Page 1:GeForce GTX 570: Now That's More Like It
Page 2:Meet Nvidia’s GeForce GTX 570
Page 3:Tessellation: Unigine Gives Us Synthetic Numbers
Page 4:Tessellation: HAWX 2 Gives Us Real-World Numbers
Page 5:The BS Of Benchmarking
Page 6:Test Hardware And Software
Page 7:Benchmark Results: 3DMark Vantage
Page 8:Benchmark Results: Metro 2033 (DX 11)
Page 9:Benchmark Results: Lost Planet 2 (DX 11)
Page 10:Benchmark Results: Aliens Vs. Predator (DX 11)
Page 11:Benchmark Results: Battlefield: Bad Company 2 (DX 11)
Page 12:Benchmark Results: DiRT 2 (DX 11)
Page 13:Benchmark Results: Just Cause 2 (DX 11)
Page 14:Power Consumption And Noise
The BS Of Benchmarking
There is a lot of jockeying for position that goes on behind the scenes in the graphics business. This isn’t new; it’s been going on for years. Things really escalated ahead of the Radeon HD 6800-series launch, though. Nvidia was playing around with pricing and individual overclocked cards, AMD tried to discredit upcoming discounts on GeForce GTX 400-series boards, Nvidia pushed the HAWX 2 demo super hard before the game was available to play, and, apparently, AMD started making some changes to its drivers.
As a general rule, I try to remove myself from anything that doesn’t directly impact the way you and I both play games. After all, do the results of a game that hasn’t launched yet mean anything when they come from a demo? Does it matter if price cuts are temporary if you can realize the value of a discount today? So much of the stupid crap that goes on doesn’t matter, at the end of the day.
An Issue Of Image Quality
Well, a while back, Nvidia approached me about an optimization that had found its way into AMD’s driver, which would demote FP16 render targets, improving performance quantifiably. But after talking to both parties, there were three reasons I had a hard time getting too worked up: 1) the optimization affected a handful of games that nobody plays anymore, 2) identifying the quality difference required diffing the images—in other words, the changes were so slight that you had to use an application to compare the per-pixel differences, and 3) most important, AMD offered a check-box in its driver to disable Catalyst AI, the mechanism by which the surface format optimization was enabled.
Catalyst 10.10: Here's the option to disable Catalyst AI explicitly.
Now, you could say that the principle of changing the image from what the developer intended, in and of itself, is reason to take up arms against AMD. Indeed, altering image quality is a slippery slope, and what I might find inoffensive might immediately be noticeable to you. That just didn’t seem like a battle worth fighting, though.
And We’re Sliding…
But maybe it was, after all. In its most recent Catalyst 10.10e hotfix release, AMD removed the option to disable Catalyst AI. Instead, there’s now an option to turn that specific Surface Format Optimization on or off. The Catalyst AI texture filtering slider only gives the option of Performance, Quality, and High Quality, though, with no explanation of what’s being tuned. Without the option to turn off Catalyst AI altogether, I was less comfortable with optimizations AMD might make in the background for the sake of its performance (this works itself out in the end; keep reading).
What’s more, the change’s timing was suspect. I didn’t have time to dig too far into Nvidia’s recent accusation that AMD is tinkering with texture filtering quality in its newest drivers, mostly impacting the Radeon HD 6800-series cards. However, a number of German Web sites announced that they’d disable Catalyst AI entirely for testing Radeon HD 5800-series cards in order to achieve comparable image quality. And then the option to disable Catalyst AI disappears entirely? Things that make you go hmmm…
Perhaps even that wouldn’t have been so bad. But then AMD holds a conference call with the German Web sites that rode its back over quality concerns first brought to light after Radeon HD 6850 and 6870 launched. Fortunately, translating German is something we do pretty damn well (special thanks to Benjamin Kraft at Tom’s Hardware DE).
The recounting of the meeting published by ht4u.net said that the new High Quality setting in Catalyst AI purportedly disables all optimizations, akin to what the Disable Catalyst AI option did in the past, while the Quality option strikes a balance between what AMD considers performance and image quality. In other words, the company says it is not purposely doing anything in software that would diminish image quality. We asked AMD for comment on the new slider, and it confirmed that setting High Quality indeed turns off optimizations, addressing the first concern we brought up.
At the same time, though, AMD acknowledged to the Germans an image quality issue that was causing the shimmering originally reported by those sites with the 6800-series cards, even with High Quality toggled. The conclusion the site drew was that it’d take a hardware revision to correct the issue. AMD tells us it can fix the problem via drivers, and that it’ll involve blurring the textures so they don’t shimmer.
With all of that said, I went through seven different modern DirectX 11 titles looking for problematic areas that’d make for easy demonstrations, and, despite knowing about the issues and squinting at a 30” screen, came away with very little conclusive in the real world. These are subtle differences we’re talking about here.
Representatives at Nvidia did shoot over a handful of different videos comparing GeForce GTX 570 to Radeon HD 5870 and Radeon HD 6870 using Crysis Warhead. In them, the shimmering is painfully clear. Additionally, we were directed to a multiplayer map in Battlefield: Bad Company 2 that’d show it. But running around the single-player campaign didn’t turn up anything blatant.
Moving forward, I'm inclined to use AMD's High Quality Catalyst AI setting for graphics card reviews, if only to get the best possible filtering quality. Unfortunately, Nvidia is going to have to compete against the Quality slider position today, though, as it sent its GeForce GTX 570 with far too little time to retest everything we ran for the GeForce GTX 580 review less than a month ago. For the time being, we're going to have to leave everything at default and point out the observed and confirmed image quality issue currently affecting Radeon HD 6800-series cards. This may or may not become a factor in your buying decision, but right now, the bottom line is that Nvidia offers better texture filtering, regardless of whether you’re one of the folks who can appreciate it.
Why do we care if the difference is so nuanced? Because we have to take any deviation in image quality seriously—we don’t want to see Nvidia drop its own texture filtering default to help boost performance (yes, Nvidia has its own set of filtering-based quality settings it could tool with as well). Hopefully, that sort of tit-for-tat isn’t on the table in Santa Clara.
- GeForce GTX 570: Now That's More Like It
- Meet Nvidia’s GeForce GTX 570
- Tessellation: Unigine Gives Us Synthetic Numbers
- Tessellation: HAWX 2 Gives Us Real-World Numbers
- The BS Of Benchmarking
- Test Hardware And Software
- Benchmark Results: 3DMark Vantage
- Benchmark Results: Metro 2033 (DX 11)
- Benchmark Results: Lost Planet 2 (DX 11)
- Benchmark Results: Aliens Vs. Predator (DX 11)
- Benchmark Results: Battlefield: Bad Company 2 (DX 11)
- Benchmark Results: DiRT 2 (DX 11)
- Benchmark Results: Just Cause 2 (DX 11)
- Power Consumption And Noise