feipoa wrote:

For these tests, do not use DRAM. Play with DRAM after we get the corruption taken care of. […]
Show full quote

For these tests, do not use DRAM. Play with DRAM after we get the corruption taken care of.

I think your issues are related to caching conventional memory in Windows. You cannot do this with the SXL in Windows.

Please try this on a fresh boot, in this order. Please run all 3 lines right after the other, then test in Windows.

cyrix.exe -e -i1 -b -f- -m-
cyrix.exe -xA000,128 -xC000,256
cyrix.exe -cd

Run cyrix.exe -q and provide a screenshot before and after running this series of commands. Now check for Window corruption. If there is still corruption, try:

cyrix.exe -e -i1 -b -f- -m-
cyrix.exe -r
cyrix.exe -cd

Hey Feipoa,

so I did the tests and I got the following results:
cyrix.exe -e -i1 -b -f- -m-
cyrix.exe -xA000,128 -xC000,256
cyrix.exe -cd
- OK, no corruption

then I tried a few variations (using flush instead of BARB):
cyrix.exe -i1 -f -m-
cyrix.exe -xA000,128 -xC000,256
cyrix.exe -cd
- OK, no corruption

then I added DRAM to both, and still did not get corruption, which is good, but it seems "cyrix.exe -xA000,128 -xC000,256" somehow affects performance, I'm guessing of the cache - using dosbench, with DRAM, I could only get by with a score of 65 before getting parity errors by trying to squeeze more out using DRAM. If I use just cyrix.exe -i1 -f -cd then enabling DRAM, I get a dosbench score of 90.

I did a few benchmarks in windows 3.11 using Wintune and the RAM, and graphics tests were much higher than any other configuration (I think cause I can use the correct VGA driver instead of the generic, however the CPU performance was maybe 1/4 less than other tests - even a CX486 Drx2 with 1KB of cache was ranked as being faster.

I will do a Wintune benchmark later using cyrix.exe -i1 -f -m-, cyrix.exe -xA000,128 -xC000,256 and cyrix.exe -cd with the board specific driver, and then I'll try it with
cyrix.exe -i1 -f -cd with the generic windows VGA driver and see which one has the best overall performance in windows. - unless you think there is a way to optimise this arrangement:

cyrix.exe -i1 -f -m-
cyrix.exe -xA000,128 -xC000,256
cyrix.exe -cd

Thanks!!

Could you please provide a screenshot with cyrix.exe -q prior to running any command line? The command-line options needed depend on what your system BIOS is setting up by default. For example, you might not need -e, you might need -i1 -i2 -i3 -i4, etc.

There is generally little benefit in using FLUSH over BARB, but I have recorded it. If your system hardware does not properly support flush, BARB is needed. Also, if you use a SCSI controller card with the SXL, you will need to use BARB (not flush) in for Windows use.

This command, -xA000,128 -xC000,256 , disables caching of conventional memory. If you are in DOS, you probably don't need to assert this line. It is the fact that conventional memory is not being cache that you are getting reduced DOS benchmark results. The same applies to cyrix.exe -m- in that sometimes software is residing on a 1 MB boundary and the results appear lower. You may not need -m- when in DOS. You can create a batch file to run the appropriate commands for when you enter Windows.

Once booted up, to remove -xA000,128 -xC000,256 , you would type: cyrix.exe -i1 -i2 , but do check that your BIOS hasn't put more restrictions into the other 2 address spaces, namely, i3 and i4.

You can also try -r instead of -xA000,128 -xC000,256

You'll also want to run chkcpu to ensure that your SXL CPU is clock doubling. The DRx2 clock doubles by default.

Could you please provide a screenshot with cyrix.exe -q prior to running any command line?

feipoa wrote:

Could you please provide a screenshot with cyrix.exe -q prior to running any command line? The command-line options needed depe […]
Show full quote

Could you please provide a screenshot with cyrix.exe -q prior to running any command line? The command-line options needed depend on what your system BIOS is setting up by default. For example, you might not need -e, you might need -i1 -i2 -i3 -i4, etc.

There is generally little benefit in using FLUSH over BARB, but I have recorded it. If your system hardware does not properly support flush, BARB is needed. Also, if you use a SCSI controller card with the SXL, you will need to use BARB (not flush) in for Windows use.

