since he is running the P3 at FSB 151 I would expect the PCI clock to also be overclocked, around 37MHz? should help a little,
also the 9250 is an AGP GPU I guess, so no need for bridge chips and so on...

my 3d03 CPU result is probably limited by the PCI bus, if you consider a 8400GS would not cause that because of GPU performance and the CPU even going from 1800 to 800MHz hardly affected it, now why is it lower than yours? not sure, maybe something with my chipset or the bridge chip being worse for that...

as for the rest, it goes from CPU limited, to GPU limited and bandwidth limited, it's not very easy to understand,

@SPBHM, found this datasheet for the PLX chip.

If you want to tweak the bridge registers(especially 0C, 0D, 48, 69, 100F) using wcpredit, you can take a look at chapter 16. Unfortunately the last register 100F can only be tweaked with some other utility. However my experience was that by tweaking the Pericom chip, I didn't see any drastic improvement.

Aquamark 3 Results


Looks like my system doesn't like full scale vegetarian.

Expendable timedemo1

Radeon 9800 Pro:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 84 Seconds
Gameframes - 11226
Lowest FPS - 95 fps
Highest FPS - 193 fps
Average FPS - 132.940476 fps

Radeon 9250PCI:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 97 Seconds
Gameframes - 11226
Lowest FPS - 81 fps
Highest FPS - 160 fps
Average FPS - 114.948454 fps

Thanks Standard Def Steve.

@SPBHM, would you be able to run the expendable timedemo with your default settings as well as underclocking to 800 MHz to have a proper comparison.

I'm trying to single out the issue by sorting and analysing all the bench results(to the extent possible) you guys have provided me.

Just for fun(to keep myself in good spirits by seeing high FPS in my system), I got crazy and did a FRAPS run on Direct3D tests in the DxDiag 🤣 . This is what I got,

Direct3D 7 : 100 to 130 FPS
Direct3D 8 : 630 to 710 FPS
Direct3D 9 : 580 to 620 FPS
Why so much difference between Direct3D 7 & 8/9??

Edit: I guess, I should open a thread to see other systems' results from these tests with probably 5 or even 6 digit FPS scores.

gandhig wrote:

@SPBHM, would you be able to run the expendable timedemo with your default settings as well as underclocking to 800 MHz to have a proper comparison.

Interesting result here, I've used the Demo version

Just to make sure, my settings were default, only changed from 16 to 32bit so:
Launcher 640x480 32 bit colour

Only "Use Low Resolution Movies" selected
in game settings attached

results:

K8 "Sempron Palermo" @ 1800MHz, ram at 450 CL3

Geforce 8400GS(G98) PCI stock:

Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 133 Seconds
Gameframes - 11226
Lowest FPS - 52 fps
Highest FPS - 140 fps
Average FPS - 84.218045 fps

Geforce 8400GS(G98) PCI 18% OC:

Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 132 Seconds
Gameframes - 11226
Lowest FPS - 52 fps
Highest FPS - 138 fps
Average FPS - 84.477273 fps

K8 "Sempron Palermo" @ 800MHz, ram at 200 CL3

Geforce 8400GS(G98) PCI stock:

Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 230 Seconds
Gameframes - 11226
Lowest FPS - 31 fps
Highest FPS - 80 fps
Average FPS - 48.665217 fps

so results are bellow a 9250 PCI and 1.5GHz Tualatin, but the test scales well with CPU clock, but it's not affected by a 18% OC on the GPU.

modern geforces with PCI to PCIE bridge are more CPU dependent compared to older "native" PCI cards?
or just modern PCI geforces are "slow" for this game?
or no love for k8?

