Bring the timing down as much as possible and try again, you'll be surprised by the difference it makes 😉 If your memory can sustain the lowest values with 0 WS you'll see a sysinfo result of at least 41-42. Regarding the cache, yes that's 128K. Have you checked for cache size configuration jumpers? I don't know if it has, just throwing in an idea.

i swapped in two sets of 128kb -15ns cache(each set made up of 4*32kb and a 8kb tag ram of course), but the board only reported "cache ram bad, cache disabled". this is unlikely for both sets of cache to be bad, as they were removed from other boards.
so these are my guessing:
1. the original cache that came with this board are remarked, they are actually four 8kb chips rather than 32kb.
2. the bios is either bugged or hacked to display 128kb when 32kb is actually plugged, and would not work with real 128kb.

konc wrote:

Bring the timing down as much as possible and try again, you'll be surprised by the difference it makes 😉 If your memory can sustain the lowest values with 0 WS you'll see a sysinfo result of at least 41-42. Regarding the cache, yes that's 128K. Have you checked for cache size configuration jumpers? I don't know if it has, just throwing in an idea.

i looked for cache size jumpers but there were none, there is a small chart printed on board telling you to plug four 8kb and a 8kb tag to form 32kb, and four 32kb and a 8kb tag to form 128kb, but never mentioned any jumpers.

bad cpu jumper setting can also fail cache to work

or using a cpu not supported with other cpu jumper setting makes cache fail

progress: i replaced the bios chip with the one from the corroded board(since they have same chipset and even identical layout), and it reported 32kb cache at post!
so my two guessings are right:
1. the original cache that came with this board are remarked from 8kb per chip to 32kb per chip.
2. the original bios was hacked to report 128kb cache when 32kb is actually plugged.

however the replaced bios still reports "cache ram bad, cache disabled" when i plugged 4*32kb cache chips, so i have to believe that they are indeed bad and use 32kb cache for a few days before my purchase arrives.
since 32kb can only cache for 8mb of ram, i set cacheable size to 8mb in bios and the tests all finish without crashing(they crashed a lot when i set cacheable size to 16mb before). the cachechk results are:
cache: 47us/kb(21.3mb/s)
cached ram: 115us/kb(8.7mb/s)
uncached ram(or cache disabled): 67us/kb(15mb/s)
also changing dram wait stats doesn't seem to affect cached ram speed. they affect uncached ram speed a bit but 1wait is the fastest, while 0wait and 2wait are both a bit slower at 74us/kb.
is this kind of performance normal? can anyone give your 386dx40 cachechk results for comparision?

sure, here are my results:
cache: 37.8mb/s
cached ram: 8.7mb/s
uncached ram(or cache disabled): 14.9mb/s

What are your lowest possible/reliable timing values and norton's sysinfo result with them?

konc wrote:

sure, here are my results: cache: 37.8mb/s cached ram: 8.7mb/s uncached ram(or cache disabled): 14.9mb/s […]
Show full quote

sure, here are my results:
cache: 37.8mb/s
cached ram: 8.7mb/s
uncached ram(or cache disabled): 14.9mb/s

What are your lowest possible/reliable timing values and norton's sysinfo result with them?

well, looks my ram performance is inline with yours, except that my cache is running at slowest 3-2-2-2 timing as the sram chips are marked -20ns(they are faked anyway). i'll see if i can get away with 2-1-1-1 when my purchase of -15ns cache and -12ns tag arrives.

by the way, is there any 386 chipsets without cached dram penalty?
that is, in common sense, the chipset would only access dram after a cache miss, therefore the total time of a cache-miss read would be cachetime+dramtime.
however, a better designed chipset shall initiate a cache read and a dram read at a same time, and discard the dram access once cache hits, thereby eliminating cached dram penalty.

bad news: the purchased sram chips arrived, but the damn board still reports "cache ram bad, cannot enable cache" at post. i can't believe that several sets of sram chips are all bad(even if just 1 out of 4 chips is bad), so i assume that something is wrong with this board that it wouldn't work with 128kb cache.
and there is no cache-related jumper on the board, i tried two unidentified jumpers but they don't do any good.

I have that board. Since i got it without cache, added some chips last year, and it wasn't without problems - got many "cache ram bad" messages, in the end cleaning (scraping) each IC's leg with a razor cleared the errors, but i didn't think to do some benchmarks before and after. I don't think i will remove the chips again (legs got flimsy after so much pushing and pulling), but in the next few days i can run some tests/benchmarks if you are interested.

quicknick wrote:

I have that board. Since i got it without cache, added some chips last year, and it wasn't without problems - got many "cache ram bad" messages, in the end cleaning (scraping) each IC's leg with a razor cleared the errors, but i didn't think to do some benchmarks before and after. I don't think i will remove the chips again (legs got flimsy after so much pushing and pulling), but in the next few days i can run some tests/benchmarks if you are interested.

oh, did you add 32kb or 128kb of cache? the board never reported any problem when i plugged the original 32kb cache back, and the purchased caches' legs are in excellent condition, so i doubt if the problem is caused by legs contact.

I have installed 5 pieces of UM61256FK-15 (tag ram identical to the other 4 chips), so i guess that's 128KB cache. It will be a few days until i can run some tests on that board.

Here are my results (speedsys and cachechk with cache enabled and then disabled, landmark only with cache enabled).
Tightest cache timings, main memory 1WS.

https://imgur.com/a/IRovBjv