Elite Group (ECS) si54p aio

[NickS]

That all looks reasonable at first sight. Try the memory/cache with SiSoftware Sandra and see what that reveals.

[schvankus]

hmmmmm, seems Sandra Professional 2003 requires Windows 98 and up. I've got 95B so it wouldn't install. I have a spare drive with 98 on it that I can throw in unless there's a different version of Sandra that can run under my existing OS.

[NickS]

I think I can find an older version, but it'll take a day or two

[Denniss]

Maybe Aida32 works with Win95 - it's a comparable to Sandra but much better
www.aida32.hu

[schvankus]

I ran AIDA32 Personal System with 24MB RAM. Here are the benchmarks:

Memory Read: 46 MB/s
Memory Write: unknown - AIDA32 was never able to complete test

Then I reinstalled the 128MB of EDO RAM. It literally took 5 minutes for AIDA to load the program and data! Here are the benchmarks:

Memory Read: 49 MB/s
Memory Write: 68 MB/s

Please advise if there's any other info from AIDA that you need.

Cheers

[schvankus]

Problem solved!

It appears my problem lies with the external cache. Here's an explanation form Bytemare off the Experts-Exchange forum:

Ok bad news first:
32MB is your maximal RAM-size for best performance

Now lets explain:
Cache-RAM needs a TAG-RAM to address the Cachelines.
All TX-compatible Chipsets (an SIS is one) can not cache more then 64MB due to the Limits of TAG-RAM/Cache itself.

In Write-Through Cache-working-mode everything is fine since the CACHE and the RAM are synchronised (no additonal information needs to be stored in the Cache).

If you set Write-Back for the Cache-working-mode, there will be a need for the so called DIRTY-BIT to sign which cachline has been changed and which not. This DIRTY-BIT will be stored in the TAG-RAM and reduces the adressable Cachlines by 2. So in this mode only the half (32 MB) is chachable.

If you CAN NOT change this BIOS-feature you are lost. One reason is, that the performance is 1-2% better in this mode. Maybe a BIOS-upgrade will help.

OK now what will happen if you have 64MB installed or even 128MB:
The system adresses the RAM from TOP to down so first the NOT cached RAM is used. And that is VERY BAD! especally more noncached RAM is installed.

The slowdown is about 40% up to that speed if you disbale completely the L2-Cache.

My System has 128MB (cacheable 64MB) with a P200 and runs like a P100 so the speed is about 50%

If you really NEED that more RAM (large numberchruncher) to avoid swapping to Disk - than you should go for it - but if you NOT NEED it really than stay with 32MB or find a way to switch to WRITE-THROUGH L2 and install 64MB.

Hope this brings the light in the mystery

I've learned a lot from reading all of your posts. Thank you again for all your help.

g

[Rainbow]

I don't think that the L2 cache has that big impact on performance... My MSI MS-5120 board (with Pentium 133MHz CPU, 128MB RAM and WD 30GB HDD) has faulty L2 cache (causing memory errors when enabled) so I disabled it. I haven't noticed any performance loss. It takes around 50 seconds from pressing the power button to finish loading the last program in StartUp (Windows 95 OSR2).

[schvankus]

This is what I know:

The RAM passed POST
The RAM works fine in other machines
The BIOS settings are correct
The motherboard is physically ok (works with four SIMMs, just not four 32MB SIMMs).
The problem exhibits itself in both 95B and 98SE
The problem is there with both an 850MB HDD and a 40GB HDD.

Bytemare's explanation seems the most plausable.

If you have another I'd be very interested in hearing it!

Thanks

g

[NickS]

How is the performance if you disable external cache in the BIOS with 24 Mb ? Is it similar ?
Hope you've got big space in your mailbox.....

[schvankus]

sorry i took so long to respond.

i ended up putting the 128mb back in and disabling the external cache to see if this was indeed the solution. the result: it still didn't solve the problem.

i put the 24mb back in and enabled the cache and it works fine.

i have no explanation for why this is happening. i feel i'm the only one in the world with this problem!

where shall i go from here???

[NickS]

I tried to send you an old demo of Sandra, but it was rejected - maybe there wasn't room in your mailbox (2Mb). I'll see if I can post it online somewhere. That would give you disk throughput, memory/cache throughput and a summary of the cache state among other things.
Another thing to find and try is Ray Van Tassle's Cachechk program.

