I have a weird problem with my DRx2 that might boil down to L1 coherency..

This is my setup:
- FIC 386-SC-HQ mainboard with the Symphony Haydn chipset
- MR BIOS v1.61
- 4 x 4 MB ram modules with 50 ns
- 128 kB cache (12ns tag and 4 x 32 kB 15 ns chips)

So I played around with my 386 CPU collection, and the motherboard operates rock solid with the AMD DX 40, Ti DLC 33, Ti DLC 40 and TI SXL2-50. No crashes, not even when I use QEMM and I can tweak the memory/cache/ISA bus settings pretty well without problems.

Now comes my newest CPU, the DRx2 33/66. I noticed that when I boot with QEMM the PC would lock up during booting to DOS with cryptic error messages. When using only HIMEM the PC could run all the benchmarks, gold memory test and DOOM without problems, but I would get lockups when trying to play Privateer (almost reproducibly at the same spot).

I tried various cyrix.exe settings, although they should not differ at all compared to the 486DLC or SXL2 in my opinion. And those work with my cyrix.exe settings. But anyways, I tried and the only thing that stopped the crashes was if I enabled KEN. Unfortunately this would make the system just as slow as when I would disable the internal L1 cache completely.

This is weird as there should not be a difference regarding cache coherency between the DRx2 and the DLC or the SXL. Especially the DLC 33 which also has the same base clock frequency as the DRx2

Anybody has an idea what I could try?

JohnBourno wrote on 2022-02-09, 22:03:

This is weird as there should not be a difference regarding cache coherency between the DRx2 and the DLC or the SXL. Especially the DLC 33 which also has the same base clock frequency as the DRx2

Perhaps there is a difference. IIRC the DRx2 was an improved DRu2 design, which in turn was using some sort of interposer PCB with extra chips to deal with cache coherency issue. DRx2 was supposed to have this implemented in the CPU silicon.
If this is anything like the weird flex circuits on some SXL chips then the purpose would be to delay flushing cache on bus hold - a way to differentiate between memory refresh cycles and actual DMA transfers. Then there are some Cyrix patents on detecting bus activity via the CPU control lines - but that in turn assumes the chipset will drive those lines during DMA.

That being said, DMAs that write to memory are not all that common in PCs. One good example is floppy reads, but that's pretty much it. Sound capture can also use DMA but games don't do that. So maybe it's just some timing issue or general instability of that particular DRx2 chip, and perhaps only in this mobo. That happens.

BTW, enabling KEN on mobo where this signal is not connected will leave the pin just floating. Depending what state it tends to fall into due to leakage or nearby electric field, it can just be permanently disabled. Which would turn off storing data in L1, obviously. And without internal cache the system works it seems, so that's the culprit.

First, try using BARB cache flushing. Chances are you were using FLUSH input, which like KEN is not connected, but it defaulted to inactive and without floppy reads you'd never know you have broken L1 flushing. That might even work on DLC chips sometimes due to their very small 1KiB cache that tends to just get invalidated by the executed code anyway. If BARB alone doesn't work, try disabling caching of the first 64k in each megabyte - that is A20 mask workaround. If setting both these doesn't help, you just have unlucky combination of CPU and mobo, or perhaps even a faulty CPU (but my bet would be on interaction with the chipset).

Didn't think about the different silicon stuff. That might very well be a difference between the CPU L1 behavior.

I tried all the combinations of BARB, FLUSH, 64kb caching, A20M and quite a few non-cachable areas. Besides KEN not much of a difference in regards of stability and speed. Tested the DMA with the DMA_TST.EXE and it was successful, so that should be a good sign.

Weird: I downclocked the CPU to 25/50 and it runs perfectly stable. Perhaps the L1 was damaged by the previous owner ...

Again, it might be a timing issue between mobo signals and the CPU. Sure, it could be degraded and not work well at 66MHz, but you have to keep in mind these CPUs were not exactly known for their excellent compatibility.
BTW, does your mobo BIOS recognize the Cyrix chips, and did you ever try to overclock the SXL2 to 66MHz?

You're probably right with the compatibility. About the 66 MHz .. this should in theory be only relevant "inside" the CPU right? So all the other signals on the motherboard would be clocked at 33 MHz. Perhaps the lockups stem from a synchronization problem of the 66 MHz L1 cache and the 33 MHz L2 cache.

Do you think it's safe to overclock the SXL2-50 to 33 MHz base frequency? Seems like quite a big step .. even though I have a big heatsink and a good CPU fan...

