Like Ailuros said, in OpenGL it's only 4 textures per pass. And it doesn't have the sophisticated combiners of the competition.ET said:Tagrineth said:DX7 - Kyro - 8 textures per pass.
Yeah, forgot about that. Thanks! Now, it would have been interesting to see Carmack add a Kryo code path to Doom3.
DX6 style MT?
- Thread starter K.I.L.E.R
- Start date
ET said:Yeah, forgot about that. Thanks! Now, it would have been interesting to see Carmack add a Kryo code path to Doom3.
It's quite tricky to do vector normalization without cubemap support.
Of cos ARB1 path (no specular) might be possible to fit.
Last time Carmack was asked he said he may try it if he find the drivers good enough.
Ailuros said:Doom3 on KYRO? Even if a software fallback mode is possible, I don't think I personally would have as much patience after all to play it on a K2.
Keep in mind the Kyro series is hellishly fast at stencil ops.
The problem is the lack of cube maps.
I seem to remember a .plan a while ago with Carmack saying he was floored by the performance of the Kyro series in an alpha build, but didn't like some odd problems in the drivers, so he decided to wait and see how PowerVR responded before making a PowerVR path... and that was the end of that.
Got a link to that statement, because I obviously missed it?
In any case TBDR´s do have an advantage with stencil ops, but that´s still not good enough to render a game like Doom3 (what I´d call) adequately.
If just 32 stencil operations per clock on a meger 175MHz seem to impress you, then just wait and see what the future might hold, besides all other improvements
In any case TBDR´s do have an advantage with stencil ops, but that´s still not good enough to render a game like Doom3 (what I´d call) adequately.
If just 32 stencil operations per clock on a meger 175MHz seem to impress you, then just wait and see what the future might hold, besides all other improvements
Ailuros said:Got a link to that statement, because I obviously missed it?
In any case TBDR´s do have an advantage with stencil ops, but that´s still not good enough to render a game like Doom3 (what I´d call) adequately.
If just 32 stencil operations per clock on a meger 175MHz seem to impress you, then just wait and see what the future might hold, besides all other improvements
175 * 32 = 5,600 stencil ops per second. Almost 50% higher than GeForce FX 5800. (500 * 8 = 4,000; 5600/4000 = 1.4; 40% faster)
Kyro series could slam through the stencil passes like crazy. 8)
Gee, isn't that a bit humiliating? A 175MHz two-pipeline budget core doing something faster than a Jet-Engine Cooled 500MHz top-of-the-line chip?
Chalnoth said:
Oh, I know. I said 'doing something faster', not everything.
We also don't know for sure just how much quality the NV3x path uses; if it's FX12 then there really isn't too much difference (yes, there is some, but it isn't nearly as much as even switching to FP16).
But the blazing stencil performance of the Kyro series would probably put K2 way ahead of GeForce2, possibly level with GeForce3... heck, it could conceivably approach Ti4200, depending on just how complicated the scene gets. K2 would whip through the stencil passes like it's nobody's business.
Testiculus Giganticus
Newcomer
I don`t really think that the kyro can be in the same league as gf3 in doom.Why?Well, it may do stencil ops faster BUT, it can do no pixel shaders-means it will do everything in quite a few passes, no cubemaps-workaround would, if implemented would certainly take quality down a notch and perhaps even performance, no T&L-it means the CPU will get some extra work.
As part of the binning process it automatically breaks down multiple passes into a single pass (in many respects).
Testiculus Giganticus said:
First we have exhibit A: 8)
Of course, not everything gets collapsed, but still, I'd guess thanks to this it'd need maybe... half as many passes as a regular DX7 renderer?
UT2003 uses render-to-texture (another thing Kyro ain't no slouch at) in place of cube maps, and the performance is pretty decent (~20fps at any resolution on my P3 800 - very badly CPU limited thanks to no T&L, but UT needs a lot more transform power than DOOM3 does)
DOOM3 doesn't even really use pixel shaders, all it uses them for is collapsing passes, which Kyro can do on its own anyway.
And multipassing isn't that bad anyway, considering the Kyro goes through most of the passes (all the stencil ones) several times faster than the GeForce3.
Testiculus Giganticus
Newcomer
I did not know that-my knowledge of TBRs is limited Though, i still have doubts about it being able to collapse passes by itself as often and as efficiently as a gf3 running the NV20 path that is specifically optimized for it.Stencil passes are NOT the only thing going on in a typical doom3 scene, and I don`t exactly see how the k2 is at least as fast as the gf3 at doing, say, dot products?Feel free to prove me wrong
Though, i still have doubts about it being able to collapse passes by itself as often and as efficiently as a gf3 running the NV20 path that is specifically optimized for it.Stencil passes are NOT the only thing going on in a typical doom3 scene, and I don`t exactly see how the k2 is at least as fast as the gf3 at doing, say, dot products?Feel free to prove me wrong
The NV20 path still operates at a max of 4 textures per pass - KYRO can handle 8, although there was a time when the OGL layer could only cope with 4 (not sure if this was changed); so, at worst it will be able to handle the same number of textures per pass. Even where it does need to multipass, all the geometry for every layer is binned before any per tile rendering occurs, so where multipassing does occur its done at a tile level and so the rendering element of multipassing is pretty much for free because the checks and blends are far cheaper on tile of a TBR than they are in external memory with limited bandwidth on an IMR.
Disclaimer (Before Ben or Chalnoth chimes in): Assuming geometry levels fit within the bin.
Disclaimer (Before Ben or Chalnoth chimes in): Assuming geometry levels fit within the bin.
Testiculus Giganticus
Newcomer
Don`t you mean NV20 pass?
Testiculus Giganticus said:
Uhhh, yeah! Updated.
Tagrineth said:Ailuros said:Got a link to that statement, because I obviously missed it?
In any case TBDR´s do have an advantage with stencil ops, but that´s still not good enough to render a game like Doom3 (what I´d call) adequately.
If just 32 stencil operations per clock on a meger 175MHz seem to impress you, then just wait and see what the future might hold, besides all other improvements
175 * 32 = 5,600 stencil ops per second. Almost 50% higher than GeForce FX 5800. (500 * 8 = 4,000; 5600/4000 = 1.4; 40% faster)
Kyro series could slam through the stencil passes like crazy. 8)
Gee, isn't that a bit humiliating? A 175MHz two-pipeline budget core doing something faster than a Jet-Engine Cooled 500MHz top-of-the-line chip?
I didn´t say otherwise, rather the contrary. Now increase the clockspeed according to your imagination and multiply with say 128Z/stencil operations/clock; add to that all other possible improvements.
Do you still think that a K2 is really that impressive after all in comparative terms to a high end TBDR?
psurge said:
AFAIK it´s 8.
Tagrineth,
PS: to put things into perspective, I don´t consider even a GF3 to present optimal playability in D3. Faster or more capable than X,Y,Z it might even turn out, but if I want to enjoy a game I get rather on the demanding side. Granted that Doom3 will actually manage to impress me. More than just a few games lately have been severe letdowns in gameplay.
Ailuros said:I didn´t say otherwise, rather the contrary. Now increase the clockspeed according to your imagination and multiply with say 128Z/stencil operations/clock; add to that all other possible improvements.
Do you still think that a K2 is really that impressive after all in comparative terms to a high end TBDR?
Um... I don't follow at all.
Why would it have to compare to a high end TBDR? It IS the highest end TBDR available today...
Similar threads
