extract the actual game frame times and then calculate min fps

 

[Oxford]

I've never seen single digit minimum frame rates in test results from anything but low-end cards.

There definitely is a way to avoid the described situation. And, it is important to not just see the average. Do you think your scenario is what explains the results I posted? If so, please specifically explain how.

Also, some benchmarks have "money shots", where the benchmark will show some nice (but excessively hard to render) vistas which can drop the framerate dramatically but will not affect general gameplay that much.

But if a card like the 480 manages to keep a game above 30 fps and another card that people claim is faster goes down to 15, isn't that at newsworthy? I want to know why the 480 can so dramatically outperform the 580 in minimum frame rate in Unigine (and to a lesser degree in StarCraft). If a card is powerful enough to handle "money shots", maybe it will be more playable in future titles even if the average framerate is lower than with other cards. That's worth finding out.

If the average framerate masks stuttering problems then it's just as much of a problem as a minimum frame rate test exaggerating a card's potential slowness.

Take Anandtech's reviews where every tiny aspect of a design is examined when a new chip comes out, and yet when it comes time to test the cards the minimum frame rates are only provided for Crysis. It doesn't make much sense. And, when the GF110 was reviewed, there was a vague mention of die shrinkage, without (as I recall) much detail about what was cut.