gandhig wrote:

Mine is a Pericom chip. Assuming that the default win7 driver(pci.sys) is sufficient for your PLX chip(couldn't find much in their website, atleast the pericom website had FAQ section), then only the bridge vendor can do something about the code 10 error. Chances are however remote to get any meaningful support from them. But I'm sure that if that is resolved, your graphics card will get detected and drivers will get installed.

the driver should support, but unfortunately it always gives me error 10, and from what I was able to find via google it's a common problem with these 8400GS's and there is no known solution and people just give up.

SPBHM wrote:

I didn't test with 3dMark05 but with 3dMark06 and yeah the CPU tests were terrible compared to a slide show even, something around 0.09 FPS. Even a Core2Duo E7500 scored only 0.83
For some reason, the cuda tool didn't run on my system. I found similar tools and their results on my system are in the previous pages of this thread. I guess we shouldn't read much into these synthetic benchmarks.

If I remember correctly the 3dm06 CPU test is compeltely different, while the 03 and 05 is similar, but as I said, with a PCIE8400GS and the same CPU it should be getting 10x higher score on 3dm03 CPU test, while the GPU part of it seems acceptable, so there is something going on

I installed HalfLife1 and the lowest i got was 21 FPS standing still(snapshot attached) inside the office complex till now. The lowest i got in the intro was 32 FPS. If your Sempron 1.8GHz platform gets 48 FPS, then it seems that my P3 850MHz platform getting 32 is normal. It's only that the overall gameplay is somewhat jerky though it doesn't look like stuttering. If only someone can confirm that a P3 850 MHz coppermine @100 FSB with PC133 3-3-3-6 timings will not go below 30 FPS at any point in HalfLife1(other than during loading, obviously) with a decent native PCI/AGP graphics card (Voodoo or Geforce FX5500/6200) on a VIA Chipset motherboard!!!

well, the only other slower PC I have at the moment running this game is my Pentium II 400 with the voodoo 4, at the same place the 8400GS + sempron was getting the 48 the PII was at 34 with 3dfx mini gl driver and around 24 with the default OpenGL... but it's also the CD version, win9x... not really comparable I guess,

duh, missed the attachment again.

@SPBHM, I will try both 3dm03 & 05. If I find any win7 system, I will try my gpu. Thanks for sharing your results.

gandhig wrote:

duh, missed the attachment again.

768p.jpg

@SPBHM, I will try both 3dm03 & 05. If I find any win7 system, I will try my gpu. Thanks for sharing your results.

I don't have a game save as far as that at the moment but,
I've just realized I can underclock my CPU to emulate something closer to the P3

so I underclocked it to 800MHz (minimum I can get, and should still be faster than a 1GHz p3) with the memory at 200 (previously it was 1800 and 450)

checked the demo1 again at 640x480 with the same VGA OC and...

8400GS PCI 18% core/memory OC 640x480 k8 800MHz/DDR200
260.978 seconds 86.689 fps

big drop compared to the same settings at 1.8GHz:

8400GS PCI 18% core/memory OC 640x480 k8 1.8GHz/DDR450
121.252 seconds 186.586 fps

and the place where I took the screenshot at 51FPS and 58FPS with the VGA OC now went down to around 34 at 640x480 and a little lower at 1024x768 (but the difference was smaller compared to the runs with the CPU at 1.8GHz);

at least with these PCI to PCIE geforces it looks like you need a really fast CPU to avoid some framerate drops like this, it would be interesting to know how it performs with a similar CPU and AGP card.

@SPBHM, I really appreciate the efforts you are putting to help me identify the cause and the downclocking was damn good thinking. I guess then your conclusion is right about the GPU's with PCI bridge(still don't understand why) & low range CPU. It would be nice to compare it with a native PCI/AGP for reconfirmation.

This is what i got with the demo1 timedemo at 1024x768, OpenGL Default, HD models disabled
304.149 seconds 74.385 fps

I have attached the SAV file which will load the exact position where i got the lowest FPS.

