New DX10 Benchmarks Do More Bad than Good

NIMBY writes "An interesting editorial over at PC Perspective looks at the changing status between modern game developers and companies like AMD and NVIDIA that depend on their work to show off their products. Recently, both AMD and NVIDIA separately helped in releasing DX10 benchmarks based on upcoming games that show the other hardware vendor in a negative light. But what went on behind the scenes? Can any collaboration these companies use actually be trusted by reviewers and the public to base a purchasing decision on? The author thinks the one source of resolution to this is have honest game developers take a stance for the gamer."

Comments

John Carmack

[dsanfte]

John Carmack used to be pretty good at cutting through the marketing crap and telling it like it was. Let's ask him.

Re:John Carmack

[peragrin]

Use OpenGL.

Re:John Carmack

[Applekid]

He seems to be less anti-DirectX these days:

"JC: DX9 has its act together well. I like the version of DirectX on the 360. Microsoft is doing well with DX10 on tightening the specs and the exactness."

Of course, he's still calls it like it is:

"The new features are not exactly well-thought-out. Most developers are pretty happy with DX9. The changes with DX10 aren't as radical. It's not like getting pixel shaders for the first time. Single-pass shaders are nice with DX10, but it's a smaller change. "

Re:John Carmack

[GrievousMistake]

For extra value we should also ask Theo de Raadt for a comment. And it would make a good House episode. "So what you're saying, Mr. NVIDIA, is that you got that driver bug from a public toilet seat?"

Re:John Carmack

[Anonymous Coward]

They made the specifications slicker and more exact, but did't add any new breakthrough features. Not contradicting at all really.

It's like "they reduced the needs for annual service on the car by 10%, but didn't add a turbocharger".

DX10

[Richard McBeef]

DX10 or for the uninformed, Derendering eXtraction (10 megapixels/second) is a standard benchmark for measuring the performance of GPUs or Gradient Pixilization Units. Pretty much this is what the video card companies all base their prices on with price being directly related to how many pixels can be gradiated per unit (usually about 30 cents per pixel/ounce).

quit already with 'optimized' drivers

[BattleTroll]

I'm getting tired of the back and forth between AMD and Nvidia. Drop the whole 'optimized' drivers crap and give us cards that work great out of the box. This entire trend of releasing per-game tweaked drivers is just hurting consumers. I shouldn't have to wait for Nvidia to tweak their drivers to get the best performance out of one of their cards. I shouldn't have to download new drivers every time a new games comes out. The whole reason you create your cards based on a known standard is to avoid this mess.

Stop fucking around and do it right the first time.

How hard is that?

Re:quit already with 'optimized' drivers

[ChronosWS]

This is only a problem if in the course of 'optimizing' for a particular use case they degrade performance in all of the other cases. There may be times, if the case is particularly widely used, that it might even be worth a small perf hit in one area to gain a large benefit in another.

You've got to remember, these guys live and die by sales. They *have* to look good in the numbers because that's what sells their cards at the top end. At the low end, no consumer cares either way as price dominates, but like automobiles, people assume that the tech from the top end trickles down to their lowly mass-market video hardware in some fashion, so it ends up still being relevant, if less directly so.

Also, if you have looked at most of these benchmarks, the difference between best and 2nd best is usually quite small, on the order of a couple percent. The bragging rights of being able to claim you can run your game at 150fps while other plebeans can only run at 140fps is just that - bragging rights. There is no practical effect on game play until framerates drop below 30fps. And the top end graphics hardware these days is not the bottleneck at resolutions of 1280x1024 and below, so really, these guys are chasing numbers in the rarified air of super high resolution monitors and games which use every trick in the book, which is an extremly small set of games actually played.

But that is what sells. And in any case, the competition between ATI and nVidia is good even if those of us who 'know' see their number-chasing as pointless. Let them do their thing, and reward or punish them at the counter as you see fit.

Re:quit already with 'optimized' drivers

[LWATCDR]

Do it right the first time? Spoken like someone that has never written software.
Every time you write a piece of code you improve it you may find a new way of doing something.
Also as more and more programs come out that use the driver they people that write them will gain a better understanding of how they are used. That will help them optimize a code path. It could be as simple as selecting which branch tends to be used more and making that the default path.
The something is used the more performance you can get out of it.

Shows how bad DX-10 really is

[Nom du Keyboard]

To me, this just goes to show what a bad standard/interface DX10 really is. Looks to me like if you make calls to it in one way, ATI shines, but call it another way and its all Nvidia -- yet both cards+drivers allegedly comply with the standard. It sounds like trying to compare floating point benchmarks on AMD Athlon versus Intel Core 2. Depending on how you arrange the numbers and call the various floating point extensions can make all the difference.

And there's no indication here if someone is using corked drivers that favor one game over the other.

What I'd like to see is a benchmark rundown of each function in DX10, along with some realistic estimate of how much each function is called in normal game play. If different games favor different functions, then say so. Only then might I have some idea of how the two graphics powerhouses measure up against each other.

And if you have some reasonable way of testing common sequences of calls, show that as well.

Does it really matter?

[Xest]

I'm not sure benchmarks really matter. It's not as if either of the cards are so bad that you're getting screwed by buying one instead of the other.

I've been using dedicated graphics cards since my old 3dfx Orchid Righteous 3D, since then I've had various ATI/nVidia cards and I've never been in the situation where I've thought "damn I wish I bought the other company's card".

I used to be someone that thought it was great to get 3 more fps than the other guy but when I came to realise that whilst I got 3 more fps in one game, and he got 5 more fps in another game that was OpenGL instead of DirectX or whatever. It became obvious that it's not as clear cut as one card is better than the other in terms of frame rates, it depends on the graphics API, the driver releases, the OS, the other hardware in the system, the game settings and on and on. Personally I prefer nVidia, but that's only because they have better developer support and I've had a better experience with the quality of their drivers over ATI's, image quality, features and frames per second has never once been an issue for me and I'm sure this is true for all but those people who think that getting an extra 3 more fps in game X actually makes the blindest bit of difference in the world.

Re:I thing, you thing, we all thing

[Traa]

To sink :-)

As my Comp Sci lecture loved saying...

[pls_call_again]

As my third year computer design lecture loved saying: There are lies; then there are damn Lies; and then there are benchmarks.