Win 95 memory problem with AMD and 32K

[georg]

I have a system with a AMD-X5-133ADW processor and a retail upgrade W95 (4.00.950). It has always run fine with two 8MB and one 4MB SIMM for a total of 20MB RAM. Recently acquiring two more 8MB sticks, I upgraded RAM to 32K. BIOS is set to autodetect. DOS reports memory OK as 32K, but Windows throws up a system protection error. Any combination of sticks in the first three slots passes memory tests and benchmarks. The 4th slot is a problem with Windows, but not DOS. I read of a problem with AMD and 32K or more of memory and downloaded an update file (iosupd.exe) from Microsoft, but that did not fix it. The Microsoft site suggests it is a problem with some systems when IE 4.0 is not installed. Why? Before I go back to the original config, anyone have any experience with this?

[dencorso]

Try using HIMEMX.EXE instead of HIMEM.SYS. If that doesn't solve your problem, try to use HIMEMX.EXE /MAX=20480 and see if it works (if it does, the problem is really too much memory, if not it may lie elsewhere...). For more on this idea, see also xRayeR on HIMEMX HTH

Edited June 29, 2008 by dencorso

[thydreamwalker]

Use Caution With IE4 as it may cause w95system to not load properly and Be prepared to Re-install if necessary***This occurred in Win3.1 and Win95 systems with 32mbRam on my PI machines and a 486 too! The Larger the Ram above 32mb=IE4 was okay as it uses a lot of resources........

[georg]

Neither sub HIMEMX.EXE for HIMEM.SYS or the HIMEMX.EXE /MAX=20480 setting changed things. Right now I'm back to the two original 8MB sticks in slots 1 & 2, and the 4MB stick in slot 4 so assume SIMM mismatch is the problem. The new 8MB sticks are dated 28 Feb 1997, labeled EDO, chips are 60ns. The motherboard is Elpina 386/486 V.I.P. by Hsing Tech Enterprise Co., Ltd., sim to PC Chips M919, with AMI socketed BIOS dated 10 Oct 1994 and UMC chipset. Older sticks not labeled. Chips are 70ns. Could also be need for 95 drivers for UMC chipset, although everything has always run OK.

[Steven W]

Check the BIOS and see if there is a access time or speed setting for the memory set it to 70ns. This is "access speed" so the higher the number the slower the speed.

You may also look and see if the BIOS will allow enabling/disabling Parity Checking, if so disable it.

Mixing and matching SIMMS just will not work sometimes.

Edited June 30, 2008 by Steven W

[Steven W]

I've given this a bit more thought... Sounds like a Win95 problem, with Dos seeing it and all.

[dencorso]

Neither sub HIMEMX.EXE for HIMEM.SYS or the HIMEMX.EXE /MAX=20480 setting changed things. Right now I'm back to the two original 8MB sticks in slots 1 & 2, and the 4MB stick in slot 4 so assume SIMM mismatch is the problem. The new 8MB sticks are dated 28 Feb 1997, labeled EDO, chips are 60ns. The motherboard is Elpina 386/486 V.I.P. by Hsing Tech Enterprise Co., Ltd., sim to PC Chips M919, with AMI socketed BIOS dated 10 Oct 1994 and UMC chipset. Older sticks not labeled. Chips are 70ns. Could also be need for 95 drivers for UMC chipset, although everything has always run OK.

OK. So much for that. Now I've got three different questions:

0)I understood you told us the problem only occurred with all 4 slots in use. Why not use 3 x 8 MiB?

1) If that works, using just the first three slots, can you use indifferently either two old sticks plus one newer stick, or one old stick and two newer? If so the problem is not the timings. If not, then it definitely is the timmings.

2) Using only the old sticks, can you work with any of these configurations: slots 1 used, 2 used, 3 free and 4 used or

slots 1 used, 2 free, 3 used and 4 used, or else slots 1 free, 2 used, 3 used and 4 used? If all three cause you problems, it is the 4th slot that is defective...

Edited June 30, 2008 by dencorso

[PassingBy]

Hmmm .....

