I don't care about particular models, it just have to properly represent the chipset. Lynx3D (some device for Windows) Mpact 3600 with D-Sub Oak Warp 5 OKI TrianGL Permedia AGP SiS 6326 revision D Starfighter PCI 24 MB, and any Intel740 AGP with less then 8MB Yamaha RPA (YGV611 / YGV612) ------------ from later cards i752, Neon 250, Xabre 80 and 600, any Volari or S3 Chrome
I am just a nostalgic enthusiast, assessing knowledge from the internet. Forums at Beyond3d and Vogons were most helpful. I had my own little web about graphics in czech since 2004, but stopped after some years. Now I feel I can write something again, this time in English. Not a native speaker, excuse my mistakes.
Yes I can, while knowing it is futile. When you ask most guys, even the educated ones, to choose faster cards of two they will usually answer quickly. We all have our opinions despite the complexity behind the hardware. I merely expanded the question to all desktop gaming 3D cards. There are many mistakes, generalizations, it is never-ending work and you can just ignore it.
No, in fact I disregard synthetic benchmarks. I spent lot of time reading many reviews. Because early 3d accelerators were not properly reviewed I am trying to examine them now on this very web. Serious business.
FLOPS most probably are. I don't go with designers math if real operations are too far from theoretical. Some custom math is in place for certain architectures. When pixel shaders started they were not scalar like now. One ancient pixel shader could perform one instruction on all components. Now we have scalar units working on a single component. I needed something consistent for my table and decided to translate scalar numbers into old pixel shader counts. Sucks I know.
Definitions of the pixel pipeline vary. I use the one saying one pixel pipeline should output one pixel per clock. In 3d games it mostly means textured, shaded, whatever pixel. I know pixel pipeline counting is outdated, ROPs are often a limiter. Time will tell if I will come up with better formula.
I use only generic, common models. Particular specification deviations from single vendor are not included, such entries would overpopulate the table and IMO would ruin it. This rule of course is not applied for some old unique SKUs. If I missed any relevant card, you are most welcome to inform me.