SPBHM wrote:

If I remember correctly the 3dm06 CPU test is compeltely different, while the 03 and 05 is similar, but as I said, with a PCIE8400GS and the same CPU it should be getting 10x higher score on 3dm03 CPU test, while the GPU part of it seems acceptable, so there is something going on

I noticed that about the 3D'03 CPU test as well. It just seems to really hate PCI video cards. On my PIII platform (1400-S, overclocked to 1586MHz, 151MHz FSB, 1.5GB 2-2-2 SDRAM, VIA 694X, WinXP Pro SP3):
-With Radeon 9800Pro AGP, CPU score is 330.
-With GeForce FX5200 AGP, CPU score is 347.
-With Radeon 9250 PCI (128-bit), CPU score is 144.

Same thing happens on a P4 2.4GHz/i845/1GB CL2 DDR266 platform:
-With Radeon 9800Pro AGP: 435
-With Radeon 9250 PCI: 152

It has to be a bandwidth limitation

gandhig wrote:

@SPBHM, I really appreciate the efforts you are putting to help me identify the cause and the downclocking was damn good thinkin […]
Show full quote

@SPBHM, I really appreciate the efforts you are putting to help me identify the cause and the downclocking was damn good thinking. I guess then your conclusion is right about the GPU's with PCI bridge(still don't understand why) & low range CPU. It would be nice to compare it with a native PCI/AGP for reconfirmation.

This is what i got with the demo1 timedemo at 1024x768, OpenGL Default, HD models disabled
304.149 seconds 74.385 fps

I have attached the SAV file which will load the exact position where i got the lowest FPS.

Half-Life-009.zip

now that was really interesting, let me start with the demo1 at the same settings (1024 hd models off)
264.678 seconds 85.478 fps (card was running with the usual 18% OC, CPU at 800MHz)

now I loaded you save game and, without moving the mouse the framerate was 27-28
now I tried it again at 1800MHz and it was 60-61, BUT, I decided to turn the flashlight and run around a little, it went from 60 to 40 and under 30, managed to see 22fps running back to that place with it on, turned off the flashlight and we are back to 60.

now I did the same with 800MHz, again start at 27-28, run around and it seems to stay consistently at low 20s, managed to see 18FPS.

Standard Def Steve wrote:

[…]
Show full quote

SPBHM wrote:

If I remember correctly the 3dm06 CPU test is compeltely different, while the 03 and 05 is similar, but as I said, with a PCIE8400GS and the same CPU it should be getting 10x higher score on 3dm03 CPU test, while the GPU part of it seems acceptable, so there is something going on

I noticed that about the 3D'03 CPU test as well. It just seems to really hate PCI video cards. On my PIII platform (1400-S, overclocked to 1586MHz, 151MHz FSB, 1.5GB 2-2-2 SDRAM, VIA 694X, WinXP Pro SP3):
-With Radeon 9800Pro AGP, CPU score is 330.
-With GeForce FX5200 AGP, CPU score is 347.
-With Radeon 9250 PCI (128-bit), CPU score is 144.

Same thing happens on a P4 2.4GHz/i845/1GB CL2 DDR266 platform:
-With Radeon 9800Pro AGP: 435
-With Radeon 9250 PCI: 152

It has to be a bandwidth limitation

good to see a confirmation, in my case it's even worse, the score on 03 was... 77! for a 1.8GHz k8... no way...

@SPBHM, your results definitely indicating the same issue. Here are my 3dmark results:

3DMark03:
Score=4295; GT1=52.2; GT2=50.09; GT3=28.37; GT4=18.73; CPU Marks=90; CPU Test1=10.80; CPU Test2=1.49; Single/Multi Texturing=1078/4434
Vertex/Pixel2.0 Shader=16.60/54.24; RagTroll=12.16

3DMark05:
Score=1865; GT1=8.59; GT2=5.45; GT3=8.88; GT4=18.73; CPU Marks=441; CPU Test1=0.26; CPU Test2=0.34;

3DMark06:
Score=921; GT1=2.76; GT2=3.43; HDR1=4.45; HDR2=3.72; CPU Marks=284; CPU Test1=0.09; CPU Test2=0.14;

3DMark06: (Core2Duo 3.0GHz with GT520 PCI)
Score=1758; GT1=4.70; GT2=5.55; HDR1=7.07; HDR2=5.96; CPU Marks=2628; CPU Test1=0.84; CPU Test2=1.32;

1280x1024 seems to be the sweet spot for my system setup, maybe even the result of PCI bandwidth limitation. My avg FPS from Demo1 holds at 74. 1080p drops it to 65 FPS. 8xAA at 768p dropped it to 45 FPS.