[schvankus]

Here are the Cachechk results:

CACHECHK V7 11/23/98 Copyright (c) 1995-98 by Ray Van Tassle. (-h for help)
CMOS reports: conv_mem= 640K, ext_mem= 23,552K, Total RAM= 24,192K
"GenuineIntel" Pentium Clocked at 100.0 MHz
Reading from memory.
MegaByte#: --------- Memory Access Block sizes (KB)-----
1 2 4 8 16 32 64 128 256 512 1024 2048 4096 <-- KB
0: 8 8 8 8 15 15 15 15 15 15 -- -- -- us/KB
2: 8 8 8 8 15 15 15 15 15 15 15 15 15 us/KB
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Timer messed up! fff9756b 0002898b 0009fff9
19 <--- same as above.
20 21 22 23 <--- same as above.

Extra tests----
Wrt 24 24 24 24 24 24 24 24 24 24 24 24 24<-Writing

This machine seems to have one cache!? [reading]
!! cache is 8KB-- 136.0 MB/s 7.7 ns/byte (179%)


>>>> If you think you do have L2 cache, you might have FAKE CACHE chips! <<<<


2.9 clks
Main memory speed -- 75.6 MB/s 13.9 ns/byte (100%) [reading] 5.3 clks
Effective RAM access time (read ) is 111ns (a RAM bank is 8 bytes wide).
Effective RAM access time (write) is 183ns (a RAM bank is 8 bytes wide).
"GenuineIntel" Pentium Clocked at 100.0 MHz. Cache ENABLED.
Options: -t0

CACHECHK V7 11/23/98 Copyright (c) 1995-98 by Ray Van Tassle. (-h for help)
CMOS reports: conv_mem= 640K, ext_mem= 23,552K, Total RAM= 24,192K
"GenuineIntel" Pentium Clocked at 100.0 MHz
Writing to memory.
MegaByte#: --------- Memory Access Block sizes (KB)-----
1 2 4 8 16 32 64 128 256 512 1024 2048 4096 <-- KB
2: 24 24 24 24 24 24 24 24 24 24 24 24 24 us/KB
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 <--- same as above.
20: 24 24 24 24 24 24 24 24 24 24 24 24 24 us/KB
21 <--- same as above.
22: 24 24 24 24 24 24 24 24 24 24 24 24 -- us/KB
23: 24 24 24 24 24 24 24 24 24 24 24 -- -- us/KB

Extra tests----
Main memory speed -- 45.7 MB/s 23.0 ns/byte (100%) [writing] 8.8 clks
Effective RAM access time (write) is 184ns (a RAM bank is 8 bytes wide).
"GenuineIntel" Pentium Clocked at 100.0 MHz. Cache ENABLED.
Options: -t0
CACHECHK V7 11/23/98 Copyright (c) 1995-98 by Ray Van Tassle. (-h for help)
CMOS reports: conv_mem= 640K, ext_mem= 23,552K, Total RAM= 24,192K
"GenuineIntel" Pentium Clocked at 100.0 MHz
Reading from memory. Cache DISABLED. Cache disabled
MegaByte#: --------- Memory Access Block sizes (KB)-----
1 2 4 8 16 32 64 128 256 512 1024 2048 4096 <-- KB
0: 35 34 32 32 33 34 33 31 32 34 -- -- -- us/KB
2: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
3: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
4: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
5 6 <--- same as above.
7: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
8: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
9: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
10: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
11: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
12: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
13: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
14: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
15: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
16: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
17: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
18: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
19: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
20: 36 36 36 36 36 36 36 36 36 36 36 36 36 us/KB
21: 36 36 36 36 36 36 36 36 36 36 36 36 -- us/KB
22: 36 36 36 36 36 36 36 36 36 36 36 36 -- us/KB
23: 36 36 36 36 36 36 36 36 36 36 35 -

[NickS]

OK, a couple of things:
- Where are the results for megabyte 0 in the second run ?
- what was the supposed state of internal and external cache for each run ? The diagnostic isn't very clear.
- What happens when you use the 128 MB ?