gandhig wrote:
2d bus transfer rate is abnormally low.
sorry about the long wait, I haven't really read the newer posts so I don't know if this is still relevant, but since I did some testing...
let me share a few results I have here;
PC used was a Sempron 1.8GHz (Venice, K8 with 256K l2)
chipset was the SiS 761GX,
Windows XP SP3
Final Reality
Sis Mirage 761GX IGP - 8400GS 512Mb PCI- Voodoo 4 4500 PCI (Windows drivers) - Voodoo 4 4500 PCI (Amigamerlin r11)
2D transfer rate test (MB/s) - 241.28 - 27.04 - 78.47 - 82.38
3d transfer rate test (MB/s)- 254.6 - 30.81 - 68.74 - 72.25
as you can see the IGP have a huge advantage (now to be honest, I don't know how relevant or reliable this 1997 benchmark is running with newer hardware and windows XP), but even the Voodoo 4, using the same PCI slot is much faster, I think there is a lot to do with software optimization, the same voodoo 4 4500 on my SiS 530 board gets around 20-30MB/s, while using a 440BX board it gives me close to 100MB/s, different windows and drivers version also impact this significantly; the 8400GS originally was PCIE 16x card, that's a very high bandwidth, while PCI 33MHz is so limited... so I think Nvidia don't really care about the PCI version, it's not even made by them, but by their partners (bridge on the same PCB to PCIE 1x), also if I remember correctly, Radeons work a lot better with low bandwidth, while using special optimizations like Optimus might change this (I'm basing this on what I remember about the guys testing external GPUs on laptops using PCIE x1 1.0)
rest of results on Final Reality:
Sis Mirage 761GX IGP - 8400GS 512Mb PCI - Voodoo 4 4500 PCI (Windows drivers) - Voodoo 4 4500 PCI (Amigamerlin r11)
3d tests
25pixel (Kpoly/s) - 484.04 - 361.21 - 564.28 - 103.76
Robots (image/s) - 113.65 - 110.50 - 90.97 - 177.94
fill rate (Mpixels/s)- 21.11 - 21.45 - 21.07 - 129.40
City scene (images/s) - 136.94 - 142.26 - 152.66 - 268.52
2d image processing
Radial Blur (images/s) - 245.22 - 247.9 - 248.79 - 251.95
Chaos zoomer (images/s) - 451.86 - 484.84 - 468.92 - 469.15
now let's try 3dmark2001SE
1024x768 - Software T&L (apart from the specified run) 32bit - 32bit/24bit texture compressed (only the Mirage ran with 32bit, the others with 24)
SiS Mirage 761GX IGP - 8400GS PCI - 8400GS PCI Pure Hardware T&L - V4 4500 PCI (Amigamerlin r11)
GT1low - 25.5 - 13.0 - 115.4 - 24.4
GT1High - 13.8 - 7.7 - 37.2 - 9.0
GT2low - 23.0 - 32.0 - 84.8 - 32.2
GT2High - 12.3 - 8.8 - 56.9 - 10.8
GT3Low - 22.8 - 59.0 - 152.0 - 33.5
GT3High - 13.8 - 20.0 - 83.7 - 17.4
GT4 - xx - 13.5 - 35.1 - xx
FillrateSingle - 111.2 - 961.3 - 963.4 - 151.1
Fillratemulti - 195.7 - 2641.1 - 2640.0 - 289.2
HighPoly1l - 4.2 - 1.0 - 10.4 - 3.0
hipoly8l - 3.1 - 1.0 - 10.2 - 1.1
EnvBumpMapping - xx - 49.5 - 58.5 - xx
DOT3BumpMapping - 13.7 - 25.0 - 21.8 -xx
Vertex Shader - 21.0 - 2.8 - 78.4 - 11.4
Pixel Shader - xx - 31.4 - 33.9 - xx
AdvPixelShader - xx - 140.0 - 173.9 - xx
Point Sprites - 1.5 - 0.8 - 37.0 - 0.5
NOW
640x480 - Software T&L* (same thing) 16bit - compressed 16bit
SiS Mirage 761GX IGP - 8400GS PCI - 8400GS PCI Pure Hardware T&L - V4 4500 PCI (Amigamerlin r11)
GT1low - 58.8 - 13.3 - 132.5 - 34.7
GT1High - 23.8 - 7.9 - 40.1 - 14.9
GT2low - 53.4 - 38.3 - 95.1 - 43.4
GT2High - 21.3 - 8.8 - 60.8 - 12.2
GT3Low - 74.0 - 61.9 185.9 - 83.2
GT3High - 37.6 - 21.2 - 95.3 - 37.6
GT4 - xx - 14.7 - 49.8 - xx
FillrateSingle - 211.7 - 1031.7 - 1031.7 - 267.0
Fillratemulti - 351.9 - 2483.1 - 2483.4 - 327.8
HighPoly1l - 4.8 - 1.0 - 10.4 - 3.0
hipoly8l - 3.3 - 1.0 - 10.2 - 1.1
EnvBumpMapping - xx - 49.5 - 58.8 - xx
DOT3BumpMapping - 39.1 - 25.0 - 21.7 - xx
Vertex Shader - 40.5 - 2.8 - 81.0 - 11.7
Pixel Shader - xx - 32.2 - 34.1 - xx
AdvPixelShader - xx - 159.6 - 271.6 - xx
Point Sprites - 1.9 - 0.8 - 57.1 - 0.5
I would really like to have a 8400GS PCIE to see if it looks the same, I mean so extremely slow with software T&L, or if it's related to the low bandwidth from the PCI bus,
the Geforce tests were all made using Forceware 178, the latest version 335 I think can run 3dmark and the results are around the same, but it can't run final reality, the SIS IGP is using the latest driver from SiS website.
I also managed to run the OpenCL test from Aida 4.2 (I think) and I'm pretty sure the memory read/write speed is limited by the PCI bus more than anything
Memory Read 99.88MB/s
Memory Write 61.46MB/s
while my PCIE 16x 2.0 card gives me around 6GB/s
again, it would be great to have a 8400GS PCIE to compare.
I honestly don't think there is much hope for these newer PCI cards to perform well compared to their PCIE x16 versions, but in some conditions it can compared to slower GPUs with higher bandwidth (like the 3dmark test using hardware T&L shows), or DXVA with MPCHC or other similar software for HD videos;
for gaming or general usage... better use an older native PCI solution or specially AGP
what really kills this card for me is that even on "youtube" once you go over a certain resolution in full screen it's just to slow, even the IGP with no hardware acceleration can do a better job.
edit: I would like to add that I rarely tried any gaming with this card, but Doom 3 was running quite ok, considering the slow PCI and the fact that my 8400GS is the G98 model with only 8sps, 800x600 high (no AA) gave me 46FPS on the timedemo, if my memory is right this comparable or faster than what a high end AGP8x (pre Geforce 6800/Radeon X800) card could achieve in this game,