This command, -xA000,128 -xC000,256 , disables caching of conventional memory. If you are in DOS, you probably don't need to assert this line. It is the fact that conventional memory is not being cache that you are getting reduced DOS benchmark results. The same applies to cyrix.exe -m- in that sometimes software is residing on a 1 MB boundary and the results appear lower. You may not need -m- when in DOS. You can create a batch file to run the appropriate commands for when you enter Windows.

Once booted up, to remove -xA000,128 -xC000,256 , you would type: cyrix.exe -i1 -i2 , but do check that your BIOS hasn't put more restrictions into the other 2 address spaces, namely, i3 and i4.

You can also try -r instead of -xA000,128 -xC000,256

You'll also want to run chkcpu to ensure that your SXL CPU is clock doubling. The DRx2 clock doubles by default.

Could you please provide a screenshot with cyrix.exe -q prior to running any command line?

Hey Feipoa,

Sure, sorry, I thought it was going to be simple 😀 - here's the system on boot with nothing in autoexec.bat (as you can see I turned off all the Cyrix statements)

There are no BIOS settings available for memory - this computer is really really old - from 1987 - there's a CMOS that saves the system settings, but quite literally I can only specify the hard disk type eg., I'm using "Type 38" and I can specify the floppy drive - 3.5" vs 5.25", and the time and date. There are no memory settings or bus speed settings or anything like that on a true or more modern BIOS.

The good news is that the computer does not use SCSI (it was one of the first to use IDE). Not sure how to test if the computer "correctly" uses the flush pin or not.

for all of the tests I've been using Phil's Computer Labs Dos Benchmark Pack (https://www.philscomputerlab.com/dos-benchmark-pack.html) and using chkcpu to look for doubling and cache enabled / not enabled, as well as the cache check program (interestingly, the cache check does not always detect the 8KB L1 cache, but on other software I do see that cache is "enabled" - but the cache check program is the only one that goes through to test it).

Let me know what you think.

From the cyrix.exe -q screenshot, the L1 cache is enabled by default (-e), however the whole memory range is set to non-cacheable (thus -i1 is needed, and only -i1). It is OK to omit the -e statement in there if you wish, or you can remove it. Given the age of the computer, I'd leave it.

From a fresh boot, this is what I'd run for Windows

cyrix.exe -e -i1 -m- -f
cyrix.exe -cd -xA000,128 -xC000,256

You can also try putting it into one line, but I've occasionally run into issues when -11 and -xA... was needed.
cyrix.exe -e -i1 -m- -f -cd -xA000,128 -xC000,256

Run some benchmarks between BARB and FLUSH to see if one is better or more stable than the other.

cyrix.exe -e -i1 -m- -f -cd -xA000,128 -xC000,256 for FLUSH

and

cyrix.exe -e -i1 -m- -b -cd -xA000,128 -xC000,256 for BARB

Again, if you run into problems with execution, break it up as before,

cyrix.exe -e -i1 -m- -f for FLUSH
cyrix.exe -cd -xA000,128 -xC000,256

and

cyrix.exe -e -i1 -m- -b for BARB
cyrix.exe -cd -xA000,128 -xC000,256

Given the age of your computer, I'm confused as to why FLUSH is working. Why would the FLUSH pin on the motherboard be connected?

ph4nt0m wrote:

The TI486SXL2 is a 3.3V part actually. TI offered it as 3.3V only, 3.3V with 5V tolerant I/O and 5V only. Clock speeds were the same and the manufacturing process was also the same, so it was down to their marketing dept what label to put on. Although 5V made for the power consumption almost 3 times higher. Therefore no good overclocking.

The Compaq Portable 386 was sold with the Intel 386DX-20. I doubt it can go much higher on the system bus, so the TI486SXL-40 is out of question most likely. I suggest to install a TI486SXL2-50 with a heat spreader and try various oscillators to see how far it can go.

I have tested the SXL-40 and SXL2-50 in the PGA-168 form factor at 3.6 V and both worked. Unfortunately, the SXL2-50 would not run at 66 MHz in the board I tested it in.