ShaderMark NV40 vs R420
- Thread starter Bob3D
- Start date
Wont be for a while yet though. Still another DX 9.0c beta to come which is due 'real soon now'.
Last edited by a moderator:
Ante P said:Diplo said:
They are wrong, I've updated them in an upcoming article:
By the way the second 6800 Ultra (the one on the right) is the Extreme version.
It is interesting to note if you look at the differences in the PS2 benchmarks they almost line up perfectly with the clock disparity.
As for the PS 1.x benchmarks I have no idea wtf is going on there. Could it be a driver issue?
Mintmaster
Veteran
Does anyone know what's up with the memory controller? R3xx had much better efficiency, especially with multitexturing. I can only think of a few reasons:
- GDDR3 has a longer latency or something that wasn't entirely accounted for
- ATI had to sacrifice some FIFO sizes to keep the die size in check (which is only 1.5 times R3xx's die).
- The memory controller is tuned to actual game performance instead of the predictable patterns of theoretical tests
Can someone please run FillrateTester with the 9500 PRO? It has a bandwidth to fillrate ratio that's similar to the X800 XT, and I'd expect the X800 XT's numbers to be close to 4 times the 9500 PRO's numbers when fillrate or bandwidth bound. I've looked everywhere and can't find anyone with a 9500 PRO who has run it.
- GDDR3 has a longer latency or something that wasn't entirely accounted for
- ATI had to sacrifice some FIFO sizes to keep the die size in check (which is only 1.5 times R3xx's die).
- The memory controller is tuned to actual game performance instead of the predictable patterns of theoretical tests
Can someone please run FillrateTester with the 9500 PRO? It has a bandwidth to fillrate ratio that's similar to the X800 XT, and I'd expect the X800 XT's numbers to be close to 4 times the 9500 PRO's numbers when fillrate or bandwidth bound. I've looked everywhere and can't find anyone with a 9500 PRO who has run it.
Tridam said:Our fillrate results are a bit different :
Some results on the X800Pro seem to be too low. ATI said me that the reason is probably a lack of optimisation of the memory access pattern on the X800Pro with the current drivers.
what resolution, refreshrate and z-format are you running?
I'm running 1024x768, 60 Hz and 24 bit Z 8 bit stencil for all boards.
By the way what does the 16 bits Z "lockable" mean?
Ante P said:Tridam said:Our fillrate results are a bit different :
Some results on the X800Pro seem to be too low. ATI said me that the reason is probably a lack of optimisation of the memory access pattern on the X800Pro with the current drivers.
what resolution, refreshrate and z-format are you running?
I'm running 1024x768, 60 Hz and 24 bit Z 8 bit stencil for all boards.
By the way what does the 16 bits Z "lockable" mean?
1024x768 D24S8
I'm not sure but I think that the driver can use more than the app specified Z-buffer bits. When the format is lockable it means that the driver can't do that.
OpenGL guy
Veteran
Lockable means just that: The application is free to lock the Z buffer.Ante P said:what resolution, refreshrate and z-format are you running?Tridam said:Our fillrate results are a bit different :
Some results on the X800Pro seem to be too low. ATI said me that the reason is probably a lack of optimisation of the memory access pattern on the X800Pro with the current drivers.
I'm running 1024x768, 60 Hz and 24 bit Z 8 bit stencil for all boards.
By the way what does the 16 bits Z "lockable" mean?
Mintmaster said:
Code:
Fillrate Tester
--------------------------
Display adapter: RADEON 9500 PRO
Driver version: 6.14.10.6436
Display mode: 1024x768 A8R8G8B8 60Hz
Z-Buffer format: D24S8
--------------------------
FFP - Pure fillrate 1850.9 2047.7
FFP - Z pixel rate 1593.4 1879.6
FFP - Single texture 1215.3 1369.5
FFP - Dual texture 826.4 961.2
FFP - Triple texture 497.9 564.9
FFP - Quad texture 411.7 466.3
PS 1.1 - Simple 1073.1 1423.8
PS 1.4 - Simple 1073.1 1423.8
PS 2.0 - Simple 1073.1 1423.8
PS 2.0 PP - Simple 1073.1 1423.8
PS 2.0 - Longer 541.9 721.2
PS 2.0 PP - Longer 542.0 721.3
PS 2.0 - Longer 4 Registers 542.0 721.1
PS 2.0 PP - Longer 4 Registers 541.9 721.1
PS 2.0 - Per Pixel Lighting 136.2 181.44
PS 2.0 PP - Per Pixel Lighting 136.2 181.44Second column is with the card clocked at 366/297 (my running config).
Looking at the pure fillrate test of the 9500 PRO, sinnce at 366/297 this test is bandwidth limited for a 8 pipes 128 bits card with these clocks, the efficiency is 2047 / ( 297 * 8 ) = 86%. For the X800 XT the efficiency is 70%.
There is a problem for sure.
Sorry for my bad english of course
There is a problem for sure.
Sorry for my bad english of course
Just ran shadermark again with the 61.11 drivers and I saw no improvement whatsoever
I do get artifacting (like when overclocking) in shadermark with the 6800 Ultra Extreme... not good
all games I've tried are fine though but it did crash once on my yesterday while in windows
I think the board I got doesn't really handle the new clockspeed....
I do get artifacting (like when overclocking) in shadermark with the 6800 Ultra Extreme... not good
all games I've tried are fine though but it did crash once on my yesterday while in windows
I think the board I got doesn't really handle the new clockspeed....
Ante P said:
You could try ati tool and see whether that detects artifacts. The program wont be able to change your cards clocks seeming as it only supports AIT cards for overclocking, but it should be able to scan for artifacts at the speed you have the card at using its ps2.0 rotating cube.
