NV30 - last chance to speculate - reactorcritcal specs
- Thread starter g__day
- Start date
Sharkfood said:
With AA turned on is the real claim. Which means a 3DMark (yes, the current version) score of around, what, 15k with AA enabled? Not sure if that's at the default res. of 10x7 or 16x12. . .gotta' be careful with the little details on these marketing claims.
I agree it can be misleading to make claims about "effective" bandwidth, at least if there are no performance measurements to support those claims. But let's say, just for argument's sake, that NV30 spanks R300 on bandwidth intensive tasks, even though it has less *physical* bandwidth. If that happens, it seems that either:
- It is fair to make claims about effective bandwidth to the extent that those claims are supported by actual performance.
OR
- Bandwidth has become an irrelevant spec because it no longer is an indicator of performance. I mean, who cares that one card has more physical bandwidth than another, if that bandwidth does not lead to higher performance?
- It is fair to make claims about effective bandwidth to the extent that those claims are supported by actual performance.
OR
- Bandwidth has become an irrelevant spec because it no longer is an indicator of performance. I mean, who cares that one card has more physical bandwidth than another, if that bandwidth does not lead to higher performance?
Randell said:
Hey, you're assuming I read your posts. 8)
pocketmoon_ said:
That's something I've wanted for some time now. I suggested something similar about 1.5 years ago on opengl.org by using "perlin noise textures", but I think better solution would be to add a new fragment/vertex shader instruction, like
PRLN reg0.x, reg1.xyz;
Would be really awesome.
Coincidentally, I'm working on a inifinite terrain demo right now using perlin noise to build the terrain. If we had a perlin noise instruction in the vertex shader all my tricks I'm doing would be unneccesary.
How do you know how that bandwith was estimated?antlers4 said:9700 has 19.2 Gb/sec physical bandwidth, z-buffer compresses up to 24x1 and has early-Z rejection. Does that mean it has 4608 Gb effective bandwidth? Nearly 10x nVidia's figure.
We don't even know if that 48 Gb/s is a real figure from nvidia...
ciao,
Marco
Randell said:
Easy. . .I was joking.
Hellbinder
Banned
You guys are all assuming that Nv30's FSAA is actually going to be better than the 9700. As it is Ati Only takes about a 30% hit on performance with 4x FSAA and 8x Aniso in most games.
You are all buying into that 48gb effective stuff. Need i remind you that the Nv30 really has 16GB raw and only a 128bit bus? The 9700 has more raw bandwidth uses MSAA, has 24:1 frame buffer compression (peak) and uses hyper Z III doing FSAA.
The only thing that the nv30 has going for it as far as i can see is Raw clock speed. But it still is not a miracle cure for such low real bandwidth where FSAA is considered. Then you have to consider the 2 TMU's are potentially be eating a lot of that bandwidth up.
i just dont see any basis at all for these extreme figures for FSAA performance. 18,000 in 3dmark with FSAA+Ansio. its just not going to happen if the current info holds up. Ati has very advanced bandwidth saving features and even at 400mhz it aint going to get no 18,000.
You guys sure do assume a lot. I just want to know based on what?
You are all buying into that 48gb effective stuff. Need i remind you that the Nv30 really has 16GB raw and only a 128bit bus? The 9700 has more raw bandwidth uses MSAA, has 24:1 frame buffer compression (peak) and uses hyper Z III doing FSAA.
The only thing that the nv30 has going for it as far as i can see is Raw clock speed. But it still is not a miracle cure for such low real bandwidth where FSAA is considered. Then you have to consider the 2 TMU's are potentially be eating a lot of that bandwidth up.
i just dont see any basis at all for these extreme figures for FSAA performance. 18,000 in 3dmark with FSAA+Ansio. its just not going to happen if the current info holds up. Ati has very advanced bandwidth saving features and even at 400mhz it aint going to get no 18,000.
You guys sure do assume a lot. I just want to know based on what?
Johnathan256
Newcomer
You also assume alot.