Any SXL(2) should be run with at least a heatsink I think. I mean these will work without one, for a time at least, but run hot. Possibly close to DX2 kind of hot - 80C and that's in open air. Why do that, and risk what is now a rare chip, when active cooling (even temporary one) is easy to build by yourself? A small Pentium era heatsink with a 12V fan running at 5V is enough to drop the temperature to less than 45C. Mounting is an issue but for testing you can just put it on top of the CPU and secure with some sticky tape on the sides. I use cotton strings - not a perfect solution but works well enough.

Anyway, yes the doubler is in the CPU and the bus should remain at 33MHz, in theory. But it will affect the timing, and maybe even shape, of all signal transitions. A few ns of difference can mean a shift from "works" to "mostly works" like in your case. Also, apparently the co-processor intreface is running at core speed even though it's I/O? I haven't actually tested with a scope but the results I got from my test setup using SXLC2 would suggest that. The first thing you notice is how everything seems to work at 25MHz but with clock doubling a 40MHz NPU can't keep up anymore. Why would that be if the bus timings remained exactly the same?

There are few 386 motherboards that can handle 50 or even 55mhz base frequency, so you can run sxl2 cpu without clock doubling.
This delivers best performance.
There is another thread with shared perf metrics for sxl2 and drx2 cpus.

New surprising development! While testing different base clocks I accidentally overclocked the DRx2 to 43 MHz base clock. I didn't realize it immediately, but was surprised how fast Privateer ran. And it runs without a hitch @43/87. Thanks to the cooler & fan, the CPU doesn't get very hot and it runs everything totally stable. (37 FPS with 3D Bench and 9,2 FPS with PCPlayer)

So this really seems to confirm that the instability seems to be due to specific frequencies on the board, and not due to some wrong L1 settings.

Huh, I didn't see this thread had been revived until now. For others who stumble upon this thread, please try playing with the cyrix.exe settings first. It looked to me like another user was trying to do a hardware mod without having gone through the cyrix.exe options first. View here for help: Register settings for various CPUs

I have a document from a newsgroup that discusses all the less known mods to get these Cyrix DLC chips working. I know I saved it, but after 25 minutes searching, I cannot find it on my hard drive. I'm pretty sure I have it printed and can scan it if you think it important.

JohnBourno, I would be shocked if your DRx2 is really stable at 87 MHz. Can you show us a photo of your heatsink/fan? Did you try installing Windows 95 at 87 MHz and running some Windows apps, like IE5?

feipoa wrote on 2022-02-13, 23:43:

I have a document from a newsgroup that discusses all the less known mods to get these Cyrix DLC chips working. I know I saved it, but after 25 minutes searching, I cannot find it on my hard drive. I'm pretty sure I have it printed and can scan it if you think it important.

Oh, I'd super grateful if you could send it if you stumble upon it!

feipoa wrote on 2022-02-13, 23:43:

JohnBourno, I would be shocked if your DRx2 is really stable at 87 MHz. Can you show us a photo of your heatsink/fan? Did you try installing Windows 95 at 87 MHz and running some Windows apps, like IE5?

Didn't try the Windows stuff yet. Will try it once I have the CD-Rom working again. I fear the laser is misaligned.

The heatsink is the normal heatsink that was installed on the CPU, and the small fan is one that I bought recently. Nothing fancy. I also did the touch test and the CPU only gets gradually warmer if I remove the fan. Based on my experience I'd say the CPU is not really clocked at 43/87, but I see the exceptional PCPlayer and 3D Bench results and they confirm that the PC is *fast*.

I'm currently recording videos and will create a youtube video about the upgraded I did to this PC: AMD DX40 -> BIOS tweaks -> more L2 cache -> Ti DLC 40 -> DRx2 33/66 -> DRx2 43/87

feipoa wrote on 2022-02-13, 23:43:

Did you try installing Windows 95 at 87 MHz and running some Windows apps, like IE5?

so I just finished successfully installing Win95 @40/80 MHz.

The reason for 40/80 MHz was that the first time I tried the win95 install I got errors. I swapped the DRx2 with my SXL2 but still got errors. Swapped the 486DLC in -> errors. Then as a last resort the AMD40 -> errors. After fiddling around I found out that the memory bank 0 was damaged. Moved the RAM to bank 1 and started installing win95. Success with the AMD40, success with the 486DLC, success with the SXL2 @40 MHz and finally success with the DRx2 @40/80 MHz.

I don't have IE5 installed, but IE 4.4 starts fine. Is there any significant difference between the two stability wise?

