Meanwhile with a P-III 600/256MB/440BX/AGP2x enabled @default bench setting, Catalyst 4.3 drivers:

3DMark2000 with "T&L enabled"... get a 2060 score

3DMark2000 without T&L and with P-III option get a 2477 score (???) with even higher Polygon Count (like 500Ktriangles/s more, on a P3-600 Katmai I can't explain this I'd expect that from a faster Pentium III not the early core on a 440BX).

3DMark2000 without T&L and without P-III option and with Software T&L get a 2286 score and seems having only a bit lower Polygon count of the T&L hw test results (even lower in the 8 lights test) but higher frame/s in the Adventure test.

What does it mean? The T&L seems actually being there cause there's no much difference from 1 Lights Polygons test and 8 lights while instead in the Software T&L the value decrease quite more. Instead the Pentium III option (SSE?) surpass them both.. (EDIT: see last posts)

The Treemark NV original T&L test start with default basic setting (command " -tltest ") and results in 8,4 fps @ 1024 16bit and 8.3 fps @ 640 16bit, this doesn't sound promising and more like a cpu limit..
(polygons per frame 33972 polygons per second 279145.)

I don't remember the command option to test it disabling a generic hw T&L unit, I think there was an option for that. Anyway result seems low! The original Geforce 256 I think was around 30 fps maybe higher in the simple test wasn't it? Even the S3 Savage2K T&L I think was faster.. I don't think it might be a real T&L more a driver problem. I understand this wasn't a gamer card of course but I wondered if it could have had a slow T&L hw unit simply disabled by sw or not, because also another thing the CPU load during the -tltest test still around 98%. Is the Geforce 256 on a time correct machine has similar cpu usage in that test?

I ran also 3DMark2001 @ default settings with the same system and with "T&L enabled" (it even gives the "Pure T&L hardware" option this sound even less promising..) result in a bit higher score 1100 compared to the Software D3D T&L 1050 points and both with similar 1,5MPolygons/s simple test. Maybe the right test would be to use a slow Celeron 333 Slot cpu and to see if the "GPU" is actually accelerating this rendering part. But I didn't want to change the cpu cause the Slot bus solder points force
The same test @ 640x480-16bit with T&L enabled get a 1168 score, without it with Software T&L get 1160 score, more or less 1 to 2 fps more for the "hw T&L". I don't get if it's actually an hardware unit or a driver thing for compatibility. 3DMark2000 shows some difference anyway from Hardware T&L and Software T&L and that I can't explain.. if it was a software rendering it should be similar like indeed is in 3DMark2001, but maybe that's not the right bench to test this card.
Sounds like a CPU limited scenario more than a hw T&L one.

I was reading old reviews about this and it seems to be said it lacked the T&L but it doesn't sound like a definitive info more based on the final core size and the general slow speed compared to the time correct alternatives (Geforce2 MX and their own R100 SDR). But reading even more old discussions, it might also be that some drivers around might have left that option in the registry for testing but not actually supporting any T&L; the one I've installed found are 2004 ver. 4.14.01.9138. But I wonder if might be some T&L functions remained in the RV100 beside the common memories I also was thinking not having much left of the R100 design, it'd be a nice thing to know at the end it might have had that unit but disabled but I don't think so. I try using latest 6.2 drivers too see if that driver option is disabled.

No, with the Catalyst 6.2 driver the T&L option is still there.. but I don't know if it's cause the previous driver installation. Anyway again 3DMark2000 at 640x480 16bit, Double buffer:

T&L "Hardware" "on"...

High polygon count (1/4/8 light): 2482 Ktriangles/s - 2064 - 1737 (much increased!)

Software D3D T&L:

High polygon count (1/4/8 light): 1125 Ktriangles/s - 1008 - 903 (much lower!)

