Mumak wrote on 2022-02-26, 08:18:
Thanks, this is interesting. It seems that the crash happening when Method 1 or 2 is enabled happens during normal floating-point math. So I assume when HWiNFO performs a soft-reset, some BIOS bug workarounds don't get applied or the FPU isn't re-initialized properly.
FPU clock is determined by measuring real FPU instruction throughput/latency. So are you sure it's running at 33 MHz and the 2x PLL works? Perhaps you can try to confirm using other FPU test/bench tools.
Although I have not probed the CPU and FPU on the Evergreen unit, I would be surprised if Evergreen only enabled clock doubling for the ALU and not the FPU. I ran some benchmarks using various configurations with Landmark v2. First is with a Harris CS80C286-16 and a Cyrix FasMath Cx-82S87-NP-SV co-processor.
And compared it with the Evergreen SLC unit.
Harris CS80C286-16 and a Cyrix FasMath Cx-82S87-NP-SV co-processor, both at 16.7 MHz
Landmark FPU = 24.03
Cyrix Cx486SLC2-66 and a Cyrix Cx87SLC-33, at their respective speeds
Landmark FPU = 45.35
The FPU results are about double with double the FPU frequency.
One may wonder... how much gain is the SLC at 2x providing to the FPU? Before I had the SLC2 soldered on, I had an ordinary SLC soldered on. With the SLC at 33 MHz, the Landmark FPU score was 41.20. So one may deduce that the SLC at 66 vs. 33 MHz, aids the Cx87SLC FPU by about 10%. You may also wonder, how much benefit does the SLC's L1 provide to the FPU? I ran Landmark with the SLC2-66's L1 disabled, and the FPU result was 39.5 (-10%). Adding these conditions in attempt to match that of the cacheless 286, we'd subtract 10% from 41.2 to approximate the SLC-33 without L1 and another 10% to approximate a SLC-16.7 (but FPU still at 33.3 MHz) and arrive at a simulated Landmark FPU score of 33.37. 33.37 isn't exactly double that of 24 from the 286, so it is hard to draw a conclusion. It may be that the SLC was really designed to run without L1 cache, and one cannot say there is a 1:1 relationship between the Harris 286 and a Cyrix SLC at the same frequency.
Unfortunately, there doesn't appear to be any documented DIP switch settings for this Evergreen unit. I suppose I could start flipping switches to see if any of them increase the FPU score, which might indicate that FPU clock doubling is disabled. I also tried to locate the datasheet for the Cx87SLC but couldn't locate it.