I'm going to test the voodoo and later edit this, if it works.

edit: it works, and rather well:

K8 "Sempron Palermo" @ 1800MHz, ram at 450 CL3

Voodoo 4 4500 PCI stock:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 114 Seconds
Gameframes - 11226
Lowest FPS - 60 fps
Highest FPS - 165 fps
Average FPS - 98.114035 fps

Voodoo 4 4500 PCI @ 184MHz
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 103 Seconds
Gameframes - 11226
Lowest FPS - 69 fps
Highest FPS - 167 fps
Average FPS - 108.708738 fps

K8 "Sempron Palermo" @ 800MHz, ram at 200 CL3

Voodoo 4 4500 PCI @ 184MHz:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 150 Seconds
Gameframes - 11226
Lowest FPS - 54 fps
Highest FPS - 103 fps
Average FPS - 74.620000 fps

Pentium II 400, ram at 100 cl2

Voodoo 4 4500 PCI @ 184MHz:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 224 Seconds
Gameframes - 11226
Lowest FPS - 33 fps
Highest FPS - 79 fps
Average FPS - 49.924107 fps

gandhig wrote:

Why so much difference between Direct3D 7 & 8/9??

Because D3D8/9 use shaders to speed up things as opposed to the (emulated) fixed-function pipeline of D3D7, presumably?

SPBHM wrote:

Pentium II 400, ram at 100 cl2 Voodoo 4 4500 PCI @ 184MHz: Expendable Timedemo Report -------------------------- Running at - 6 […]
Show full quote

Pentium II 400, ram at 100 cl2
Voodoo 4 4500 PCI @ 184MHz:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 224 Seconds
Gameframes - 11226
Lowest FPS - 33 fps
Highest FPS - 79 fps
Average FPS - 49.924107 fps

I don't understand what makes my system(min/max/avg=21/73/35) slower than even a PII 400? Just to be sure and rule out the factor of Voodoo's awesome performance in old games, can I trouble you to try the 8400GS PCI on your PII 400 system? BTW the PII 400 system is not running WinXP right, in which case that might be another variable. If the 8400GS PCI in the PII system brings down the Lowest FPS, say lower than 15, then I guess my chances are next to NIL for an improvement in mine. Only the min FPS is of more concern to me.

d1stortion wrote:

Because D3D8/9 use shaders to speed up things as opposed to the (emulated) fixed-function pipeline of D3D7, presumably?

Thanks for clearing that one, d1stortion.

gandhig wrote:

SPBHM wrote:

Pentium II 400, ram at 100 cl2 Voodoo 4 4500 PCI @ 184MHz: Expendable Timedemo Report -------------------------- Running at - 6 […]
Show full quote

Pentium II 400, ram at 100 cl2
Voodoo 4 4500 PCI @ 184MHz:
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 224 Seconds
Gameframes - 11226
Lowest FPS - 33 fps
Highest FPS - 79 fps
Average FPS - 49.924107 fps

I don't understand what makes my system(min/max/avg=21/73/35) slower than even a PII 400? Just to be sure and rule out the factor of Voodoo's awesome performance in old games, can I trouble you to try the 8400GS PCI on your PII 400 system? BTW the PII 400 system is not running WinXP right, in which case that might be another variable. If the 8400GS PCI in the PII system brings down the Lowest FPS, say lower than 15, then I guess my chances are next to NIL for an improvement in mine. Only the min FPS is of more concern to me.

unfortunately I'm unable to get the 8400GS to work on the Asus P2B I have, perhaps it's something like the problem you had having to edit the bios, regardless it wont work, insert the 8400GS and I got an error bip and the PC boots with no video at all.

but I'm certain it would be lower than yours, considering the performance of both cards with the K8 at 800Mhz

the Voodoo4 4500 PCI (and also the 9250 PCI), seem to perform a lot better in this than the newer geforces, (considering the huge difference in specs),

a 8400GS should never be slower than a 9250, (GPU vs GPU) if some other limitation was not involved
and the 8400GS running with a faster CPU,

Geforce 8400GS(G98) PCI stock(K8 1.8GHz):

Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 133 Seconds
Gameframes - 11226
Lowest FPS - 52 fps
Highest FPS - 140 fps
Average FPS - 84.218045 fps