Pentium III SSE flag only:

High polygon count (1/4/8 light): 1776 Ktriangles/s - 1648 - 1501

I don't understand here.. previously the SSE option gave higher polygon count, with latest driver is the contrary with lower triangles rate and higher for the "T&L" theorically on..

But the Treemark T&L test doesn't show any improments in both 640x480 and 1024x768 modes, still 8,3 fps result and this sound like a CPU limited result. Do anyone remembers this benchmark command line options?

3DMark2001 SE with 6.2 drivers, 640x480 16bit double buffer:

"Hardware T&L on"...
High polygon count (1/8 lights): 2,2 MTriangles/s - 1,2 (higher!)

Software T&L
High polygon count (1/8 lights): 1,1 MTriangles/s - 0,7 (lower...)

@ 1024x768 default same score..

Before with the 4.3 driver it resulted both in 1,5 Mt/s with "hw" or sw T&L in simple 1 light test. I can't understand why now this difference.. why software T&L now seems slower and hardware T&L that might be a compatibility driver thing (?) is higher.. I'm now running the whole test to see the final scores. I might try to reinstall the config again and to see how installing only the final drivers works. If it's CPU limited score should be more or less identical enabling or not those options and instead before they were quite similar, now much different but I can't explain the lower Software T&L polygon count.
Win 98 is still the same with 2004 Dx9/patches config.

3DMark2000 beside lower final scores with 6.2 drivers, result in a similar situation with "T&L on" having a strangely much higher (similar to the previous Pentium III option results because now that is slower too) while the Software T&L option get a much slower polygons/s and no reasons I can understand for it. It's difficult to say what's happening, I suppose it didn't have the T&L unit but drivers changed results seems strange beside that "Hardware T&L" option is not clear if a compatibility oriented Kyro 2 style "feature", a real enabled theorically existing T&L slow unit or a driver error having to support a lot of different cards. I think to remember it's not the first time I've seen that "Hardware T&L" option in my past tests with this card.

Any opinions/suggestions?

Anyway I installed the Celeron 333Mhz and now there's no SSE or Software T&L ran by a fast FPU of the P-III confusing the situation.. let's see the polygons results..

3Dmark2000 640x480 16 bit Polygons Count test:

D3D "Hardware" T&L "on": like 140 Ktriangles/s (1 light) not even mentioning the other two tests..
D3D Software T&L: around 450 Ktriangles/s (1 light) .. with a Celeron 333Mhz I doubt that it could be faster than an hw T&L unit ever designed in any gpu.

I think it does not have an hw T&L unit left from the R100 core after all.. strangely now the polygons/s is lower than Software D3D I wonder if needed SSE after all. Maybe that option was there for an error or for compatibility reasons.

Treemark simple test 640x480 16bit: 2,66 fps - 1024x768 32bit: 2,68 fps..I don't know.. sounds like anything T&L related run on the CPU if it decreased that much isn't it?

Remove the cooler and see what chip is underneath, dunno why you havent started with this 😀

Dolenc wrote on 2022-02-01, 13:52:

Remove the cooler and see what chip is underneath, dunno why you havent started with this 😀

I was thinking to do it but seems having a strong tape on it, the card anyway is quite cold during test, too much for a R100 core and the thin heatsink plus the reduced back PCB core area I'd have no doubts is a Radeon VE chip. 😉

I'll try anyway. 😉

My guess is that the video driver is emulating hardware t&l. So in the end you get different kinds of software t&l but not true hardware acceleration. On Radeon 7500 there is a big difference when hardware t&l is enabled, it goes from ~4M triangles/sec with software only to ~14M with hardware acceleration (in 3Dmark01 tests).

bakemono wrote on 2022-02-01, 15:07:

My guess is that the video driver is emulating hardware t&l. So in the end you get different kinds of software t&l but not true hardware acceleration. On Radeon 7500 there is a big difference when hardware t&l is enabled, it goes from ~4M triangles/sec with software only to ~14M with hardware acceleration (in 3Dmark01 tests).

