interpreting cachechk output

[cmc]

Hi Folks

In some of my other posts I describe a 486dx2 box I built. I finally have it up using various scrounged parts and an IDE/compact flash converter. Even managed to install FreeDOS after some trouble with bad sectors on old floppies.

Anyway I've been worried for a while about the memory speed, and whether everything is cached properly. I ran CACHECHK v5. The output is attached.

I'm not sure how to interpret it. So I have 5 sticks of RAM. 4x1mb 30-pin sticks in bank 0, and 1x32mb 72-pin stick in bank 1. The board is a DTK PKM-0038s, similar but not exactly the same as this one:
http://stason.org/TULARC/pc/motherboards/D/DTK-COMPUTER-INC-486-PKM-0038S-VER-3.html

From what I can tell, the 1mb stick seem to be about twice as fast as the 32mb stick. Is this accurate? If so, what could be the cause? I *think* all the RAM is cached, as the small memory chunk reads are 16us... BUT, then increasing to 83us for larger chunk sizes. For the top 4mb (which I guess is the 4x1 sticks), the reads are only 41us. Then of course you see the "REALLY SLOW" warnings in the log file.

Perhaps the 32mb stick is just slow? It is an ebay item for $5, so how much can I expect anyway:
http://www.ebay.com/itm/350478815084

 

[cmc]

Forgot to mention that the cache size is 128k. I would like to upgrade it, but cache is *still* expensive!

 

[Uniballer]

I have found CTCM.EXE version 7 to be a better cache test than CACHECHK.EXE for my purposes. The report you provided looks like the L2 cache miss penalty is huge, but that only the first 32MB are getting cached by the level 2 cache (see how it goes straight from 16 [speed of L1 cache] to 41 [speed of memory above 32MB] with no 30 in between). The cache misses above 32MB are faster (41) than below 32MB (83). Weird.

The RAM you bought on ebay sounds like it should be as fast as your board can go. What chips are on the SIMM? What does your BIOS say about the RAM timing (wait states, etc)? I might try eliminating the 30-pin SIMMs and see if anything changed.

 

[cmc]

Thanks, I'll try out CTCM.EXE. I'll post the results plus some close-up pictures of the chips once I get home.

I really can't explain why the first 32mb chip would be so much slower. But it could be that the board only caches up to 32mb, or that 128k is not enough cache, or I suppose even that the board will only cache on certain memory banks depending on the configuration. Four 30-pin SIMMs are in one bank, and then there are two 72-pin slots, each one of those slots being one bank. So bank 0 is 4x30-pin slots, bank 1 is 1x72-pin slot, and bank 2 is 1x72-pin slot.

One odd thing is that the board seems to *require* that bank 0 (the 30 pin slots) be full before it can use bank 1. If I take out the 4x1mb 30-pin SIMMs, the computer refuses to boot, and gives the classic memory beep-of-doom. What I can try though it to remove the 32mb 72-pin and see if it caches the 30-pin SIMMs.

 

[cmc]

More odd behavior. I ran CACHECHK.EXE with the 32mb stick removed, leaving only 4x1mb 30-pin SIMMs. See the attached file. It is cached, but then it is 83us on 256k and above chunks.

Then put the 32mb stick back in and ran CTCM.EXE. Report attached. Not sure what to make of that output.

Unfortunately the BIOS does not let me pick memory timings. At least I could not find an option for it.

So it appears that the board is slowing down any RAM that it caches?

 

[cmc]

A couple more reports. This time with L2 cache disabled in the BIOS. It appears to actually be faster when the cache is disabled!

 

[cmc]

Here is what SPEEDSYS says about it, cache enabled and disabled. Benchmarks seem to confirm that overall it is slower with cache enabled.

But I need to try it out in practice, to see if it is really slower.

 

[cmc]

Sorry to spam the boards... two last pictures: