aquishix wrote:

I will soon be running benchmark comparisons that bear on these questions. […]
Show full quote

Scali wrote:

My guess would be 'no'. I bought my 486DX2-66 in 'phases'... Initially I bought the CPU+motherboard, and re-used the memory and […]
Show full quote

The Serpent Rider wrote:

Are there any games which clearly benefit from having VLB card on a 386 system?

My guess would be 'no'.
I bought my 486DX2-66 in 'phases'... Initially I bought the CPU+motherboard, and re-used the memory and ISA VGA card from my 386SX-16.
Even with the DX2-66, the ISA Paradise card I had was doing quite a nice job in most games.
I later upgraded to a Diamond Speedstar Pro VLB, which certainly was faster, but not orders of magnitude.

A friend of mine also later got an old 486 machine with ISA only, but with an ET4000 installed. He got even better performance than my Paradise did.

Based on these experiences, I would say that even on a 486DX2-66, the move from ISA to VLB is not that large.
Extrapolating that to the much slower 386DX-33 or 40, I don't think a fast ISA card would be much of a bottleneck at all, in most games.
Yes, in raw bandwidth, the VLB card would be better, but the CPU probably cannot saturate that bandwidth anyway. Besides, a game does a lot more than just writing to videomemory, so the real-world advantage is much smaller than the difference in bandwidth would suggest.

Having said that, I would certainly be interested in benchmarks from a 386 with VLB, compared to a fast ISA card, such as the ET4000.
If you were to run a game such as DOOM, can you see a difference?

I will soon be running benchmark comparisons that bear on these questions.

I've got a Opti 495SX motherboard with a soldered-on Am386DX-40 CPU, but it only has one VLB slot. Instead of having a fancy VLB video card in there, I've got a VLB multi I/O card, which seems like a better choice for improving system performance. I just snagged a motherboard from Israel on eBay which appears to be the only one for sale in the world which has the same CPU on board, and has --two-- VLB slots. I already figured that what you said in this post is correct, re: video card performance on a 386 system. I just figured it couldn't hurt to have two VLB slots, and I couldn't find any motherboards for sale that are comparable to my Opti 495SX that I currently have (I.e., having just 1 VLB slot.)

The problem with my system is that the CPU benchmarks are just coming in really low. The benchmark numbers for my CPU are showing the performance ofa. 386SX-25 or lower, and I can't explain why. I've done absolutely everything I could think of, up to and including permuting jumper settings, adjusting BIOS settings, and upgrading the cache from 20ns to 15ns, and doubling the amount of cache from 128KiB to 256KiB. Nothing has budged the performance even a tiny bit. Warcraft I runs like complete crap on this system and it's pretty frustrating.

So, I think I'm going to get an ISA-only motherboard with an Am386DX-40 CPU while that other motherboard ships from Israel, and see if, paired with an ET4000 card (which is also on its way) that I can get decent system performance out of it. I'm not stuck on having VLB cards in my 386 system, but I didn't expect to face these weird problems, either.

Thoughts?

Check the jumper for turbo function.

alvaro84 wrote:

aquishix wrote:

I'm not stuck on having VLB cards in my 386 system, but I didn't expect to face these weird problems, either.

Thoughts?

Back when I had a 386/VLB board with the OPTi495XLC chipset it was an underperformer too. And even though I found the reason in relaxed cache and memory timings I couldn't rectify it. I tried a different set of cache chips but couldn't seem to get it working with tighter timings so it remained slow in comparison to my other 386 boards. What's even more interesting later I got a small board with the same chipset but without a VLB slot and it performs just like my faster ALI ones.

I was so disappointed with the VLB386 that I got rid of it in the end. It was either slow or unstable - and the other, faster ones performed at least as well with fast ISA cards.

Excellent! That's very encouraging information and gives me more confidence that I'm honing in on the root problem.

I really appreciate it. =)

Find a proper manual or jumper settings for your board.

Then check jumpers for:

- system bus speed (unless you have an oscillator - what is its speed?)
- turbo jumper!!

Only these two settings have mayor impact on the system performance. BIOS settings can have around 10-15%.

Post the photo of your board and results from Speedsys, NSSI.

kixs wrote:

Find a proper manual or jumper settings for your board. […]
Show full quote

Find a proper manual or jumper settings for your board.

Then check jumpers for:

- system bus speed (unless you have an oscillator - what is its speed?)
- turbo jumper!!

Only these two settings have mayor impact on the system performance. BIOS settings can have around 10-15%.

Post the photo of your board and results from Speedsys, NSSI.

It's definitely not the turbo button; I wasn't born yesterday. 😉 When the turbo button is pushed, the system crawls to about the speed of an XT Turbo or maybe a low-end 286. Otherwise it performs like a low-end 386.

I'm pretty sure it's not the bus speed either, but not quite as confident about that. The oscillator is rated at ~14.3MHz, as you can see in one of the photos I'm going to attach to this message.

As you requested, I ran the two benchmarks you mention.

I couldn't get a good picture showing this, but part of the silkscreening reads: "VER-1.2 495SX"

I can't find the exact manual/jumper settings for this motherboard, but the two closest I've been able to find are:

https://stason.org/TULARC/pc/motherboards/I/I … OPTI-495SX.html

https://stason.org/TULARC/pc/motherboards/D/D … OPTI-495SX.html

NSSI reports that it's a Dataexpert motherboard, which is plausible, but the closest Dataexpert motherboard listed on stason.org has JP15 and JP16 in between the ISA slots instead of JP51 and JP52...

Here are my jumper settings:

• JP1 closed User-selectable
  JP2 1&2 User-selectable
  JP3 2&3 User-selectable
  JP4 1&2 User-selectable
  JP5 1&2 User-selectable
  JP6 closed User-selectable
  JP7 2&3 ? User-selectable
  JP8 2&3 Hard-wired
  JP9 2&3 Hard-wired
  JP10 2&3 Hard-wired
  JP11 2&3 Hard-wired
  JP12 2&3 ? User-selectable
  JP13 2&3 User-selectable
  JP14 2&3 User-selectable
  JP15 .
  JP16 .
  JP17 closed User-selectable
  JP18 all open User-selectable
  JP19 .
  JP20 open User-selectable
  JP21 .
  JP22 open User-selectable
  JP23 open Hard-wired
  JP24 1&2 User-selectable
  .
  .
  .
  JP51 open User-selectable
  JP52 open User-selectable

kixs wrote:

Post the photo of your board and results from Speedsys, NSSI.

Here's the next batch of photos. I pushed the turbo button on one of these to show you what I mean about how the performance drops off a cliff. 😉

kixs wrote:

Post the photo (...)

And here's the last batch.

(Sorry for the image spamming, guys!)

You said performance of your board is like 386SX-25. By the benchmark results it looks more like DX-33 or even better. Speedsys result of 8.11 is in range of 386DX-40.

This is your board:
https://stason.org/TULARC/pc/motherboards/D/D … OPTI-495SX.html

My guess there could be some setting in the BIOS that drops the performance. Play around with it or post screenshots of all BIOS settings.

PS:
Turbo button is the most common mistake when performance difference is too big - like 386sx levels.

After seeing these numbers I'd guess memory + cache timing.

kixs wrote:

You said performance of your board is like 386SX-25. By the benchmark results it looks more like DX-33 or even better. Speedsys […]
Show full quote

You said performance of your board is like 386SX-25. By the benchmark results it looks more like DX-33 or even better. Speedsys result of 8.11 is in range of 386DX-40.

This is your board:
https://stason.org/TULARC/pc/motherboards/D/D … OPTI-495SX.html

My guess there could be some setting in the BIOS that drops the performance. Play around with it or post screenshots of all BIOS settings.

PS:
Turbo button is the most common mistake when performance difference is too big - like 386sx levels.

1) That is NOT my motherboard. As I pointed out, my motherboard has JP51 and JP52, not JP15 and JP16. Whether or not it's crucial that it's not quite my motherboard is another question, however. I suspect that some small company in Taiwan made a knock-off of that DataExpert 495SX motherboard that is very similar, but not quite identical.

2) The benchmark results are not reflecting reality, though. I distinctly remember playing two video games back in ~1994 on my 386SX40: One Must Fall 2097, and Warcraft I. Both games ran very well, even with sound. Neither game runs anywhere close to acceptably on this Am386DX-40, even with sound disabled. Perhaps there's a weird resource conflict (IRQ?) that doesn't manifest when running a synthetic benchmark. