PowerVR SDK Updated - Includes stunning new Benchmark !

[Teasy]

I just do have to say that all of those PVR demos are optimized for the Kyro. Never have vendor-specific benches been any good. DOOM3, on the other hand, is being optimized for immediate-mode renderers

But the only way Fablemark is optimised for Kyro is that its using allot of stencil buffering, and this use of stencil buffering is very similar to Doom 3's use of stencil buffering. So if you say Fablemark is optimised for TBR's then how can Doom 3 be optimised for IMR's?

I particularly doubt that it will be able to handle the polycounts.

AFAIK Doom 3's polygon counts aren't even going to be that high. Isn't he actually using quite normal quality polygon models and just using a nice little trick to make them look more high poly? Also considering the style of the game (no big outdoor area's) I can't see the game having huge polygon counts. Unless anyone knows different?, if so I'd like to know the expected polygon counts for this game.

Actually considering the number of passes used in the game for most cards out their surely allot of HW T&L cards might end up faster just using software T&L? Because with HW T&L your going to have to do the T&L for every pass, E.G. a GTS would need to do the T&L around 4 or more times. So perhaps allot of HW T&L cards may end up without the use of their HW T&L engines?

 

[DemoCoder]

The T&L for many of the other passes consists of different vertices, not the same vertex 4 times. E.g. passes 2 .. N are stencil passes are various effects.

We will see heavy use of shadowing way before Rev's "2 years". A huge number of games on the GameCube and X-Box are doing this *right now*. Doom3 is doing it, and hence, every game based on that engine. Star Wars Galaxies, EQ2, etc. You can bet your bottom dollar that everyone else is working to copy Carmack's techniques (low-poly imposters, heavy use of shadow volumes) I've already seen another engine advertisement which is copying D3's character model trick.

 

[KimB]

Thats not optimising the engine for it though!

Nope, it's optimizing the game for it.

Given what John Carmack has stated, he's probably actually spent more time playing around with ATI's hardware, since he's been saying that ATI's hardware has more driver problems (If it works the first time on nVidia hardware, and not on ATI's, whose hardware are you going to spend more time with?).

Still, I really think JC is trying his best to make both ATI's and nVidia's video cards work as well as he possibly can, and I seriously doubt he's going to consciously optimize the engine more for one architecture than the other.

But, the fact remains that the game will most likely run best on GF4 hardware because that's what the designers are using (And yes, I _am_ leaving out unreleased hardware here...).

 

[psurge]

NVidia has stated that future hardware will allow a different stencil op to be executed depending on whether or not a triangle is front-facing or back-facing.


Serge

 

[KimB]

But the only way Fablemark is optimised for Kyro is that its using allot of stencil buffering, and this use of stencil buffering is very similar to Doom 3's use of stencil buffering. So if you say Fablemark is optimised for TBR's then how can Doom 3 be optimised for IMR's?

I don't know how similar JC's techniques are to fablemark, but, given the performance scores, I'd say they're wildly different. Otherwise JC would be hailing the Kyro 2 as the best card for DOOM3, and that by a wide margin. He hasn't even mentioned the Kyro 2, and we do know for certain that he's spent tons of time designing the engine to run well on GeForce3/4's and the Radeon 8500. Fablemark was designed pretty much exclusively on Kyro hardware, and we can be almost assured that there are things in there that were placed specifically because they run like crap on IMR's.

AFAIK Doom 3's polygon counts aren't even going to be that high.

No, because you need to retransform everything multiple times. This is where the Kyro will likely have fits. The resultant pressure on the T&L pipelines is going to be rather severe.

Actually considering the number of passes used in the game for most cards out their surely allot of HW T&L cards might end up faster just using software T&L?

You're going to have to re-do the T&L anyway, hardware or no. Many of the passes will involve rendering from different viewpoints (for shadows).

 

[Ty]

We will see heavy use of shadowing way before Rev's "2 years". A huge number of games on the GameCube and X-Box are doing this *right now*. Doom3 is doing it, and hence, every game based on that engine. Star Wars Galaxies, EQ2, etc. You can bet your bottom dollar that everyone else is working to copy Carmack's techniques (low-poly imposters, heavy use of shadow volumes) I've already seen another engine advertisement which is copying D3's character model trick.

My italics above but I wasn't sure if you were stating that SWG and EQ2 were based on the Doom3 engine or that they were just going to end up similar to it. Anyhow both of those engines are built in-house. I believe EQ2's engine will be similar to Doom3's technique but not SWG's.

 

[Tahir2]

For Kristof :

Radeon 8500LE using 6093 Drivers for Win XP.

 

[LittlePenny]

Nice work on the fire demo Kristof. Was FableMark done by one person or a bunch of people?

 

[Kristof]

Nice work on the fire demo Kristof. Was FableMark done by one person or a bunch of people?

2 People did it, one programmer and one artist, with at least 4 more people nagging about little details and features they would like to see added In case you are wondering I was one of the people that did the nagging

misae, the white is caused by inaccuracy in the vertex position, the initial lighting passes ar done with no texture (hence the white look) the texture is blended in at the last stage but gets an incorrect/inaccurate depth value and is incorrectly declared hidden by the hardware. Did you notice any weird staircase effects on some objects ? That issue is a bit hard to see with only 320x240 resolution.

 

[Randell]

Kristof,

yes I got staircased flashing textures. But disabling T&L and HyperZ cured it.

8500LE with 9031's on Win98SE

 

[Tahir2]

yes i noticed the staircase effect too.

nice benchmark by the way.

 

[Reverend]

Rev, DoomIII and FableMark have a lot in common in how they do lighting...<snip>

I know what you're saying re comparisons between this benchmark and D3. Now, you mentioned DOOM3, not me, in drawing comparisons. Let me get this correct :

According to you, this benchmark represents (closely?) D3's use of lighting and is therefore an indication of lighting performance specifically in D3 for whatever card is used for benchmarking it. Yes?

The benchmark gives no other analytical results other than a framerate figure.

So, according to your post (and Simon's reply to my post), especially mentioning D3 (you, not Simon) and in doing so leading me to believe this benchmark is an indication of performance in D3, am I to suppose a GF4Ti4600+AthlonXP2000+ will give me 26fps in D3? I would believe the number will be less in D3 since I think all other game considerations will have to be factored in. A GF4Ti4600 will run D3 very poorly then!

Okay, leaving all the acting above aside...

IMO, this benchmark should be taken for what it is. I do not think it was wise to to mention Doom3 at all. It may be useful for programmers but it is not useful for the public. It may even be of more use to programmers/developers if the results output additional data instead of just a framerate. As it is, looking at the results by various folks here (that clearly shows a KYRO beating all others) and with the mention of DOOM3 as a comparison of lighting schemes, it is not healthy.

 

[Dave Baumann]

So, according to your post (and Simon's reply to my post), especially mentioning D3 (you, not Simon) and in doing so leading me to believe this benchmark is an indication of performance in D3, am I to suppose a GF4Ti4600+AthlonXP2000+ will give me 26fps in D3?

You’re taking that well out of context Rev. Of course you can’t conclude the performance of DoomIII from this benchmark, but then that not what being said. The value of the benchmark comes when you have a suitably diverse set of benchmark results from different systems/video cards – then you can may gain an understanding of what certain video cards are like relative to another under circumstances that are like to be used in games, such as doomIII, that utilise similar features.

 

[PC-Engine]

Will D3 be TnL limited?

 

[Jerry Cornelius]

I think the test is very relevent to Doom III. Doom III is going to make significant use of stencil buffer, tilers are faster at stencil buffering than IMR'ers. The only piece of the puzzle left is whether the Doom III engine does things in such a way that causes the massive slowdowns you get when using stencil shadows in Quake III.

The Day of the Tiler Approacheth. I'd just die laughing if Unreal II and Doom III are playable on a Kyro II. I guess there's not much chance of that with out pixel shaders.

 

[Reznor007]

Nice work on the fire demo Kristof. Was FableMark done by one person or a bunch of people?

2 People did it, one programmer and one artist, with at least 4 more people nagging about little details and features they would like to see added In case you are wondering I was one of the people that did the nagging

misae, the white is caused by inaccuracy in the vertex position, the initial lighting passes ar done with no texture (hence the white look) the texture is blended in at the last stage but gets an incorrect/inaccurate depth value and is incorrectly declared hidden by the hardware. Did you notice any weird staircase effects on some objects ? That issue is a bit hard to see with only 320x240 resolution.

In case you were wondering...it works perfectly on Voodoo5 using the 1.04 drivers...although very slow (5-8FPS average).

 

[Doomtrooper]

Not bad for 2 year old technology....is Geometry assist enabled ?

 

[Reverend]

You’re taking that well out of context Rev. Of course you can’t conclude the performance of DoomIII from this benchmark, but then that not what being said. The value of the benchmark comes when you have a suitably diverse set of benchmark results from different systems/video cards – then you can may gain an understanding of what certain video cards are like relative to another under circumstances that are like to be used in games, such as doomIII, that utilise similar features.

There are some results given by some folks in this thread, aren't there? I see Kyro2s running quite a bit ahead of a GF4Ti4600!



There's a sentence in my post about my play-acting a bit in my post.

My point is that Doom3 should not be brought into this thread by Kristof. In this thread, Kyros are tops, Kristof then brings in Doom3 for comparison... and we know what Carmack told me about the Kyro WRT Doom3.

Enough, I've said my piece and I'm not the type to argue .

 

[Jerry Cornelius]

and we know what Carmack told me about the Kyro WRT Doom3.

Was this is the first round of questions?

 

[Kristof]

My point is that Doom3 should not be brought into this thread by Kristof. In this thread, Kyros are tops, Kristof then brings in Doom3 for comparison... and we know what Carmack told me about the Kyro WRT Doom3.

Maybe we should see it in a larger context... JC mentioned the issue that Stencil were slow (better he said that hardware is currently not optimised to handle the rather simple complexity of stencil pixels vs textured/shaded pixels), I brought this into this tread since FableMark illustrates JCs point. I am not implying that KYRO will rock, or not, with DoomIII. What FableMark does seem to imply is that PowerVRs Deferred Rendering System seems to have the ability to process Stencils very quickly.