Radeon 9250PCI (Tualatin 1.5GHz):
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 97 Seconds
Gameframes - 11226
Lowest FPS - 81 fps
Highest FPS - 160 fps
Average FPS - 114.948454 fps

for exemple on 3dm03
8400GS = Single/Multi Texturing=931/2951
9250 = Single/Multi Texturing=538/937

now looking at the voodoo, it should never beat the 8400GS
3dm01
8400GS:
FillrateSingle - 963.4
Fillratemulti - 2640.0

Voodoo 4 4500
FillrateSingle - 151.1
Fillratemulti - 289.2

But it's exactly what it's doing,
Voodoo 4 4500 PCI stock(K8 1.8GHz):
Expendable Timedemo Report
--------------------------
Running at - 640 x 480 x 32
Total Time - 114 Seconds
Gameframes - 11226
Lowest FPS - 60 fps
Highest FPS - 165 fps
Average FPS - 98.114035 fps

now that's also strangely close to the 9250 considering the GPU differences, but between CPU differences and driver optimizations you can perhaps explain this?

back to V4 vs 8400gs,it's specially obvious that something is wrong with the 8400GS performance if you compare both with the CPU at 800MHz
74 for the voodoo, 48 for the 8400GS, exact same PC and PCI slot, same image quality as far as I can tell...

to be really conclusive I would need a standard 8400GS G98, with PCI E, and the motherboard I'm using even supports PCIE 16x, a shame I don't have the card,

one thing is that the Voodoo 4 probably have some sort of driver optimization for this benchmark, and the Radeon 9000s are more likely to have than 2008-2010 Geforces, but I don't really think the difference would be so significant, without another factor (my guess is, bridge chip + lack of driver optimization for extremely low bandwidth, and also somehow higher CPU dependency)

P3 850 should absolutely have a higher result compared to what you have, but, a k8 800Mhz should also have a way better result than 48FPS I get with the 8400, as shown by the voodoo, and old tests

(using a geforce 2 GTS)

I'm certain the 800Mhz K8 can beat k7 at 1000MHz for gaming.

@SPBHM, your post hit the nail pretty hard in the coffin. I guess, it has been sufficiently proved that GPU's with PCI bridge onboard will have inferior(considerable) performance in comparison with native PCI GPU's(till proved otherwise, doubt anyone will).

8400GS PCI's lower performance against 9250 PCI(even with better CPU) might be due to more L2 cache on the Tualatin 1400-S. I even did a run with L2 cache disabled on my system with results of (min/max/avg=10/40/18). Thought of requesting you and Standard Def Steve, anyway it matters little now.

Think I have come to the crossroads. I have to decide between buying a native PCI GPU(i.e. if I can get one) for my present system (or) manage a fast enough system to have playable performance in conjunction with the GT520 PCI anomaly. I don't prefer buying online, so local market it is.

That aside, I will however try my best to get to the root of the issue as to what is the 'X' factor for this low performance, till something gives up. Maybe again knock the doors on the bridge vendor.

I think l2 cache is not a factor here, you have to consider the cache on the k8 sempron is probably faster than the cache on the tualatin, and the rest, the core, the memory controller (FSB 150 external vs ondie controller, SDR150 vs DDR450) and so on...
so I can only think the problem (8400GS slower than 9250) is on PCI/bridge bandwidth related stuff, and/or software optimization;
disabling all the cache is very different from having fast 256KB instead of fast 512KB (that's the only difference between the CPU I'm using and a Athlon 64 Venice, which was what, the 4th revision of A64), I have used multiple A64s and K8 Semprons over the years, performance was always not far, a little clock difference and you compensate the cut down l2.

let me quote Xbitlabs regarding l2 and k8.

"Before comparing the performance of the new AMD Sempron 2600+ to other CPUs, we wanted to check out the influence of the size of the L2 cache on the performance of the system with a K8 processor. When testing CPUs on ClawHammer and NewCastle cores earlier, we had noted that the reduction of the L2 cache from 1 megabyte to 512 kilobytes had had a very slight impact on the performance. There are several reasons for that.