I don't know, but I've had more cases of IE5 crash on my SXL2-66 compared to IE4. Nonetheless, impressive being able to run a DRx2 at 80 MHz.

JohnBourno wrote on 2022-02-16, 20:38:

feipoa wrote on 2022-02-13, 23:43:

I have a document from a newsgroup that discusses all the less known mods to get these Cyrix DLC chips working. I know I saved it, but after 25 minutes searching, I cannot find it on my hard drive. I'm pretty sure I have it printed and can scan it if you think it important.

Oh, I'd super grateful if you could send it if you stumble upon it!

Attached is the document I had downloaded from some newsgroup long ago concerning the Cyrix DLC mods. It looks to be printed by a cheap bubble jet printer I had in the late 90's, but I scanned it in 1-bit so it won't be obvious to the viewer. I tried to cut out most of my scribbled in notes, but some are still in there.

feipoa wrote on 2022-02-26, 10:44:

JohnBourno wrote on 2022-02-16, 20:38:

feipoa wrote on 2022-02-13, 23:43:

I have a document from a newsgroup that discusses all the less known mods to get these Cyrix DLC chips working. I know I saved it, but after 25 minutes searching, I cannot find it on my hard drive. I'm pretty sure I have it printed and can scan it if you think it important.

Oh, I'd super grateful if you could send it if you stumble upon it!

Attached is the document I had downloaded from some newsgroup long ago concerning the Cyrix DLC mods. It looks to be printed by a cheap bubble jet printer I had in the late 90's, but I scanned it in 1-bit so it won't be obvious to the viewer. I tried to cut out most of my scribbled in notes, but some are still in there.

Cyrix_DLC_hardware_mods_by_Ernie_van_der_Meer.pdf

Thanks for posting this. My board won't even post with a DLC in it, this should help get it running.

When I originally scanned that file, it looked great on my desktop, but I viewed it on my laptop just now and it was overly contrasted. I've since rescanned it with a lower 1-bit conversion threshold and re-attached the file to my previous post. Hopefully you all will find it more agreeable now.

Thanks a lot! Might try this with my other mainboard that didn't like the TI DLC cpu.

Meanwhile I'm back to square 1 with the DRx2 .. I got instability issues again. But like the time before the instability didn't go away when I changed the CPUs. Now I'm trying a PC Chips mainboard but still the same problems. Drives me nuts that I get instability issues with all CPUs even when I run them within the spec frequencies. Last thing I didn't swap until now is the memory module. We'll see if that helps.

A common problem with 386 boards is the cache jumpers are set incorrectly, or sometimes they don't properly support 256K even though the manual has a setting for it. Another issue that is some boards don't like 2- and 3-chip memory. Try 9-chip.

@JohnBourno
What is the other hardware you run this on ?
The posted perf numbers feel low for an 80MHz system.
Do you have high wait states imposed or something like that ?
Still. Great turnaround. Keep it coming.

Opening these old versions of IE4 is not enough. Try to load vogons.org or other more complex web sites than google.com for example.

feipoa wrote on 2022-03-01, 01:10:

A common problem with 386 boards is the cache jumpers are set incorrectly, or sometimes they don't properly support 256K even though the manual has a setting for it. Another issue that is some boards don't like 2- and 3-chip memory. Try 9-chip.

Today I learned! Thanks for the hint, I'll try that. My current 9-chip modules are quite small and slow (80ns) but I bought four 4 MB 50 ns 9-chip modules yesterday, which I hope will help me to circle in on the root cause of the instability.

pshipkov wrote on 2022-03-01, 01:37:

The posted perf numbers feel low for an 80MHz system.

Opening these old versions of IE4 is not enough. Try to load vogons.org or other more complex web sites than google.com for example.

I checked the Phil's Ultimate VGA Benchmark Database and the fastest 386 scores were 25.9 in 3D Bench and 5.9 in PC Player Bench (320x240), so I'm pretty happy with 37.0 and 9.2 with my DRx2 CPU. It beats some 486 DX2 66 with PCI graphic cards on that list.

The PC is heavily overclocked though. The CPU runs on 40/80 MHz instead of 33/66 and the BIOS settings are on 0 WS for everything and 2-1-1-1 for the L2 cache etc..

I finally found the bug that caused the instabilities. One chip in my second set of L2 SRAM chips had also a defect. It ran OK at 33 MHz but became unreliable when faster.

Now everything is running rock solid. Installed Win95, IE5.5, Netscape and no crashes or blue screens. I don't have a network card yet, so I can't test loading websites - just the startup of the browser.

Super happy with the results! And thanks for your help 😀