(20 Jan 97 This has been updated for PPro at 233mhz - and now for glide 2.1 installation info - with correct FXMEMMAP.VXD and fastvid settings my scores have shot up (redone again! see bottom of text) DMAN's 3DFX benchamarks using a Pentium Pro 200 (at 200mhz) and D3DTEST.EXE Re the test results I posted previously Gary Tarolli of 3DFX mentioned that I should have had NO UPDATES selected in the test so I redid them and here they are .... > GARY Tarolli said: > Everyone please remember to use the "No updates" option in D3test - that > way the test will be uneffected by waiting for VSYNC and refresh rate. > This is a LOT easier than trying many different refresh rates, assuming > you can read the results on the screen. > > It seems like people are getting around 400K triangles on a P-166, > that's very encouraging! The fill rate is much improved as is intersection. (I still did it at different refresh rates just for comparison) In the tables below the number is the refresh rate, F = flat shading selected. G=Gouraud shading selected. G+B=Gouraud plus Bilinear Filtering selected. The D3D test options set were as before: Z buffer on Pers. Correct on Textures on No Updates on Shading model see below. Overdraw Order: front to back Rasterisation Mono Texture Filter: Point (plus some with Bilinear filtereing as stated). 60F 60G 60G+B ----- ----- ------- Fillrate : 32.00 32.56 32.32 Poly throughput: 444.91 433.01 433.01 Intersection: 5.23 5.25 5.25 75F 75G 75G+B ----- ----- ------- Fillrate : 31.04 31.60 31.44 Poly throughput: 437.47 430.85 419.62 Intersection: 5.08 5.10 5.09 85F 85G 85G+B ----- ----- ------- Fillrate : 30.32 30.89 30.80 Poly throughput: 446.40 433.01 433.03 Intersection: 4.97 5.00 4.99 (note PowerVR users should try renaming the D3DTEST.EXE to something else as the drivers for the PowerVR are rumoured to look for the D3DTEST.EXE and then 'cheat' to produce higher figures than normal using special techniques which are not usually used for games as these techniques produces picture tear. Renaming the EXE means it can't cheat!) As I have just re-clocked my Pentium Pro to 233mhz here are some quicky tests at 60hz only (no time!) 60F 60G 60G+B ----- ----- ------- Fillrate : 32.72 32.72 32.48 Poly throughput: 475.27 475.00 476.16 Intersection: 5.25 5.27 5.29 Also after installing the new Glide Dev Kit (with Glide 2.1 and updated .vxd's) my benchmark dropped dramatically. However I learned that the new drivers enabled USWC caching on PPro's (the job of fastvid normally). After much messing around I discovered I had to run fastvid twice in my startup. e.g. fastvid 111 2 f8000000 (2D card Stealth 3D 2000) fastvid 111 4 fe000000 (3dfx) this enables USWC for both 2D and 3D cards. 7 Jan 97 ========= The benchmarks read... Now 60F 60G 60G+B Fillrate : 32.00 32.56 32.32 Poly throughput: 494.02 476.75 476.16 Intersection: 5.23 5.26 5.25 75F 75G 75G+B Fillrate : 31.04 31.60 31.44 Poly throughput: 491.56 477.65 493.04 Intersection: 5.07 5.10 5.10 Getting nearer the 500 now. Other people have got 538kpps with even with Gouraud and Bilinear on. 16/1/97 update: =============== Latest with correct FXMEMMAP.VXD on PPro200 at 233mhz: he downloaded one just to make sure. Now (at 60hz) on a PPro at 233mhz I get: Zbuffer, Persp.Correct, Textures, No Updates, Point filter all ON. With the correct FXMEMMAP.VXD (6464 bytes) with fastvid NOT used for the 3D card (2D card only) I now get: Flat shading: Fill Rate: 32.08 mpps Polygon Throu. 614.27 kpps Intersection: 5.26 kpps Gouraud: Fill Rate: 32.80 mpps Polygon Throu. 574.00 kpps Intersection: 5.29 kpps Gouraud & Bilinear: Fill Rate: 32.48 mpps Polygon Throu. 573.28 kpps Intersection: 5.28 kpps Wow !