@StandardDefSteve, if you have this Tualatin setup running, can you load the attached half life save file and check the FPS(if you have steam version still better) with 9250 PCI and 9800 AGP. Though 9250 by itself is not comparable with 8400GS & GT520, if it delivers more FPS at the min FPS location, then we can conclude that the culprit is the PCI reverse bridge. On the other hand 9800 AGP Pro which is more or less comparable to these cards(more with GT520) can give an idea about the difference in FPS on account of AGP and PCI buses.

I don't have the original Half-Life. Any other benchmark or game?

Can you download the Half Life demo from here. I'm attaching the save file which will load the min FPS position for this version.

You have to copy the save file to the SAVE folder inside Profiles folder. Console commands are disabled and hence FRAPS is the way to go. I didn't find much difference in the FPS values between that displayed by FRAPS and ingame, so it should be comparable.

Another interesting thing was that even my core i5 laptop had only 33 FPS at that position. However that was with OpenGL, no problem with software renderer(60 FPS - restricted by intel drivers). I think we are getting somewhere.

Other Games:
Deus Ex 1 (Opening dock scene - 12 FPS)
Hitman 1 (Traditions of the trade opening scene outside the hotel - FPS in teens or low 20's)
F1'99 (Australian Grandprix start-finish line)
Unreal 1 Flyby demo (right side of the castle - 22 FPS)

Edit: UT2003 demo benchmark: Flyby/Botmatch=64.73/23.43, overall there was a lot of stop/still moments during the benchmark rendering.

Another thing I noticed, the current steam version of HL1 here is always running with some sort of anti aliasing!? also it looks like the game scales to the desktop resolution (check my 640x480 screenshot), 8400gs vs voodoo 4 attached, I didn't specify the use of AA, the nvidia control panel is using the default settings (controlled by application).
(both running at 1024x768)

I also checked this demo version on the previous post, it's quite old and it looks like it runs worse than the steam version on the 8400GS, it had some visual glitches here.

gandhig wrote:

3DMark03: Score=4295; GT1=52.2; GT2=50.09; GT3=28.37; GT4=18.73; CPU Marks=90; CPU Test1=10.80; CPU Test2=1.49; Single/Multi Tex […]
Show full quote

3DMark03:
Score=4295; GT1=52.2; GT2=50.09; GT3=28.37; GT4=18.73; CPU Marks=90; CPU Test1=10.80; CPU Test2=1.49; Single/Multi Texturing=1078/4434
Vertex/Pixel2.0 Shader=16.60/54.24; RagTroll=12.16

1280x1024 seems to be the sweet spot for my system setup, maybe even the result of PCI bandwidth limitation. My avg FPS from Demo1 holds at 74. 1080p drops it to 65 FPS. 8xAA at 768p dropped it to 45 FPS.

1080p drops my result to 37 AVG on demo1.

my 3dmark03 results
CPU at 1.8GHz, VGA with 18% OC

3DMark03:
Score=4984; GT1=122.7; GT2=46.7; GT3=31.1; GT4=23.1; CPU Marks=78; CPU Test1=10.3; CPU Test2=1.1; Single/Multi Texturing=931/2951
Vertex/Pixel2.0 Shader=15.1/61.7; RagTroll=23.4

just found my old result from with a XP 1.4@2.2GHz (shouldn't be all that different from k8 at 1.😎 and FX5900 and it was over 600 for the CPU, and 6K overall, it was faster on GT1 and GT4, and CPU results are not even comparable, rest is around the same.

and 800MHz CPU, VGA with 18% OC:
Score=4095; GT1=83.0; GT2=42.1; GT3=26.0; GT4=18.3; CPU Marks=71; CPU Test1=9.4; CPU Test2=1.0; Single/Multi Texturing=931/2949
Vertex/Pixel2.0 Shader=12.8/56.0; RagTroll=16.8

