Anonymous Coward wrote on 2024-04-05, 15:04:
The system running about the same speed as a 486DLC kind of makes sense, as it basically is a 486DLC with an extra 1kb internal cache. They were definitely a notch below proper 486s, which is why this product was not very successful.
You are right. I dug through all my documentation and datasheets and it appears that the 486S has the same core as the 486DLC with a different bus and cache. The later Cyrix 486DX is an improved design that is closer, performance wise, to Intel and AMD 486s.
However, I was also right that the internal cache was working in write-through mode. I found a small utility that lets you switch the internal cache of all Cyrix 486s between WT and WB mode: https://github.com/karcherm/cx486wb
By default, after boot up, the processor cache is set in WT mode. Switching it to WB yields a modest 5% increase in real world applications (I tried with Wolf3D and Doom).
Disruptor wrote on 2024-04-05, 16:00:
Can you post the graphical chart from SpeedSys 4.78 please?
Unfortunately I put the motherboard back in storage after removing the CPU+FPU from the socket. I want to mount it in a faster motherboard because this one had both memory and bus bottlenecks. In some tests, I was measuring lower performance than what a 486DLC scored on my Panda 386V VLB motherboard. In a couple of week, I should be able to go back to this project and I'll post screencaps from various tests.