CompuServe Thread

I also have a DX266 machine EISA with 1 meg of cache ram. It is faster by about 20% on rendering than the ISA machines. I feel compaiing CPU's in different bus designs is not a valid comparison of cpu benchmark. My 12 slot ISA bus has only 128K cache but caches only the first 16 meg. even though I can put 32 meg on board. The VLB bus has 256K cache and caches the first 64 meg. When I ran the benchmark the rendering was done in the first 16 meg with no paging so the test was valid between the two machines but not with the EISA machine. I also had a 50 Mhz. CPU with 16 meg. 256 K cache for a short time and it was running 16-20% slower than an equivalent DX2-66 with everything else constant. I returned the 50 Mhz. MB for another DX2-66 so I can't do anymore testing. >>A RAM cache upgrade is only $9.00, so I will try that and find out in about two weeks. In the meantime, does anyone have any more information about this issue? Why isn't it documented in the 3DS installation manual? That's a pretty phenomenal price to upgrade cache ram. It cost me $200 to go from 256K to 1 meg on my EISA board. I think it would be out of place for the installation manual to quote pricing on cache ram upgrades as a basis for installation. It's so market dependent. One of my machines has the bios capability to shut down the internal cache and the external cache separately. When I tried this with 3DS renderings. The most significant speed hit was with the internal cache. A DX-2 66 operates the internal cache and floating point operations at the higher speed for a greater portion of the rendering period internally than paging out to the external cache and external ram making the speed difference between the 50 Mhz bus and the 33 Mhz. bus far less significant than the speed difference between the 50 Mhz. CPU/cache/mathco. vs the 66 Mhz CPU/cache/mathco. That's the way the engineer at INTEL explained it to me respecting floating point intensive 3DS operations. On the other hand, Screen redraws are more a function of a VLB speed at 50 Mhz. vs 33 Mhz. and some boards will perform faster at 50, IF, the VLB card will function at the 50 Mhz. VLB speed. He said the problem was that VLB spec was not designed for 50 Mhz. bus and cards that work are few and buggy. This info is about 9 months old now so recent dev. might have changed. Re Topas: I hav'nt done too much recently with Topas. The interface is difficult for me and my video VESA driver needs upgrading since I can only operate in 8 bit mode or targa+ for rendering. There comment about the >16 meg of ram is true but a second bank of non-cached 16 meg will perform better than only 16 meg of cache ram with the additional going to the swap drive. In that respect their statement is misleading.

When it comes to rendering the CPU does all the work and 66 is faster than 50, no question about it. I guess if you're looking at bus speed performance the 50 is faster, but who cares when you have to render? I have a machine with 478K base memory, and it runs about 2-3 seconds slower than another machine with 636K base memory on a five minute frame.