The Serpent Rider wrote on 2020-06-04, 22:20:
Most likely Riva TNT2 or 3DLabs OEM card.
EDIT: Actually no. It's their own professional 3D chip named IBM POWER GXT2000P Graphics Accelerator. Apparently it was low cost (for IBM anyway) solution. No Direct3D support, so goodbye 90% of games. Probably very mediocre even for pro apps.
Apologies for digging up an old thread, which I discovered when trying to identify the same IBM card.
The IBM press announcement does not specify the chip maker. The absence of electrolytics with a 1998 silkscreen makes me think of 3DLabs.
I found one Linux enthusiast asserted, "IBM also offered some rebranded graphics cards from other manufacturers that were supported by AIX and Linux: GXT2000P (3D Labs Permedia 2), GXT110P (S3 Trio64V+), GXT120P (Matrox Millenium I), GXT130P (Matrox Millenium II) and GXT135P (Matrox G450)."
https://blog-geierb-de.translate.goog/debian- … &_x_tr_pto=wapp
The layout resembles a 3DLabs Permedia2, but the date is a little newer and the 10ns SGRAM leaves headroom. Comparing PC retail cards, the mismatch is an increase from 8Mb to 32Mb and the addition of active cooling. My gut leans more towards a customised Permedia2V. It looks less like a Permedia3, which would be too new anyway.
I cannot explain why Linux users preferred simple frame buffer on GXT2000P, rather than trying Permedia2 Glint drivers. Similarly, Windows drivers would need INF hacking to match IBM vendor ID and nobody seems to have posted success/failure on that platform either.
I don't think this is an experiment I'll be investing in, but if I had a GTX2000P I'd try booting it as a secondary device and forcing Linux Glint drivers because they are open source and there may be helpful documentation. If that worked, I'd copy the lessons learned to a Windows INF file.. hmm, but we need to wonder why there are no reports of anyone trying this! 🙁
Desktop timeline [ MOS 7501 → 68030 → x86(P5/MMX) → x86(K6-2) → x86(K7*) → PPC(G3*) → x86-64(K8) → x86-64(Xeon) → x86-64(i5) → x86-64(i7) ] * lost