[Dave Baumann]

OK, now that HL2 is actually here, and it has a meaningful benchmark mode on it that will obviously go in the test suite. I had kinda decided that TR would be the one to drop out seeing as it is probably the least played. However:

• TR Does still stress DX9 capabilities
• UT2004 is nearly entirely CPU limited and only DX8
• Splinter Cell isn't relfective of what people are playing and its only DX8

So, I figured I'd get your thoughts on it as well.

[Natoma]

Benchmarks should reflect not only the current and future capabilities of the card in question, but they should also reflect in part what people are playing, so I voted for Splinter Cell to go rather than UT2K4.

[Cryect]

I say drop UT2004, kinda worthless having an almost completely CPU limited benchmark that is only going to get worse with time.

[Tim Murray]

UT2004 performance is CPU bound, so it's kinda meaningless. SC is interesting because it shows weaknesses in certain hardware (NV43).

[Razor1]

sad to say it but Tomb Raider for dx 9 good amount of shader usage, although poor optimizations.

[Jawed]

Drop TR, since it's possible to make HL-2 give a graphics card a hell of a time. TR benchmarks never seemed to tally with other DX9 games as far as I could tell. And who plays it anyway?

Oh and why not introduce 1920x1080/1200 resolution in benchmarking. If ATI is promoting the latest graphics cards for high definition gaming, then they should be tested at high definition settings. Sure it's a bit forward looking for the vast majority of us, but HD is receiving a lot of attention in enthusiast gaming circles as far as I can tell.

Jawed

[Dave Baumann]

Somebody dontate me a monitor that is capable of higher resolutions than 1600x1200 and I'll gladly start testing...

[Pete]

Surely TR:AoD is the least played out of all three? And you're benching it without the latest official patch, IIRC, which is an imperfect situation in and of itself.

UT2K4 is aging, but perhaps it'd be a good candidate for testing at insane resolutions with max AA. And it's always good as a sort of driver efficiency bench, no?

Splinter Cell's sequels still use its engine, don't they? Or is the original SC not indicative of the sequels' performance?

Edit: Dave, your CRT can't go higher than 16x12, even at very low refresh rates (say, sub 60Hz)? Have you tried using Powerstrip or something to add in a few custom refresh rates, like 19x14 and 20x15?

[Natoma]

Originally Posted by DaveBaumann

Somebody dontate me a monitor that is capable of higher resolutions than 1600x1200 and I'll gladly start testing...

What monitor are you using?

[ChrisRay]

Personally. I'd lose UT2004. You dont gain much from the benches of this software these days anyway. Not with modern hardware anyways.

[Richard]

I voted for UT2k4 as it's incredibly CPU limited and because I think there's nothing that game feature-wise has that you can't find in the rest of your benchmark suite.

My second choice to drop would be TR since we're starting to have enough DX9 titles for benchmarking.

My personal view on video card reviews (and even more so on B3D's reviews) is that what counts is the engine, not the actual game. if the number of players was a very important criterium then CS Classic would be in all benchmarks.

[Richard]

Originally Posted by Pete

Splinter Cell's sequels still use its engine, don't they? Or is the original SC not indicative of the sequels' performance?

Pandora Tomorrow uses pretty much the same engine, except with incredibly low resolution shadow maps (but has more light sources on any given map). On the other hand I think it requires more vertex muscle. SC3 is based on an entirely new renderer but it's not out yet.

[John Reynolds]

Originally Posted by DaveBaumann

Somebody dontate me a monitor that is capable of higher resolutions than 1600x1200 and I'll gladly start testing...

I've been thinking about buying a LCD, so if you want to pay for the shipping to have my KDS 21" CRT (2048x1536 @ 60hz) you can have it. :P

[Inane_Dork]

Easy: Tomb Raider. It has DX9 stuff, yes, but who cares how it performs on their machine? The other two are much more interesting, as far more people will actually care about the results.

[Reverend]

UT2K4 and SC should go. The former because it becomes a waste of time for reviewers and the latter because we'll have a ne- , er, never mind.

