Hi,

Are you sure that 16Mb TNT2 is a V770? I personally don't remember V770s to come with only 16Mb RAM, and all the reputable information sources I have found only mentions the 32Mb version:
https://theretroweb.com/expansioncards/s/diamond-viper-v770
https://theretroweb.com/expansioncards/s/diam … iper-v770-ultra
https://www.vgamuseum.info/index.php/cpu/item … vidia-riva-tnt2
https://thandor.net/object/401

Obviously I could be wrong about that, and it is possible that the 16Mb version is simply too rare to be documented on these sites, but the only mention I have found online about a 16Mb version of the V770 is by E-bay sellers.

You're benchmarking with 3dMark99, I would use 2k and benchmark 800x600 and 1024x768, both in 16-bit and 32-bit color modes. If you still see no difference, something's up, but you could/should be able to see some difference between the cards methinks.

MagefromAntares wrote on Yesterday, 10:05:

Are you sure that 16Mb TNT2 is a V770?

For a while, I was confused myself as I had it listed as a 32mb part in my inventory. It is 16mb
I just pulled it and checked. It is this card here, mem chips only on one side of the board:

https://www.techpowerup.com/gpu-specs/riva-tnt2.c4130

JayAlien wrote on Yesterday, 11:26:

For a while, I was confused myself as I had it listed as a 32mb part in my inventory. It is 16mb I just pulled it and checked. I […]
Show full quote

MagefromAntares wrote on Yesterday, 10:05:

Are you sure that 16Mb TNT2 is a V770?

For a while, I was confused myself as I had it listed as a 32mb part in my inventory. It is 16mb
I just pulled it and checked. It is this card here, mem chips only on one side of the board:

https://www.techpowerup.com/gpu-specs/riva-tnt2.c4130

Hmm weird, it seems to be then a V770 with only one side populated with memory, that actually makes me remember the V550 cards, I'm not sure how the memory bus is routed in the PCB, if the half of the memory chips are removed without redesigning the memory bus it is possible that it actually only has 64-bits of memory bus to talk to the chips, even though the NV5 chip has the lines for 128-bits, this is purely conjecture, but would explain the similar results.

I would be also interested in the results of doing what @leonardo have suggested, if the performance remains so close to the M64 with 3D 2k and the resolutions given, then I would consider the possibility that by not installing those chips they have reduced the memory bus of the card to the half.

However the ways to actually prove it would be not easy for example using a scope to measure the interaction between the memory chips and the NV5 chip, however the memory bus runs at 150Mhz, and oscilloscopes above 100Mhz are generally not in the cheap category in fact quite the opposite unfortunately 🙁... The BIOS of the card might also be mined for information regarding how it configures the chip, it might set something that lowers the performance.

I'm comparing against old benchmarks on the same platform, the TNT2 M64 isn't in the computer at the moment. However, I had benchmarked it before at 16bit color depth. I ran the same bench with the TNT2 16mb. Results are basically indistinguishable from each other.

3dMark2000
1024x768 (16-bit) 800x600 (16-bit)
TNT2 M64 32MB (64bit bus) 1420 1470
TNT2 16MB (128bit bus width) 1459 1477

Half the chips could indeed lead to 64-bit bus width.

Can you post a picture of both the front and rear of the card? Perhaps you can also run AIDA32 or SiSoft Sandra to get a few extra details on the software side 😀.

Thandor wrote on Yesterday, 12:25:

Can you post a picture of both the front and rear of the card?

Of course! I also included the other TNT2 that I used to have in the computer. It's entirely possible neither of these are quite what I thought they were.

Thandor wrote on Yesterday, 12:25:

Perhaps you can also run AIDA32 or SiSoft Sandra to get a few extra details on the software side 😀.

I installed SiSoft Sandra, this is what I see for the details of the 16mb card (attached system report)

This just looks like a classic case of a cpu bottleneck to me.
Of course I may be wrong.
Crank up the resolution to see if the weaker card is exposed.

The Diamond is a 128-bit and the ASUS is a 64-bit card. There may be clock variance between the cards, hiding some of the difference. Try higher resolutions, a single texturing test would be enough.

Sandra doesn't really give much information about the card sadly.

I don't have any 3dmark99 scores on hand to say whether the M64 seems high or the V770 seems low.

I think something odd is going on there, because even a combination of CPU bottleneck and different clock speeds on the cards shouldn't make them perform equally on a PII 400Mhz.

Normally a TNT2 M64 (non-Vanta) and a standard TNT2 have the same core clock. The M64 will have 64bit memory clocked at 143Mhz and the standard TNT2 will have 128bit memory clocked at 150Mhz. Your M64 card has 6ns memory though, which should be good up to 166Mhz, while your V770 has 7ns memory, which is rated for 143Mhz. If the M64 really is a "Vanta", then I don't know what is going on because that should be a 100Mhz core and 64bit memory at 125Mhz, which is obviously going to be waaaay slower than a standard TNT2.

So, potentially, the M64 could have a higher clock speed which would bring it a little closer to the full 128bit card's performance. If it also has a higher core clock for some reason that could make the two cards trade blows in different tests depending on if memory bandwidth or pixel\texel throughput are the limiting factor. That said, I don't think it was all that common for well known companies to mess with core\memory clock speeds on TNT2 models back then. They generally stuck to whatever nvidia's spec was for each particular model (Vanta, M64, TNT2, etc.).

I would try to get some more info about both cards first. I like to use Everest 2.01 for testing card in Windows 98SE. It usually does an okay job of reporting clock speeds and other things. Along with that, Rivatuner 2.24 can sometimes be used to figure out clock speeds and specs of cards that don't show up properly in other programs.

... oh, actually, looking at your Sandra report again, it looks like you are using driver version 30.82? That is from the Geforce 4 Ti era, which is much newer than what I would use for a TNT2 on a PII 400. I would try an older driver... like maybe 3.68 from this page. Make sure to totally uninstall the current drivers first (reboot to be sure the card isn't detected again).

Thanks for the suggestions and help.

I removed the driver and used the older version 3.68 per your suggestion. It doesn't seem to have affected the benchmarks very much:

3dMark99 (800x600) 3dMark2000 (1024x768)
TNT2 16MB (128bit bus width) 3316 1459
TNT2 16MB (used older driver 3.68) 3496 1479

I've attached the Everest Report. This is confusing me, it confirms a 128-bit bus but the memory clock is 143MHz which is the clock speed of the TNT2 M64 or TNT2 M64 Pro.

The known variants are listed here:
https://www.techpowerup.com/gpu-specs/?q=tnt2

edit: Looking at the specs on techpowerup, I *think* the 32MB card may be the TNT2 M64 Pro. The 16MB card currently in the computer looks to be a vanilla TNT2 seemingly with lower memory speeds. The theoretical speed listed for both of these cards is Pixel Rate - 250.0 MPixel/s, Texture Rate - 250.0 MTexel/s so perhaps that's why I'm not really seeing any difference in performance between them.

This is shaping up nicely. Pro cards tend to be clocked faster. The stated pixel and texel fill rates are theoretical abilities of the chip alone. But if you run a pixel fill rate test at a higher resolution, you will clearly see the importance of memory bandwidth.