Ahhh ... nostalgic issues ... pcchips M919 (socket 3) ... please have a look here to see if this will help you :

http://th2chips.freeservers.com/m919/

-and-

http://th2chips.freeservers.com/m919/unoff/m919.html

I assume the M919 is a 72pin SIMMs versions ... please check that you're not mixing FP ram and EDO ram ... they don't play well together ...

Last but not least is the possibility that one of the SIMM may be having a defective region ... try the free memory tester from simmtester.com ...

If I recall correctly, unbranded (or unknown manufacturer) SIMMs back then were prone to stability and mixing issues ...

Rgds

[Steven W]

I'm kind of curious about something. You said DOS sees the last slot, but Windows doesn't. What if you configured the autoexec.bat and config.sys files in Windows the same way it is in DOS?

[georg]

Quote: (Steven W) "see if...access time or speed setting" "see if the BIOS will allow enabling/disabling Parity Checking"

No settings in BIOS for speed or parity. Once autodetect is checked, no choices. Unchecked, can make very limited changes to cache timing, and add one wait state to read and write.

Quote: (georg) "a problem with some systems when IE 4.0 is not installed. Why?"

Installed IE5.5 with no change. Uninstalled IE5.5 after test. Downloaded SP1 for 95 from Microsoft. No change.

Quote: (dencorso) " Why not use 3 x 8 MiB?"

I've lost confidence in mixing the sticks, and not much practical difference between 20MB and 24MB. Reinstalled the 60ns EDO SIMM's and rechecked benchmarks. No faster than the old sticks. Reinstalled the original config with 2 8MB sticks in slots 1 & 2 and the 4MB stick in slot 3, all 70ns.

Quote: (dencorso) "using just the first three slots, can you use indifferently either two old sticks plus one newer stick, or one old stick and two newer?"

Originally two old sticks plus one new seemed OK, but after the software updates, I got new problems, especially with two new sticks plus one old stick. Instead of the original error message, got multiple messages that vxd's were missing or corrupt. Took the new stick out and all was fine.

Quote: (dencorso) "can you work with any of these configurations"

It was in my head that the first two slots had to be filled with a matched pair. "1 used, 2 used, 3 free and 4 used" works OK with the old RAM. Changed sticks around so often that slot 3 quit working. After getting it resurrected, decided to quit while I was even and buttoned up the case. I don't think there is anything wrong with any of the slots. If all four are filled, POST is normal and DOS reports 32MB RAM. I played several DOS games, used XTPRO, memmaker, chkdsk, scandsk, etc. with no problems, but couldn't get W95 to load from its own DOS base.

Quote: (PassingBy) "check that you're not mixing FP ram and EDO ram" (and try) "simmtester" "What if you configured the autoexec.bat and config.sys files in Windows the same way it is in DOS?"

I used the BIOS memory check, the read and write benchmarks in AIDA32, and ran some apps. Don't know for sure that the old RAM is fast page. It has always worked reliably. The EDO RAM works OK by itself. There is some issue with this combination of 1994 + 1997 RAM with W95 and AMD. The problem is not intermittent. I used to mix RAM sticks willy nilly with no problems at all, but only used Intel processors with W95 and the most RAM I ever installed back then was 96MB on a Soltek board with a Pentium 166. This system boots to DOS, loading autoexec.bat and config.sys. If I want to use W95, I just type WIN at the C:\ prompt. It's all the same environment.

Scott Mueller in "Upgrading and Repairing PCs" 8th Edition (1997) says:

"Some 486 motherboards may have problems addressing memory past 16M due to DMA controller problems...because the ISA bus only allows DMA access to 16M. This should not be a problem with newer 32 bit operating systems."

That seems to imply W95 would be better at managing larger amounts of memory. When those SIMM's fell into my lap it seemed like a good idea at the time to use them for something. Should have known better.

[Steven W]

When those SIMM's fell into my lap it seemed like a good idea at the time to use them for something. Should have known better.

With stuff like this, you don't know until you try.