TRAoD is a bit difficult. Initially (i.e. when B3D first presented it to the world, ahem), it's quite unoptimized by the respective IHVs (drivers) and since its programmer follows coding "strictly by the MS books", it really shows hardware at the intrinsic level. Now, drivers are optimized to hell for it. It still has a fairly large DX9 featureset usage, something many DX9 games don't have.

Either B3D reviews hardware (i.e. use games that stress 3D hw no matter if the games are popular or not) or B3D reviews popular software.

[Geo]

Originally Posted by Reverend

UT2K4 and SC should go. The former because it becomes a waste of time for reviewers and the latter because we'll have a ne- , er, never mind.

TRAoD is a bit difficult. Initially (i.e. when B3D first presented it to the world, ahem), it's quite unoptimized by the respective IHVs (drivers) and since its programmer follows coding "strictly by the MS books", it really shows hardware at the intrinsic level. Now, drivers are optimized to hell for it. It still has a fairly large DX9 featureset usage, something many DX9 games don't have.

Either B3D reviews hardware (i.e. use games that stress 3D hw no matter if the games are popular or not) or B3D reviews popular software.

Ah yes. Good ole 'first principles'. B3D has never worried about being fashionable when it conflicted with the core mission, so why start now?

[sonix666]

Let Tomb Raider go. I can't imagine that alot of people are playing the game looking at all the negative reviews is has managed to gather. UT2k4 might be CPU limited, but it is a game being extensively used. And using CS classic as an argument against basing on player numbers is quite strange, because CS classic is fast no matter what system you use. UT2k4 is graphically more challenging for GPU and CPU.

[rwolf]

Review with them all. Don't drop anything.

[DarN]

I'm with Rev. (voted UT2004)

[chavvdarrr]

no one plays TRAOD.
It doesn't matter if it's CPU or GPU limited, what matters is if the game is played by people

[Martin Eddy]

For the types of reviews B3D does it shouldn't matter if people play the game or not. If I want to read a standard video card review there's <insert large number here> sites I could visit. I want my B3D reviews to focus on what the strengths and weaknesses of a particular card are, this is the exact same reason I like that you guys still use synthetic benchmarks.

P.S. I say get rid of Splinter Cell and UT2004 and then if HL2 turns out to be a good indicator of DX9 performance ditch TR:AOD.

[Hanners]

I would vote to dump UT2004, it isn't really indicative of anything any more from the point of view of a video card review.

[Tim]

UT2004 as it does not give us any relevant data (CPU limited) and everybody and their mother is doing UT2004 benchmarks anyway.

[Neeyik]

I see that everyone has forgotten that UT2004 is not always CPU limited - it depends very much on the graphics card being tested. The test might not prove much on something like a 850XT PE but for a 9550 it is useful. Also, UT2004 has higher texturing-to-arithmetic operations in a typical scene when compared to some of the other games we test with; that alone makes it useful as not every game on the market will be packed with shaders.

Splinter Cell annoys the crap out of me. No AA and using AF does exactly what you expect it to, regardless of what graphics card is used. A similar line of argument about AF can be put towards TRAoD but, as people have repeatedly pointed out, it does utilise DX9 features quite a lot and the graphics detail/settings can be widely adjusted (unlike SC).

I agree with people that the game itself used to test with is totally irrelevant - we're not creating full gaming situations here but rather graphical ones that have the possibility of existing in modern games. The raw frame rates aren't that important either because in real gameplay, they're not usually going to match those seen in benchmarks. What should be considered is having a range of graphics scenarios, with varying degrees of demand & feature usage and that means having a range of testing programs - just as we adjust the test settings for different levels of hardware, I believe we should do the same with the games we use. After all, who would want to see 6 pages of totally flat fill rates curves for a 9550LE ?

[anaqer]

Dump all of them, but start with TRAoD. No use of another quasi-syntetic benchmark (since a game that nobody plays is little more).