I accidentally responded to this in the wrong thread, so I will paste the comments here.

feipoa wrote:

Well, the Baby Screamer didn't even provide a boot screen, so 50 MHz is out.
Do you think the SiS Rabbit will cope with 50 MHz? Aside from the VLSI-based Baby Screamer, I have CHIPS 351/355/356, UMC 481/482, SiS Rabbit 310/320/330, and VIA 481/495. My SiS 460-based board uses a clock gen IC, so I cannot test 50 MHz on it. Since the VIA board is damn slow, I am guessing the timings are ultra conservative and that 50 MHz might work on it.

What minimum SRAM and DRAM speeds do I need for 50 MHz operation?

Anonymous Coward wrote:

I doubt SiS Rabbit would work, because it's a true 386 chipset. The VIA 481/495 and the SiS460 are both designed for 486s, so they might work okay. On a 486, 12ns cache is ideally what you would want for 50MHz, but I've seen plenty of 40MHz 386 boards that use 20ns cache (with 15ns tag). RAM should be 60ns. Faster would be better, but it's basically unobtanium (even though 50ns reportedly exists). Of course, you may be able to do better if your board allows you to insert a lot of wait states. The biggest problem with designing the 50MHz motherboards was supposedly making sure everything was properly shielded due to problems with crosstalk (at least from what I can remember). I don't have much faith in a 50MHz 386. I would think even if you can make it stable, the wait states will kill the performance.

The VIA board does have the option to interject ISA, cache, and DRAM wait states. I tried it at 50 MHz using the SXL, but it was not long-terms table. I ran it with the slowest cache and DRAM wait states, which equate to about the speed of a 40 MHz bus with cache/DRAM on turbo. I was able to run 3dbench, but after more minutes of uptime, it would no longer run 3dbench. I tried disabling the caches and adding ISA delays and waits, but after the board warms up, it won't run 3dbench. I set no 0 ws for the ISA graphics card.

That cache felt fairly warm, but not hot, which isn't usually the case. So I set the TAG to 12 ns, but it did not help. I suppose I could put in 10 ns cache throughout, but I would have to pillage it from numerous boards which use 10 ns as the TAG. I guess I'll try one of the other boards instead.

Anonymous Coward wrote:

I've owned two of those exact boards, however mine did not have the local bus slot soldered (just a regular ISA slot), though the pads are present.

Both of my boards had some kind of defect with the 82C206 system chip, and DMA did not seem to be working properly. This problem affected my ability to use soundcards and SCSI controllers, especially when used together with a 486DLC CPU. Definitely not recommended.

Mine has just an ISA slot. If I find some 486DLC or 386 CPU I'll give the board a whirl.

Anonymous Coward - this is interesting. I am testing the SXL2 and DRx2 on my VIA 481/495-based motherboard and am using a DMA SCSI card, Adaptec 1540CP. When the SXL2 or DLC are in the system , the computer will hang with DOS access to the HDD, that is, after it successfully boots from the HDD. The DRx2 has no problem with the 1540CP card. I always thought the two CPUs went hand in hand inasmuch as compatibility is concerned.

I also thought the DRx2 had the multiplier disabled like the SXL2 does. However, when I run the DRx2 in my VIA board, the system boots with multiplier enabled. Is it standard for the DRx2 to have the multiplier enabled? If so, then that explains why the DRx2 did not work in the VLSI board; it was booting up at 2x40 MHz. In the VIA board, I had to run it at 2x33 for it to run reliably.

DRx2 has enabled multi at default and it can't be changed. SXL2 has multi disabled and it must be enabled later via software.

You have to be really careful, because the register that controls the caching scheme on the DRx2 and DLC, changes the clock multiplier on the SXL.
The DRx2 chip also contains cache coherency circuitry not present in the DLC/SXL. This should not matter on boards which supposedly support the DLC, but makes a difference on pre 1992 386 motherboards.

Anonymous Coward wrote:

You have to be really careful, because the register that controls the caching scheme on the DRx2 and DLC, changes the clock multiplier on the SXL.
The DRx2 chip also contains cache coherency circuitry not present in the DLC/SXL. This should not matter on boards which supposedly support the DLC, but makes a difference on pre 1992 386 motherboards.

How does that affect SCSI DMA?

Assuming your board supports hidden refresh, or has the flush circuit wired up, then there shouldn't be DMA issues. SCSI isn't the only device that uses DMA. Using your floppy drive or sound card will also cause problems on a poorly configured DLC type chip.

Is the 82C206 chip on your motherboard made in 1993 by chance? 1993 was a really bad year to buy a 386 motherboard.