Assuming NV30 can't outperform R300 is no better than assuming NV30 will win out. It just depends on your point of view. The people in here assume only because they don't yet have facts. As do you. Well, as has been mentioned, performance at high resolutions with AA yield different results than the numbers people are used to. In the high end pc round up in the December issue of PC Magazine, the 9700 systems scored about 2.5 x the GF4 Ti 4600 scores on the P4 2.8 at 1600x1200 with 4x antialiasing. There were overclocks and possible configuration differences that make the results not quite useful, but it demonstrates that such a comparison is in the ballpark. With anisotropic filtering used, the margin would only get larger I think.
Certainly makes the nv30 claims easily possible on some system configuration somewhere, and, again as has been said, not necessitating "18,000" with FSAA and aniso. It is even meaningful as you'd buy such a card to turn up the resolution and image quality features in any case.
For reference, the figures for the systems in that round up at 1600x1200 4x AA were around 2,200 to 2,650 with the Ti 4600 systems, and 6,150 to 6,500 with the 9700 systems.
Certainly makes the nv30 claims easily possible on some system configuration somewhere, and, again as has been said, not necessitating "18,000" with FSAA and aniso. It is even meaningful as you'd buy such a card to turn up the resolution and image quality features in any case.
For reference, the figures for the systems in that round up at 1600x1200 4x AA were around 2,200 to 2,650 with the Ti 4600 systems, and 6,150 to 6,500 with the 9700 systems.
Mintmaster
Veteran
demalion:
I was just about to say the same thing. None of these numbers show a significant improvement over the 9700:
-The 9700 already is 2.5 times the speed of GF4 in Q3 (16x12, 4xFSAA)
-The 9700 already gets 3 times the GF4's 3DMarks (16x12, 4xFSAA)
-We don't know about Doom3, but it was already significantly faster than GF4 in alpha state, and it was mentioned by ID that ATI really got the performance up since then
-The 9700 already does 128-tap aniso. 16x trilinear aniso = 16 x 8 = 128x.
-I haven't seen a screenshot where 6x FSAA has any bad edges (some sites said it was even better than Matrox's 16x FSAA due to the latter's ordered grid), so I doubt 8x will be much improvement
-NVidia's shaders, from what I gather, are not that much better than ATI's. The way I see it, 8500 is to GF3/4 as NV30 is to R9700 in terms of shader functionality. They may perform better, however, due to the 8x2 config and extra math power. Without textures in the vertex pipe, NV30 won't be able to support VS 3.0.
I was actually very surprised to see this information come up again, as I thought the previous information from RC was bogus. Either this info has been leaked again (making it more likely to be true), or it's just the old rumour in new form.
I'm finding NV30 to be less impressive by the day. I don't think R400 is that far away now, either - I'm thinking around 6 months.
I was just about to say the same thing. None of these numbers show a significant improvement over the 9700:
-The 9700 already is 2.5 times the speed of GF4 in Q3 (16x12, 4xFSAA)
-The 9700 already gets 3 times the GF4's 3DMarks (16x12, 4xFSAA)
-We don't know about Doom3, but it was already significantly faster than GF4 in alpha state, and it was mentioned by ID that ATI really got the performance up since then
-The 9700 already does 128-tap aniso. 16x trilinear aniso = 16 x 8 = 128x.
-I haven't seen a screenshot where 6x FSAA has any bad edges (some sites said it was even better than Matrox's 16x FSAA due to the latter's ordered grid), so I doubt 8x will be much improvement
-NVidia's shaders, from what I gather, are not that much better than ATI's. The way I see it, 8500 is to GF3/4 as NV30 is to R9700 in terms of shader functionality. They may perform better, however, due to the 8x2 config and extra math power. Without textures in the vertex pipe, NV30 won't be able to support VS 3.0.
I was actually very surprised to see this information come up again, as I thought the previous information from RC was bogus. Either this info has been leaked again (making it more likely to be true), or it's just the old rumour in new form.
I'm finding NV30 to be less impressive by the day. I don't think R400 is that far away now, either - I'm thinking around 6 months.
Similar threads
