BAM!!!

http://www.pcguide.com/ref/mbsys/cache/charCa … eability-c.html

It depends on the chipset, but many 486 boards with 256K cache will cover up to
- 32 MB if the L2 cache is in write-back mode.
- 64 MB if the L2 cache is in write-through mode.

You are looking for this (:

BastlerMike wrote:

Some 386 and 486 board manuals mention the cacheable range. From what I have read so far, i would say there is a rule: […]
Show full quote

Some 386 and 486 board manuals mention the cacheable range.
From what I have read so far, i would say there is a rule:

32kb 8MB
64kb 16MB
128kB 32MB
256kb 64MB
512kb 128MB
1024kB 256MB

These values are for write-through operation. In write-back mode the cacheable range is reduced by half.
I don't exclude additional chipset related restrictions though...

On my M919, CTCM7 reports:
256 KB WB Cache = 32 MB cacheable
256 KB WT Cache = 128 MB cacheable

On my Biostar MB8433UUD-A v3.0 CTCM7 reports,
256 KB WB Cache = 32 MB cacheable
512 KB WB Cache = 64 MB cacheable

Has anyone else seen differently?

I second rg100's question about the Windows performance. Anyone want to comment on using Windows98SE/NT4 with,

A) 32 MB RAM only and 256 KB WB cache, where all RAM cacheable
or
B) 128 MB RAM and 256 KB WB cache, where only 32 MB RAM cacheable

Who's performance suffers the most, A or B? A will need frequent access to a swap file while B will not.

EDIT:
The webline that bob posted mentions that a 5-25% performance decrease will result from adding more RAM than is cacheable, however, it also says this,
"If you are doing heavy multitasking and notice that the system is thrashing, you will always be better off to have more memory, even uncached, instead of having the system swap a great deal to disk. Of course having all the memory cached is still preferred."

So it looks the retro guys here will have to test this out ourselves!

Thanks very much for the super fast response guys!

feipoa, I wonder if I've spotted a difference between the Biostar MB8433UUD-A v3.0 and v3.1 versions? On my v3.1 model, the PCB has the cache size options printed on it, and it says that you can select either 128Kb or 256Kb. There's no mention of a 512Kb setting; not that I can see anyway. That's a bit disappointing. I wonder if the v3.1 version is a cost reduced version of the v3.0 board?

Edit: I installed Win98SE on my Biostar MB8433UUD-A v3.1 mobo, with 128MB of EDO RAM installed, and it took a very long time. I then installed WinBench99, and that installation seemed to take forever. The sluggishness was very noticeable. I have now reduced the RAM amount to 32MB, and the sluggishness has more or less gone.

That's odd. I also have a 3.1 motherboard that I used for years until the southbridge died. It does have the 512KB option printed on the MB and I have used it with 512KB of cache for years as well.

For 8 pieces of 256 KB cache
JP5 2-3
JP6 1-2
JP7 open

For 4 pieces of 512 KB cache (that's the only supported configuration)
JP5 1-2
JP6 1-2, 3-4
JP7 1-2, 3-4

I have not noticed sluggish performance with 3.1 or 3.0 Biostars with 128 or 256 MB of FPM RAM, or even EDO RAM. But again, I've never run it with less than 128 MB. Could I be so used to sluggish that I don't know better? I doubt it. In my M919, I've set WB and WT for L2 such that all 128 MB of RAM is cacheable and I did not notice any change in performance, nor in benchmark scores.

There is a slight possibility that your TAG RAM is bad or incorrect, same goes for your cache. Double check the jumper settings as well. When you say sluggish, in what way? How much longer to open an application? You might want to time opening Sandra 2002 or something similiar in size. When cache goes bad, I've noticed that changing to the slowest cache timings makes it work again, but not on the fastest settings.

I highly recommend that you run MemTest 4.0+ for a good 3-4 hours.

Pilot error. 😊 I shone a torch on the mobo, and I can see the 512Kb setting. This info is several inches away from the other cache size information. Well, that's my excuse anyway! 😉 Sorry about that.

retro games 100 wrote:

Pilot error. 😊 I shone a torch on the mobo, and I can see the 512Kb setting. This info is several inches away from the other cache size information. Well, that's my excuse anyway! 😉 Sorry about that.

It's not like they always printed all jumper settings on a logical easy-to-find spot 😜

After a quick test with 32, 64, 96, 128 MB sticks of EDO RAM, I have determined that CTCM7 sometimes is incorrect. In L2 WT mode, it showed that all RAM was cacheable up to 128MB (32 MB for WB mode) for 256 KB of cache.

The only Windows-based benchmark programs from my testing list that show change with cached and uncached RAM are CpuMark99, SuperPi, PassMark - MathMark score, and WinTune98 - Memory test.

CpuMark99 Stand-alone showed the most significant benchmark change. A summary of test results in Windows98SE w/Cyrix 5x86-120, L2: 2-1-2, RAM: 0/0 RAM WS, and 256 KB L2 cache are shown below using CpuMark99 Stand-alone v1.0.

--L2 cache: WB/WT
32 MB RAM: 5.4/5.4
64 MB RAM: 4.2/5.5
96 MB RAM: 5.6/ 5.5 <--------- This one doesn't follow the trend
128 MB RAM: 4.2/4.2

The results showed the same pattern when tested in Windows NT4.0. It infact seems like 256KB L2 will cache up to 32MB in WB mode and 64MB in WT mode.

Under the conditions of 32 MB EDO (all cached) and 128 MB EDO (only 32 MB cached), I opened IE6, logged into gmail, and opened up this tread. I did not notice any significant speed difference between the two cases. I hope this clears some things up.