I find the DRAM timings make the biggest difference in most of my systems. In the ideal world if you could run the system bus at 50MHz, disable the L2 cache and crank the memory timings as low as possible you could have crazy fast performance. It might be possible if someone were able to fabricate SIMMs using SRAM or super fast DRAM.

Anonymous Coward wrote:

I find the DRAM timings make the biggest difference in most of my systems. In the ideal world if you could run the system bus at 50MHz, disable the L2 cache and crank the memory timings as low as possible you could have crazy fast performance. It might be possible if someone were able to fabricate SIMMs using SRAM or super fast DRAM.

why disable cache? even cache at 3-2-2-2 is still faster than 2-1-1-1 dram. and i don't think you can get -60 dram running at 2-1-1-1 timing at 50fsb. 3-1-1-1 may be possible with 50ns dram but they are hard to find, even for simms made in the 430tx and mvp3 era.
also, fast timing at high clock is not only limited by dram quality, but also by the chipset itself. the chipset may not be able to perform the cycle that fast, just like dram.
sram and dram run in different ways, so i don't think the chipset can handle it if you make a simm with sram chips.

clueless1 wrote:

Really interesting is comparing the memory/cache timing graphs in the two Speedsys screenshots.

VLB System with a typical 3-1-1-1 0/2 WS config:

A PCI system:

WT is faster with a 66 MHz 486er than WB with most games.

In your VLB system the memory bandwidth is so much higher than mine, but the results in the lower right graph are nearly identical. Maybe 72-pin vs my 30-pin memory?

My VLB system uses 30 pin SIMMs. The board is this: http://retronn.de/imports/hwgal/hw_mainboard_ … um486v_aio.html

72 pin SIMMs provide no performance benefit over 30 pin, unless they are EDO (which 486s dont' support anyway).

Thanks for the clarification AC. I'm curious what accounts for that huge bandwidth difference then. Other than memory bandwidth, the other results are very similar between mine and elianda's 486 VLB.

there was a guy who did his 486 with sram as main memory, no?
i wonder if his results are around somewhere