Interesting. Have you tried with faster cpu with which processor more or less the Software T&L speed equal the Hardware T&L?

I think I've more or less understood what might be the problem. It's a similar situation seen in the past with the Rage Mobility serie of desktop AGP cards. Different generic ATi drivers I'm trying seems to not really understand which "GPU" is while indeed the video card is a "Sapphire R7000 64M 64bit DDR AGP VGA/TVO/DVI-I" and considering the chip size, the back pcb core area, it can't be the huge R100 gpu but I'm sure the RV100, after all it's a common card not some strange revisions.
But at the end the INF files point to the generic Radeon 7000 series drivers that I think are compatible but more polished for the R100 SDR and forcing a Radeon VE driver notice it's not designed for it; I remember having seen similar situations with the Mobility-P AGP card I have with "difficult" driver installations. I might search for the original Sapphire driver to test it but the driver I'm using on a new installation and config (KT133 and Athlon 1000) seems to work like before but still the HW T&L option is there probably intended for the R100 and left there I might imagine for compatibility reasons. But I suppose this was not a "S3 Savage 2K-like" situation with a disabled T&L and as read in the old reviews, probably really lacking it in the design to reduce costs and having a cheaper still good solution. After all I can say it's not as bad as often read, it not that slow and has a very good 2D quality and supporting most video accelerations of those times. I've tested it with IDCT mpeg2 decoding and on an Athlon 1000 the mpeg2 decoding on the cpu passed from almost 50% usage to 15-20%. Obviously final quality and support depends on which sw is used, most had quite different results more or less visible.

If you would just read reviews from the time you would know RV100 did not have the TCL.

I remembered as a far memory the missing TCL unit and I read some of the time correct reviews that just wrote about the lack of it but I imagine this was probably confirmed by correct drivers clearly missing that option in benchmarks, I suppose no official specs usually listed that missing feature or any others; sometimes might happen that some features might simply be disabled even if this seems not the case; some multi core cpu had one disabled or the R300 gpu in the early 9500 Pro cards had a similar situation too. In this case once installed the card with some time correct installation setups it didn't work as expected (desktop came back to 16 colors after reboot with no errors), I installed manually the autodetected "Radeon 7000 series" driver directly with the W98 INF search and that worked as usual so I suppose this might require a specific driver version to install without effort not just the usual generic package. And if I installed the Radeon VE listed driver it would notice the wrong hardware so something felt interesting there. I suppose that "hw" t&L feature appears for that (while the driver works correctly no problem anywhere) and I was just wondering if something more of the R100 engine might have been (ipothetically) left or "enabled later" like a Savage 2K or Kyro II sw EnT&L option; after all we can guess the design may still be "based" around the R100 project. So it is probably just a driver problem and cause in those times I didn't have much interest in this card I never cared a lot about guessing how much different it was or not just like some of those reviews more focusing on the multi-monitor output, the market sector and doing benchmarks and final thoughts.

Some final tests using the CPU flags only in 3DMark, the 3DMark99 scores is 6210, the 3DMark2000 score is 2883 with an Athlon 1000/KT133 and Win 98. 😉

Some final update: I tried reinstalling the o.s. starting only with the 6.2 latest WME driver package and while now the installation worked still the "Hardware" T&L option is there in the benchmarks. The RV100 chip most probably don't have that in hardware but anyway the driver gives that, without any forced installation. And the only reason I can think of is the compatibility like a sort of EnT&L situation as seen in the original Kyro II (not SE version) where latest drivers "enabled" that software solution maybe "proprietary". But I found interesting that some benchmarks like 3DMark2001 with the "Hardware" T&L enabled compared to the generic Software T&L (the usual D3D option) show rendering polygons errors in some of the benchmarks while it doesn't happen with the other options. So still I think this is interesting also how much triangles/s rate changed from 4.3 drivers to 6.2.