feipoa wrote on 2022-08-17, 18:51:

This mod is still on my list. The ICs have been sitting in a bin; I'd have to take the motherboard out of the case though. Did you just replace those 8 chips, or was there more to the mod? Did you use solder paste, or solder the pads one-by-one with a contact-based soldering iron?

When you say that 1024K is better than 256K, did you run any numbers to demonstrate the comparison?

Are you running the PF110 CPU? My system has the PF110, so I was concerned that perhaps the higher FSB might cause issues with 1024K vs. 256K using the same memory timings. This is an issue I've witnessed on socket 3 boards.

Just two Weller soldering irons to remove the chips and one to put the replacements. Although I could do it with a heat gun, too.

The board has the necessary layout for 32-pin async SRAMs with evolutional pin-out. CE2_H pulled high to Vcc, new Vcc is pin 32, A15 and A16 routed properly. The BIOS takes care of cache size detection.

It's an overclocked P80. I don't have a PF Nx586 yet, though willing to pay a fair price for it. Also I didn't record specific benchmark stats, but the improvement was noticeable. 70ns FP memory isn't really fast, so large cache helps. Nx586 runs it on a separate bus, half the CPU clock, with an internal tag. It's not shared with the system bus. I think 10ns cache chips are an overkill in my case. 12ns or even 15ns should do just fine.

That's before the mod. Speedsys confirms there are 192Kb of cache accessible out of 256Kb.

Tiido wrote on 2022-08-17, 19:00:

This is in my todo list also, chips are ready but there is no time to deal with it...

Which CPU are you running in yours? I had been hoping that someone else would do the 1024K cache modification on something with a PF110 and preferably with a fully setup system for testing in Windows.

It has a P90 in it, so nothing special. The better CPUs cost far more than what I paid for the entire machine so I am unlikely to ever see an ugrade to it 🤣