Interesting results meljor, thanks for doing this. I recently got myself a P3B-F with a P3-600 Katmai and was wondering which AGP card to use from my collection; the charts you made are very helpful in planning my future build. Cheers 😀

Thank you. The question pops up every now and then so i thought let's test it.

Funny how m64 gets worse results than TNT1 😀 Do you have a normal TNT2 to try?

Nice benchmark, great work!

Odd though how your Voodoo3 PCI and Voodoo4 AGP achieve similar scores. I found the V3 PCI to be painfully slow compared to a V3 AGP.

Just my question too. Why not tested a TNT2 Ultra?
I do have an G400 Dualhead AGP, and if it's better than the TNT2 Ultra in my P3-500, then it's going to replace it.
I think that the G400 somehow is more capable than some people are giving it credit for.
Shure is cheaper than getting one of those Voodoo PCI Cards. Hmmmm. Wonder why! 😁

Great job none the less. Good work and thumbs up.

RacoonRider wrote:

Funny how m64 gets worse results than TNT1 😀 Do you have a normal TNT2 to try?

I don't! still looking for an ultra card. I have other cards but i couldn't be bothered as they wouldn't paint a different picture or are build in systems. Only interesting ones would be the 6800ultra and v5 5500 agp version but i didn't want to take them out of their home 🤣

jwt27 wrote:

Nice benchmark, great work!

Odd though how your Voodoo3 PCI and Voodoo4 AGP achieve similar scores. I found the V3 PCI to be painfully slow compared to a V3 AGP.

I think the v4 needs a little bit more cpu power to make a difference, but it isn't a very fast card. At current 3dfx prices the 3000/3500 are the best choice.

A while back i tested some cards on my p3-1400. In 3dmark99 (only tested 800x600) the 2000 pci scored 6049, the 3000 agp 6784, the v3 3500 did 7486 and i didn't have the 4500 back then but i simulated with v5 agp with one chip off: 8181 3dmarks

In the p3-1400 the 2000 pci was 5% faster than the agp 2000, it could be that i tested (in both cases) the sgram version which seems to be quicker. Have to check that and will test both sd- and sgram cards on the k6-2 benchmarks i'm running now.

Since the ti4200 has no external power i'm a little scared of frying the ss7 board so i might keep this one out of the tests on this platform, any comments on that?

RacoonRider wrote:

Funny how m64 gets worse results than TNT1 😀 Do you have a normal TNT2 to try?

Not so funny when you consider the M64 has a 64-bit memory bus while the original TNT is 128-bit. The M64 core might be clocked slightly higher, but not enough to make up for the throttling that the memory bus inflicts on it.

On paper, the performance of the core of the M64 is identical to the core of 128-bit TNT2, because it's the same core, but it's when it tries moving information over the memory bus that it falls down.

http://en.wikipedia.org/wiki/List_of_Nvidia_g … its#Pre-GeForce

You'll notice the fill rates are higher for the M64, but in reality it never achieves those speeds. The original TNT can go flat out at it's maximum rated speed without being hamstrung by memory bandwith constraints. The early GeForces were similarly hamstrung by a poor memory architecture that couldn't keep up with the demands of the core.

wow.... started benching on a ms-5169 ali V ss7 board and it is giving me a VERY hard Nvidia time.

Was almost done with the voodoo's and matrox cards which were running as expected but hit the first bump with TNT1. After some driver switching i got it running fine. So in went the tnt2 m64... i tried 2 of these and they completely destroyed agp support 😐

Got it up and running again and tried gf2mx and gf4 mx440 and again lockup after lockup 😵 Now even 3dmark wouldn't start anymore.

Got a few more things to try but i'm afraid i have to swich to a p5a board and do it all again. It will probably give me troubles with one of the cards but it should work with most of them.

Woow well done man! 😀
I did something like that some months ago with a Pentium 3 700mhz coppermine + Msi 370 VIA with some old GPU
I have two TNT2 Ultra if you want, I can do some benchmarks 😉

I dont see any Geforce III in this test.

This registry settings should help with Nvidia cards 😎

[HKEY_LOCAL_MACHINE\Software\ALi\ALi AGP Driver]
"AGPDataRate"="1"
"ForceSBADisable"="0"
"ForceAGPEnable"="1"
"FrameBufferSize"="C"
"FrameBufferDisable"="0"
"GATMode"="0"
"K6SetEWBEC"="2"
"UseUCForWC"="0"
"GTLBAlwaysFetch"="0"
"ResetAGPCommand"="0"
"MaxPCIRetryCounter"="0"
"AGPTiming"="0"
"PCIMode"="0"
"ForceAssertRequest"="0"
"DisablePCIReadPrefetch"="0"
"AGPDelayClock"=hex:10
"SDRAMDelayClock"=hex:10
"K6WriteAllocate"="2"
"PADInputLevel"="1"

Robin4 wrote:

I dont see any Geforce III in this test.

I have one Geforce 3 TI200
I try to do these benchmarks next weekend

falloutboy wrote:

This registry settings should help with Nvidia cards :cool: […]
Show full quote

This registry settings should help with Nvidia cards 😎

[HKEY_LOCAL_MACHINE\Software\ALi\ALi AGP Driver]
"AGPDataRate"="1"
"ForceSBADisable"="0"
"ForceAGPEnable"="1"
"FrameBufferSize"="C"
"FrameBufferDisable"="0"
"GATMode"="0"
"K6SetEWBEC"="2"
"UseUCForWC"="0"
"GTLBAlwaysFetch"="0"
"ResetAGPCommand"="0"
"MaxPCIRetryCounter"="0"
"AGPTiming"="0"
"PCIMode"="0"
"ForceAssertRequest"="0"
"DisablePCIReadPrefetch"="0"
"AGPDelayClock"=hex:10
"SDRAMDelayClock"=hex:10
"K6WriteAllocate"="2"
"PADInputLevel"="1"

I will try! Thanks for sharing
I have done some tests last summer and it's better to use old detonator version like 10.X or 6.X

sliderider wrote:

RacoonRider wrote:

You'll notice the fill rates are higher for the M64, but in reality it never achieves those speeds. The original TNT can go flat out at it's maximum rated speed without being hamstrung by memory bandwith constraints. The early GeForces were similarly hamstrung by a poor memory architecture that couldn't keep up with the demands of the core.

In reality no bottleneck is absolute. There is more to it than fillrate:bandwidth ratio. When using resolutions in which these cards usually deliver smooth framerates, the m64 often wins.
I would be also careful to call memory architecture of early GeForces poor. Yeah it was the weak point, but with exception of Kyros there weren't chips with better overall efficiency in that generation. It got the job done.

Putas wrote:

I would be also careful to call memory architecture of early GeForces poor. Yeah it was the weak point, but with exception of Kyros there weren't chips with better overall efficiency in that generation. It got the job done.

Do you consider Radeon to be outside of the early GeForce generation or not sufficiently more efficient?

definitely not sufficiently more efficient
bigger die, feature parity, less performance

Putas wrote:

definitely not sufficiently more efficient
bigger die, feature parity, less performance

🤣
Interesting. I didn't know Radeon was the larger chip. I wonder if most of that is caused by the additional efficiency improving hardware.

By changing 2 registry settings i got the cards working on the msi board. Got all the data up to the geforce4 ti (it worked fine but i wouldn't trust it long term as it takes a lot of power from the agp slot).

I included a v3 3500tv and matrox g450, excluded FX cards as they really refused to work on this board.

Coming week i will make the chart.