3dbench was like the 3DMARK back in the days 🤣
CPU, Videocard, ISA clock speed, Wait states and other settings can all affect the score. The CPU does play a huge part, but if you put a ISA card into a Pentium you will still get a poor score!
Aww, man I think I just recently threw out my Speedstar 24x that I used back in... 1994? If only I had found this site, sooner. I'll look around and see if I really did throw it out or not because it looks like people here may want it. 😜
Some may be a bit confused as to what 3D is on a PC. Quake for example is always rendered in 3D, but a "2D" card renders it in pixels, which don't make curves well. A "3D" card, regardless of doing the work itself rather than the CPU, renders in texels, which look prettier because you can make rounder shapes.
wrote:Some may be a bit confused as to what 3D is on a PC. Quake for example is always rendered in 3D, but a "2D" card renders it in pixels, which don't make curves well. A "3D" card, regardless of doing the work itself rather than the CPU, renders in texels, which look prettier because you can make rounder shapes.
I beg your pardon? Rounder shapes could be achieved with tessellation. Texels refer to texture mapping which Quake already does in software mode. Or are you talking about bilinear filtering?
^^^ I thought 3D cards inherently used a triangular unit for rendering textures rather than square pixels.
I wonder if replacing the HYB514256BJ-70'son the 9000C with -50's would give it more improvement.
The polygons drawn are based on triangles (well, except nVidia NV1), but the pixels are still square. All still true for software Quake.
About replacing the RAM: If the card is really limited by RAM latency/bandwidth then you could in theory get better performance with faster RAM. You will have to set higher clocks manually though, the card won't clock itself higher automatically.
But would you really bother tuning a Trident video card for more speed? If the RAM chips are socketed it shouldn't be too hard, but still...
Are you sure it needs to be reclocked? My cousin had one of those 90's diamond cards with the two add-on memory sockets, the typical chips being 60ns and 70ns. But he went through the trouble to find 45's and it was faster, but ran too hot. At least that's the way I remember it.
Then you remember it wrong.
The memory speed is determined by the graphics card itself, and the card doesn't know or care about the speed rating of the chips. They're either fast enough to keep up with the speed the card runs them at, or they aren't. If you want the card to run the memory faster, then you have to find some way to tell it to run the memory faster.
Now, you may not be mistaken when you remember there being a performance increase when the memory was upgraded on that Diamond card... but it had nothing to do with the speed of the chips used. Back in those days, many cards were set up such that the amount of memory was also tied to the memory bus width. For example, the Cirrus 5434 and Tseng Et4000W32 (among many others) supported a 64-bit memory bus when equipped with 2MB... but they usually only came with 1MB, which also halved the memory bus to 32-bit.
I understand what you're saying, however back in the SIMM days, I used to run sticks through a tester, which would do step-up/step-down tests, spike tests, etc. And memory never ran at its labeled speed for long--it heated up enough, internally I suppose, that it would go from 100ns to 120, sometimes as high as 140's. This leads me to believe the chips in the 40's may have in fact helped, however I don't think my cousin soldered a pair in place of the factory chips so I would have imagined it bottlenecked.