[georg]

Quote: (PassingBy) "have a look here to see if this will help you"

Thank you for those links. Lot's of good info there. Also found an AMI site in Germany that gave more precise motherboard ID from the BIOS code appears at boot. My board turns out to be more like M921 than M919, but it's the same family. Steve Dabrowski's comment "make sure you have identical 72 pin SIMMs" seems to fit. My version number 3.4B/F is a late model. The UMC chip numbers match, and the code he gives on the 8881F chip indicates EDO RAM can be used.

dencorso:

Since I took the cover off again, tried the other combos you suggested, and some others.

1 used, 2 free, 3 used and 4 used (4MB in slot 4)

DOS reports only 4MB RAM

1 free, 2 used, 3 used and 4 used (4MB in slot 4)

DOS reports only 4MB RAM

8MB sticks in 3 & 4, 4MB stick in one or 2

DOS reports 20MB and Windows runs fine.

Old RAM in 1 & 2 and 1 stick EDO in 3 or 4

On 1st try, DOS reported 24MB and Windows loaded, but retried this later with both EDO sticks in 3 or 4 and couldn't repeat. Windows would not load.

8MB Old RAM in 1, 8MB EDO in 4

Won't boot

EDO in 3 and 4

DOS reports 16MB, Windows loads, benchmarks complete OK

Following are results using the old Discover benchmarks that came with the Hurricane drive rocket:

Sieve

190.6 with 16MB FP

190.6 with 20MB FP

189.0 with 16MB EDO

Dhrystone

81.9 with 16MB FP

82.4 with 20MB FP

78.9 with 16MB EDO

ModeSwitch

23.4 with 16MB FP

23.4 with 20MB FP

19.6 with 16MB EDO

Whetstone

16.9 with 16MB FP

16.9 with 20MB FP

16.8 with 16MB EDO

Discover Mark (% of Pentium 100)

76.6 with 16MB FP

76.8 with 20MB FP

74.6 with 16MB FP

Without the Intel Triton chipset, EDO does not appear to be doing anything, so I'm buttoned up again with the same old 20MB. Still don't understand why DOS 7.0 works but W95 doesn't. The problem seems to be mixing FP and EDO rather than anything magic about 32MB, but it isn't worth buying more old RAM to prove it. SP1 appears to have removed the 3 X 8MB option, but maybe it was just a fluke that it seemed to work during the initial messing about.

[PassingBy]

Hmmm .....

Without the Intel Triton chipset, EDO does not appear to be doing anything, so I'm buttoned up again with the same old 20MB. Still don't understand why DOS 7.0 works but W95 doesn't. The problem seems to be mixing FP and EDO rather than anything magic about 32MB, but it isn't worth buying more old RAM to prove it. SP1 appears to have removed the 3 X 8MB option, but maybe it was just a fluke that it seemed to work during the initial messing about.

As I mentioned earlier, FP and EDO does not play well together ... If I recall correctly (my memory is a wee bit hazy here ... it's been quite a long while ...), it's due to different bus cycle/signal timings between the EDO and FP ... While they share the same pin counts but mixing FP with EDO makes the FP goes 'huh?' as the EDO are able to retrieve more bits within the same cycle and the FP says 'what !? where !? when !?' ...

Again, this is not something BIOS memory timings can help much ... It's like trying to make an unbuffered DDR module work with a registered DDR module (well, something like that) ... They just don't play nicely ... My last socket 3 was a generic SiS496 pci that also does not work with mixed FP+EDO ...

Memory region errors usually does not appear under DOS as DOS kernel is loaded and stays within the first 1 MB ... All the rest will become XMS/EMS ... If the defective region(s) is above 1MB then DOS never crash, the later loaded apps will kindly do that when they use the defective regions ...

Win9x is different as the kernel loads high up in memory and if this is where the defective region is located then it usually throws up BSODs very very early .....

On another note, Intel triton was intended for the socket 7 pentiums (p75-p233mmx) ..... The UMC8881 were socket 3 for the much more earlier i486 family ... The AMD X5-133 is not anywhere near the same as the AMD K5/K6, it's really just a turbo-boosted 486 .....

Rgds

[georg]

Quote: (Steven W) "you don't know until you try"

It's a good way to find out what you don't know.

PassingBy:

Thank you for your explanation. I appreciate your taking the time to do it.

My thanks to all who responded.