SPBHM wrote:

Another thing I noticed, the current steam version of HL1 here is always running with some sort of anti aliasing!? also it looks like the game scales to the desktop resolution (check my 640x480 screenshot), 8400gs vs voodoo 4 attached, I didn't specify the use of AA, the nvidia control panel is using the default settings (controlled by application).(both running at 1024x768)

It looks like that and I will check in detail tomorrow. Are both game versions (8400GS & Voodoo4) same? Books on the shelf are missing in the Voodoo one. I can't believe nvidia graphics cards taking a hit with 8xAA for old titles like HL1. Got to understand more and do some more trial runs with different settings in the nvidia control panel.

SPBHM wrote:

I also checked this demo version on the previous post, it's quite old and it looks like it runs worse than the steam version on the 8400GS, it had some visual glitches here.

Yeah it was the alpha version(pre-release). I too had the visual glitches but FPS was same, if not better, as the steam version. I don't know how much steam running in the background affects the FPS, thought it should not be affecting much. I gave the link only for Standard Def Steve to get the benchmark going so that we can have a common platform. This thread will then atleast help people to not fall for new PCI graphics cards for gaming with old hardware, if proved correct.

SPBHM wrote:

my 3dmark03 results CPU at 1.8GHz, VGA with 18% OC 3DMark03: Score=4984; GT1=122.7; GT2=46.7; GT3=31.1; GT4=23.1; CPU Marks=78; […]
Show full quote

my 3dmark03 results
CPU at 1.8GHz, VGA with 18% OC
3DMark03:
Score=4984; GT1=122.7; GT2=46.7; GT3=31.1; GT4=23.1; CPU Marks=78; CPU Test1=10.3; CPU Test2=1.1; Single/Multi Texturing=931/2951
Vertex/Pixel2.0 Shader=15.1/61.7; RagTroll=23.4

That's baffling to say the least, your CPU score, lesser than even mine. Maybe 3Dmark03 is partial to yours (on the dark side), I was about to say AMD processors. But then your Athlon XP didn't have any problem. Either it is a glitch or the bench is partial to Sempron (maybe low cache or something).

gandhig wrote:

It looks like that and I will check in detail tomorrow. Are both game versions (8400GS & Voodoo4) same? Books on the shelf are missing in the Voodoo one. I can't believe nvidia graphics cards taking a hit with 8xAA for old titles like HL1. Got to understand more and do some more trial runs with different settings in the nvidia control panel.

it's not the same version, the 8400GS PCI is running trough steam, the PII is running a CD-ROM version I got in 2002, I think it was released in 2001 "best seller series" and came with the updates released at that time, I can't say the exact version right now, as for the differences on the shelf, I noticed but than I realized it's destructible, I would need to check again.

That's baffling to say the least, your CPU score, lesser than even mine. Maybe 3Dmark03 is partial to yours (on the dark side), I was about to say AMD processors. But then your Athlon XP didn't have any problem. Either it is a glitch or the bench is partial to Sempron (maybe low cache or something).

I would think it's unrelated to the CPU, 1800 to 800MHz should have a much bigger impact if was a representative CPU score, I think it's just caused by the VGA (and PCI performance),

this K8 I'm using is a Sempron "Palermo" it's one of the last CPUs released for s754, it's basically A64 Venice with half (256kb) of the l2 enabled... it shouldn't be much slower, the AXP I used back in the day also had 256kb l2 and the rest of the CPU/platform was significantly improved from k7 to k8 anyway.

SPBHM wrote:

I would think it's unrelated to the CPU, 1800 to 800MHz should have a much bigger impact if was a representative CPU score, I think it's just caused by the VGA (and PCI performance),this K8 I'm using is a Sempron "Palermo" it's one of the last CPUs released for s754, it's basically A64 Venice with half (256kb) of the l2 enabled... it shouldn't be much slower, the AXP I used back in the day also had 256kb l2 and the rest of the CPU/platform was significantly improved from k7 to k8 anyway.

The term 'CPU score' is misleading. If what you deduced is correct(VGA & PCI performance), then I don't get how Standard Def Steve's Tualatin 1.5GHz sytem(lesser than your Sempron 1.8GHz) with 9250 PCI graphics card(again lesser than your 8400GS PCI) score almost double the points? Unless the PCI reverse bridge has something to do with it.

SPBHM wrote:

Another thing I noticed, the current steam version of HL1 here is always running with some sort of anti aliasing!? also it looks like the game scales to the desktop resolution (check my 640x480 screenshot), 8400gs vs voodoo 4 attached, I didn't specify the use of AA, the nvidia control panel is using the default settings (controlled by application).

You were right on both counts. Steam version definitely has default anti aliasing, even forcing it off from nvidia control panel did nothing. Coming to scaling, only the OpenGL resolutions upto 1024x768 under normal display mode automatically scales to 1280x960(not to desktop resolution). I was able to verify it by disabling the scaler setting in the control panel. Software renderer does not have this aspect.

I'm having trouble getting it to run. It just crashes to the desktop when I try to load a new game. I've tried Win98 compatibilty mode. This happens with the software renderer and OpenGL. Running Catalyst 7.3 (9800 Pro).

Standard Def Steve wrote:

I'm having trouble getting it to run. It just crashes to the desktop when I try to load a new game. I've tried Win98 compatibilty mode. This happens with the software renderer and OpenGL. Running Catalyst 7.3 (9800 Pro).

Now that is unfortunate. I don't have much idea about ATI cards as I never owned one, or for that matter the GT520 is the only one i bought(poor me). Is it crashing even when run in a window? Did you get the same issue with 9250 PCI too? Thanks.

Hmm, doesn't work in windowed mode either. It also doesn't work with the 9250 PCI, using older Cat 5.6 drivers.

Standard Def Steve wrote:

Hmm, doesn't work in windowed mode either. It also doesn't work with the 9250 PCI, using older Cat 5.6 drivers.

If you have Unreal 1 or Gold preferably with latest patch, can you run the flyby timedemo in Direct3D9 mode? I'm still searching for other benchmarks from that period so that CPU/Memory bottleneck will not be a factor.

edit1: Can you try half life1 demo once again with default settings(without overclocking), just in case if ATI card is not the reason for crash?

edit2: Expendable built-in timedemo, 640x480x32, min/avg/max FPS=22/36/69

edit3: Turok=90.3(Nvidia Riva 128 default); Incoming=35.2(default); Forsaken=101.25(default); Both Incoming & Forsaken rendering totally garbled.

Half Life doesn't work at default settings, either.
I'm in the process of downloading Expendable from an extremely slow site (5KB/s 😲 ). Will bench it later on.

I ran a few other benchmarks on this system as well. All benches were run on a TUV4X at 1585MHz (151x10.5), 1.5GB of memory at 151MHz 2-2-2, and WinXP SP3 with the latest updates. I used Catalyst 7.3 with the 9800 Pro and 5.6 with the 9250 PCI.

Radeon 9250 PCI running Final Reality:

The Radeon 9800 Pro running Aquamark 3:

The Radeon 9250 PCI running Aquamark 3. As with 3DMark, the CPU score is much lower when the PCI GPU is used.

Radeon 9800 Pro running 3DMark03:

Radeon 9250PCI running 3DMark03:

Radeon 9800 Pro running 3DMark05 (the 9250 wouldn't run this one)

Thanks again Standard Def Steve for bearing with the pain and the misfortune of downloading at less than dial up speeds, stretching your patience to the limit.

Final Reality Comparison:

3dMark03 Comparison:

Probably the GPU benchmark developers should specify which benchmark, tests, 'what' portion of the graphics subsystem and that will be more meaningful for identifying bottlenecks. @Standard Def Steve, could you please consider running the benchmarks with downclocking to 850 MHz, if possible?

I owe you guys a lot for the help and in general grateful to all the members of our forum for not making me feel stupid about this thread and also for refraining from harsh remarks. I look forward to the day I might be useful to this forum, as presently I don't have much to offer both in hardware & software.