The VIA board only has two chips, the 481 and 495. The BIOS on this board supports L1 cache of DLC's, so I am not sure why SCSI DMA works only with the DRx2. The board supports "Decoupled Refresh" and the BIOS is relatively feature rich. It contains an early AWARD v4.20 BIOS.

Interesting. This might be the only 386 chipset I know of where the 82C206 has been integrated into another chip. Please investigate this problem further if you have time. It sounds pretty similar to what I experience on my OPTI 495SX based board. It wouldn't surprise me in the slightest if VIA integrated a faulty DMA controller. DMA was never one of their specialties.

I am not sure exactly what to investigate. The Adaptec 1520B works on the VIA w/DLC, while the 1540C does not. 1520B doesn't have true DMA for HDD access, 1540C does. The VIA works with the 1540C if using the DRx2. All the safe L1 cache features are set. What I am going to do is run through my stack of 386 boards to see if any other boards exhibit the same anomaly. If they do, it sounds to me as if the DRx2 might be more desirable than the SXL. During the process, I am going to run a brief comparison amongst the chipsets with the 3 CPUS (DLC, SXL, DRx2) w/DOOM, 3dbench, and pcpbench.

I think you should do more tests on the soundcard when the DLC is installed in the VIA board, especially with SCSI installed.

Anonymous Coward wrote:

I think you should do more tests on the soundcard when the DLC is installed in the VIA board, especially with SCSI installed.

Are you implying that adding sound will correct the 1540C SCSI issues on the VIA board? I received some pretty impressive SXL-50 results, which ran DOOM and Quake fine. Maybe the presence of the 1540C is what was rendering the board unstable with this configuration. I don't really want to focus too much more on the VIA board at this time until I run through all my boards with a 50 MHz bus, then select the best of the lot for extended testing.

I just tested the 1520B SCSI card. DLC, SXL, and DRx2 all work fine with it, that is to say, you can navitage directories in DOS, run benchmarks, and boot to Windows 3.11 at 1024x768x65k. The 1540C card only works reliably with the DRx2.

Now that I pulled out the 1520B, I was able to run some Windows 3.11 tests on the SXL at 50 MHz. It seems stable with cache/memory at "Normal". Doom scores 4527 (16.4 fps) on the SXL-50, while the DRx2-66 scores 5384 (13.9 fps). Quake scored 3.1 fps on the SXL-50 and 2.6 fps on the DRx2-66

In your experience, what ISA cards tend to be affected by a 12.5 MHz ISA bus? I tried setting "Extended ISA wait state" to "enabled", however the benchmark scores did not decrease. I also tried setting "I/O Recovery Time" to "enabled", however the benchmark scores did not decrease. Would setting these help at all with running my ISA bus out of spec at 12.5 MHz? What about "ISA Command Delay"? I am surprised that the GD-5434 runs fine with 0 WS. Should it be set to "no 0 WS" at 12.5 MHz?

GD5434 should be designed for a 33MHz interface, so I would imagine it would have no problems keeping up with 12.5MHz at 0ws.

In my experience the cards that will be wonky over 8MHz are SCSI (usually 10MHz is okay), some ethernet, and especially sound card.

If your motherboard chipset has DMA problems, they will surface when you try to play sound fx on your soundcard. The problems should be similar to what you experienced with your 1540.

Many months ago when determining which 386 board to go with, I determined that none of them had issues with sound and DLC/SXL chips w/L1 enabled, nor issues with floppy access. All worked fine with 1520B. It is only the 1540C which is causing issues. The fact that using the DRx2 fixes these 1540C issues, I think is quite a discovery. It will be interesting to see how the 1540C works on the various other 386 boards.

From my experience my OPTi 495SX absolutely hates when I use the 1540C and SXL together. Swapping in the 1520B (or IDE) seemingly made things better, but when trying to play SFX in games the system came crashing down. My problem is likely more extreme than yours.

For the record, the ASI 495SX board I used to own worked flawlessly with either the SXL or the DRx2-66 when used with the 1540C

Did you try the OPTi 495SX with the DRx2-66 and 1540C?

I'd like to say I did, but honestly I can't remember. It's been 8 years since I sold the chip.

Well, depending on the outcome of my results, you might be looking for it again. Unless, that is, you are more than content with the IBM BL3.

The surface temperature of the cache on the VIA board gets up to 105 F, or about 41 C when running at 50 MHz. The VIA VT82C495 chipset gets up to 125 F, or about 51 C. Do you think this is acceptable? The board otherwise appears stable. I ran Win 3.11 at 1280x1024x256c and opened up several programs at once (Word Perfect, MS Word, Netscape 4, IE3, ACDSee, etc.) and let it sit for 15 min. I ran Quake a few times in loop as well as DOOM. It did not crash, nor did it produce any errors.