First, all K8 processors have an exclusive cache. It means that data stored in the L1 cache are not duplicated in the L2 cache. Thus, the total effective capacity of the cache memory is the sum of L1 and L2 caches. Second, the L1 cache of K8 processors is rather large by itself, 128 kilobytes. Third, each K8 CPU has an integrated memory controller, which helps to reduce the memory access latencies in comparison to processors that communicate with the memory via the chipset. That’s why the reduction of the L2 cache doesn't affect the performance of K8 processors as dramatically as, for example, it does with CPUs of NetBurst architecture (but to be frank, when scrutinizing the new CPUs of the Pentium 4 6xx family, we noticed that the doubling of the L2 cache brought just a minor performance gain to them)."
http://www.xbitlabs.com/articles/cpu/display/ … 00_9.html#sect0

so I think you are looking at the wrong thing to explain the 9250 outperforming the 8400 (while the 8400 have the CPU performance advantage, and it's a faster GPU).

let me mention something else I briefly mentioned earlier, with highly constrained bandwidth nvidia could do some specific optimizations as seen here
http://forum.notebookreview.com/e-gpu-externa … tml#post6715285

they did some "PCIE Compression" for their PCIE x1 connected GPUs, which worked well, but it was only enabled using optimus (a software solution for Intel + NV GPUs on the same laptop) and a x1 link, now it makes me wonder, if they had something similar for desktop PCIE x1 (and that's how the drivers see our cards I think, the bridge chip is invisible at the end for it!?), would it make performance improve a lot? probably, and/or would it impose a higher CPU dependency for the compression work? I don't know, but I doubt they have something like that working for our cards anyway.

but back to the software optimization point, it's obvious that PCI cards are not really on Nvidia's radar, it's not even an official Nvidia product, so it would be reasonable to expect poor optimization from their side, in your case I think for playing many old games you would probably be extracting more performance with a fast older AGP-era PCI card, but ideally you would have PCI Express x16 or AGP.
the GT 520 will be better than older cards for newer games, and video decoding, but thta's it I think.

Sorry, the L2 cache observation was ignorant on my part then.

I think, a game's performance which I found to be comparable with similar systems is Q3A. I'm collecting some data to substantiate it and will report later.

I have decided to concentrate more on the Q3A benchmark results as there are lots of datapoints available for comparison in addition to comparable performance of my system against that of similar systems.
First the bench results,

Settings:
1) Command line: +set cd_nocd 1 +set s_initsound 0
2) Applied 'Defaults' from Setup Menu
3) Selected 'Normal' under Graphics Settings
4) Changed Video Mode/Color Depth to 1024x768/32 bit
5) Changed LIghting to 'Vertex'

Result: 112 FPS (Got 96 FPS with Lighting set to 'Lightmap', a huge hit of 16 FPS)

From reading online, I came across the following:

a) Quake 3 runs better on Nvidia cards in general.
b) The game has its own lighting engine. So hardware T&L will be of only partial benefit, i.e. only the transform part (Is this the reason for the huge hit while enabling lightmap
c) Lighting set to Vertex means actually Lighting disabled.
Can someone confirm whether these are correct?

Basically I was comparing my results with the results of P3 800 EB/815 from this chart, though it is probably benched with Lightmap enabled. However the 133 FSB along with Intel Chipset may compensate for that.

Other observations:
I was fiddling with the 'sensor' utility in the HWiNFO32 tool to measure the GPU core load in real-time while gaming:
I started logging with a scan interval of 500ms with 3 different games:
a) Q3A = 50 % on AVG and 61% MAX
b) Unreal1 = 40 % on AVG and 60% MAX
c) Expendable = 25 % on AVG and 30% MAX
and trying to find a pattern. Any suggestions/directions are highly appreciated.

@SPBHM, Just wanted you to know that there is a possibly simple workaround for getting the 8400GS PCI working on the Asus P2B board (in Windows) without a permanent bios mod. If you are interested or it is of any use to you, please let me know.
The tool is attached and some details are in page-6 of this thread.

@feipoa, If you find time, can you check out the Matrox Parhelia with the DOS PCI tool in your ServerSet system?

this is what I get with PCI tool on the P2B + 8400GS

SPBHM wrote:

this is what I get with PCI tool on the P2B + 8400GS

I presume you ran the tool without any display, probably by loading the command in autoexec.bat or am I missing something?

