VOGONS


First post, by konc

User metadata
Rank l33t
Rank
l33t

Hello,
Assume a 386 with 128KB L1 cache on the motherboard, the usual stuff. If one would like to test stability vs SRAM timing (find out the tightest cache timing the chips+motherboard are stable with), is memtest86+ an appropriate test to run?
Of course setting the timing way lower than what's possible results in obvious problems (no boot, frequent crashes etc) but for marginal situations where the system boots and problems occur only after extensive use (=a lot of time and "luck" needed to encounter them) would memtest86+ catch them or does it test only the RAM? If this is the case what would you recommend to stress test the cache?

Reply 1 of 9, by pentiumspeed

User metadata
Rank l33t
Rank
l33t

Use the cache utility that test performance.

memtest86+ only tests the main memory only and very sensitive and very useful for hard to catch errors. That other software wouldn't even windows tests wouldn't find.

Cheers,

Great Northern aka Canada.

Reply 2 of 9, by mpe

User metadata
Rank Oldbie
Rank
Oldbie

The memtest86 implicitly exercises caches to some extent. However, most of its tests are deliberately written in a way to avoid reads/writes being cached/bufferred as it obviously defeats the purpose of the tool to test he RAM.

I usually use Doom/Quake in DOS as a simple test.

However even that doesn't give full guarantee. Better to boot into Windows and run something like Winstone.

Blog|NexGen 586|S4

Reply 3 of 9, by cyclone3d

User metadata
Rank l33t++
Rank
l33t++

Memtest86+ is ok-ish for older stuff but it takes absolutely forever to find problems on newer systems that don't have problems pop up all the time.

For that matter, if you have intermittent problems on an older system, it can take days of memtest86+ running to finally find a problem.

HCI memtest is way, way, way better from my experience. But it is for newer systems. It finds slightly flakey memory way, way, way, way, way, way faster than memtest86+ does.. if memtest86+ even finds problems at all.

I'm taking 30 minutes to an hour max with HCI memtest vs memtest86+ running for 24+ hours.

Now if you have RAM or some other issue that is causing errors all the time, memtest86+ will find those problems fairly quickly.

HCI memtest puts the system under a much higher load so the problems show up way more reliably and way quicker.

Edit:
The free version will work as far back as Windows 9x

If you buy the Deluxe Pro version, you also get a CD/USB bootable version.

It is a 32-bit program so no testing really old stuff though as far as I am aware.

https://hcidesign.com/memtest/
/end edit

Yamaha modified setupds and drivers
Yamaha XG repository
YMF7x4 Guide
Aopen AW744L II SB-LINK

Reply 4 of 9, by douglar

User metadata
Rank Oldbie
Rank
Oldbie

So basically you are suggesting that the best way to test a mainboard cache is to run demanding software that uses an active memory foot print that’s a little bigger but not too much bigger than the main board cache. Is that the idea?

Reply 6 of 9, by konc

User metadata
Rank l33t
Rank
l33t

Thank you guys for all the responses. This was also my impression

mpe wrote on 2020-02-03, 20:41:

The memtest86 implicitly exercises caches to some extent.

but we all agree that since it's not made for testing specifically the cache (rather the opposite), better use something else.

Unfortunately it's a 386 so no quake or windows 9x (dos & win3x only) to run all the suggested tests. I'm considering temporarily using another hard disk to do a minimal install of win95 and try them though. I won't run a nuclear power plant with it but it's built to stay on the desk and get used so I'm trying to determine once and for all the best values where it's 100% stable and never touch it again until something breaks.

Reply 7 of 9, by Tiido

User metadata
Rank l33t
Rank
l33t

Testing cache specifically requires chipset programming that BIOSes use to determine how much cache is there etc.
Memtest86 has no such functionality and I'm not aware of any software that can do it.

T-04YBSC, a new YMF71x based sound card & Official VOGONS thread about it
Newly made 4MB 60ns 30pin SIMMs ~
mida sa loed ? nagunii aru ei saa 😜

Reply 8 of 9, by mpe

User metadata
Rank Oldbie
Rank
Oldbie
Tiido wrote on 2020-02-04, 11:49:

Testing cache specifically requires chipset programming that BIOSes use to determine how much cache is there etc.
Memtest86 has no such functionality and I'm not aware of any software that can do it.

I don't think you need to use any chipset programming to do that. Although that would be definitely far more accurate.

Memtest86 actually has routines to detect caches, measure speed, sizes, etc. As mentioned above the main reason is that it is somewhat futile attempt of trying to avoid test data being cached as well as report it.

See - https://github.com/Distrotech/memtest86/blob/ … t86-5.01/init.c

Blog|NexGen 586|S4

Reply 9 of 9, by Tiido

User metadata
Rank l33t
Rank
l33t

Most/all chipsets when off CPU caches were a thing actually have methods to force access to go exclusively into the cache, without touching RAM, essentially turning it into the RAM itself. Other methods assume that whatever you do in one spot remains in caches but you are still at the mercy of whowever the device does with cache and RAM interactions. Cache in at least some of the modern CPUs should be possible to put into a mode where it becomes directly accessable RAM, it certainly is possible in many non x86 architectures.

T-04YBSC, a new YMF71x based sound card & Official VOGONS thread about it
Newly made 4MB 60ns 30pin SIMMs ~
mida sa loed ? nagunii aru ei saa 😜