The 480Mpixels/second fill rate of the GeForce 256 is a step forward in comparison to the TNT2 Ultra's 350Mpixels. According to NVIDIA's white papers, it should be more than capable in the polygon count department with over 15 million fully filtered and lit little blighters (triangles) per second. In reality and with the current 120MHz core speed (get your calculators out folks!) it's much closer to 10 million fully lit and filtered triangles per second (take note: there are no games today using 10million polys nor will there likely be for some time). Also, it's worth bearing in mind that the initial SDRAM version will not have enough bandwidth to reach that 480Mpixels/second theoretical peak fill rate and instead DDR (double-data rate) memory is needed. A more realistic fill rate for this more limited 128-bit memory bus bandwidth GeForce 256 seems closer to 275 MPixels per second (3DMark's Multitexturing tests showed this but further driver enhancements could affect this), which is still plenty fast.
During our summer visit to Taiwan's Computex trade show in June, we spoke to various OEMs and indeed TSMC - the chaps (and ladies mostly) that work so hard to FAB the chips for both NVIDIA and 3dfx - about the possibility of the (then NV10) being set in .18micron. It was originally hoped that TSMC could manage to shrink the die from .25mircon (TNT2) down to .18micron along with the 23million transistors but it seems as though it was a little too small, too soon and yields just aren't 'there' yet. Hence a compromise of .22micron is the order of the day (TNT2-A chips will now also be on .22micron). So because NVIDIA seized the opportunity to be first to market this month with a next generation part, the .22micron GeForce 256 does need active heat dissipation in the size and shape of a heatsink and fan combo. Every GeForce will be a little on the hot side and not particularly overclockable initially. This may yet turn out to be an interim solution as NVIDIA may or may not add other NV10 (NV15 & 20?) skews in the future when TSMC attempts to obtain better yields at the .18Micron level. The benefits would trickle down to the end user in terms of the overall chip price being a few dollars cheaper (less wafer etc…) to the OEM possibly at the .18Micron level. But this looks to be some way off. In the meantime all GeForce 256s will have the graphics clock set to a rump 120MHz.
We're not going to list a full page of specs and if you really want to (you don't have to) feel free to see them here. For the rest of you, on to those scores. Let the games begin eh?
For this first SE GeForce 256 test run, we dusted off several different PC set ups and prepped them via ScanDisk and a Defrag for our benchmark suite. The test rigs included:
Processors: Intel Pentium III 600MHz, Celeron 300A @450MHz
Memory: 128MB of WinTec PC100 RAM
Motherboard: Abit BE6 2.0
Hard Drive: Quantum Fireball 8.4 Gig HD.
Sound Card: Creative Sound Blaster Live! Value
CD/DVD: Creative DXr5
3COM LAN Card.
Windows 98 Second Edition with DirectX 6.1 & 7 installed
Monitor: Sony Trinitron 400PS
Haggen Dazs: Two tubs of
We also tested on the AMD Athlon 650MHz power-horse, which included the following specifications:
AMD Athlon 650, 512KB L2 Cache (1/2 core speed)
AMD Production Mainboard Rev. B with 750 chipset
Western Digital Expert WD41800 18GB UDMA 66 HD
128MB of EMS HSDRAM
CD/DVD: Creative DXr5
3COM LAN Card
Windows 98 Second Edition with DirectX 6.1 & 7 installed
We also used the following video cards with the following sets of drivers:
NVIDIA GeFORCE 256 32MB DDR-SGRAM -3.47 reference drivers
Guillemot 3D Prophet- For Direct 3D and DX6 benchmarks we were instructed to use the NVIDIA 2.08 Detonator drivers. For all other OpenGL benchmarks we were given NVIDIA's 3.38 GeForce 256 drivers.
Guillemot MAXI Gamer Xentor 32 with NVIDIA 2.08 Detonator Drivers
Guillemot MAXI Gamer Xentor 32 SE, with NVIDIA 2.08 Detonator Drivers and Falcon Northwest's own overclocking utility.
3dfx Voodoo3 3500 with 3dfx driver version 1.02.18
And some tub-thumping sounds from The Orb, Prodigy and Fluke CD's to keep us up all night benchmarking.
We choose a full suite of benchmarks in order to put the GeForce through its paces. You'll recognize the games/synthetic benchmarks used and should you feel the need to reproduce any of the ensuing scores (you can if you really, really want to but you don't have to or anything) then we've kept it fairly clean and simple, so you can. For our OpenGL tests we used Quake 2 version 3.20 and ran demo1.dm2 as well as Brett Jacob's Crusher.dm2 timedemo. For Quake III (version 1.08) we used the q3demo1 timedemo and choose to test the 'NORMAL' and 'HIGH QUALITY' settings. We also used our own 'SEHQ' setting, which is merely the 'HIGH QUALITY' setting cranked up to 1024x768. For our D3D tests we ran the Expendable Timedemo but with the DDR drivers optimized for DX7 (not DX6), we think the D3D scores will be going up significantly. All tests were run until a consistent score was obtained and with V-Sync turned 'off'.