Which NVidia chips support HW-accel rendering to memory?

I would like to experiment with server-side opengl rendering in response to
web requests on Windows 2000 and Linux. My current card is a TNT2, but I
would like to purchase a GeForce2 or GeForce3 to experiment properly, since
AFAIK the TNT definitely does not accelerate rendering to a bitmap. When I
looked into this last, specialty CAD chips (FireGL, etc.) were the only
options, but I gather that NVidia supports this feature now, too.

I’m interested in multithread contention for the GPU, reuse of display lists
and other caches, and whether the final-stage conversion to JPEG or PNG will
ultimately make the whole exercise impractical.

Can anyone tell me whether the GeForce2 or (only) GeForce3 can render to an
offscreen, in-memory buffer?

Is there a specific reference on the comparative features of the two chip
generations for this particular task?