Since the 8400GS PCI and the onboard PCI bridge are not in the list,
1) Can you post the snapshot of IO, Memory & IRQ pages by running the msinfo32 with 8400GS PCI in your Sempron system (This is required for identifying the resource requirements of the graphics card). If you can post the PCI tool output of this Sempron system by booting to DOS, that will help a little more.
2) Can you also post the snapshot of IO, Memory & IRQ pages by running the msinfo32 with only Voodoo 4 4500 PCI in your ASUS P2B system(This will be helpful to avoid resource conflicts while manually assigning the resources to the 8400GS PCI later).

gandhig wrote:

I presume you ran the tool without any display, probably by loading the command in autoexec.bat or am I missing something? […]
Show full quote

SPBHM wrote:

this is what I get with PCI tool on the P2B + 8400GS

I presume you ran the tool without any display, probably by loading the command in autoexec.bat or am I missing something?

Since the 8400GS PCI and the onboard PCI bridge are not in the list,
1) Can you post the snapshot of IO, Memory & IRQ pages by running the msinfo32 with 8400GS PCI in your Sempron system (This is required for identifying the resource requirements of the graphics card). If you can post the PCI tool output of this Sempron system by booting to DOS, that will help a little more.
2) Can you also post the snapshot of IO, Memory & IRQ pages by running the msinfo32 with only Voodoo 4 4500 PCI in your ASUS P2B system(This will be helpful to avoid resource conflicts while manually assigning the resources to the 8400GS PCI later).

I've just typed the commands with no display

files are attached but my Windows is not in English,
also for the Sempron PC the onboard VGA is actually always active, I have it disabled on windows trough the device manager, but I have no way of disabling it via bios when using just a PCI Video Card.

Also, the P2B is running 98, but I can easily install XP ( for the 8400GS drivers)

Hi!

Could you please share how you do the benchmarks? Basically what command line options and what how could one benchmark the same way to compare with your results?

Especially Expandable.

Thank you!

for Expendable add -timedemo to go.exe, it generates a .txt file on the game folder with the results

Cool SPBHM, I will check these logs and revert.
Mau1wurf1977, I have not changed anything in the in-game colorful setup menu. In the external setup menu, the following only was changed:
Resolution : 640 x 480 x 32
Use low resolution movie : enabled (not for any particular reason)
BTW, enabling and disabling other options didn't affect the end results very much for me atleast except disabling sound obviously.
Edit : I don't have access to the system this weekend as I'm out. Maybe someone can recommend an ideal setup for this expendable benching

SPBHM, I have attached the files. These files have to be extracted and added to your DOS boot disk. All the commands are added to the AUTOEXEC.BAT.

However, I missed out in asking an info that is needed. It is the slot number of the PCI slot in which you will insert the 8400GS.

To find out the ID of PCI slot(Bus:Device:Function format), you can boot to DOS with only Voodoo PCI card so that you will have the display and then run the command "pcitool /B". This will list out all the devices in brief with the above B:D:F format (for example, 00:02:00). If you know the Vendor and Device ID of this Voodoo card, you can easily find out the B:D:F for that particular slot/card from the list that is displayed.

Once you find out the B:D:F of the particular PCI slot, you have to replace the 'x' in the attached PCITCMD.INI with the actual Device number. The lines starting with the ';' are comments. Then it is time to remove the Voodoo and insert the 8400GS in the same PCI slot and simply boot to DOS. If I'm unable to convey this to you clearly, just run the pcitool /B command with only the Voodoo card, note down the results and post it. I will modify the files as required and resend it.

This is how it works. The AUTOEXEC.BAT will first run the pcitool to manually assign the resources to the PCI bridge and then chainload Win98 OS. I'm presuming that you never got any display out of the 8400GS PCI when inserted in your P2B system. If everything goes well you will get the 'Safe mode' type VGA display from the 8400GS PCI. Then if you want you can install XP for loading drivers for 8400GS.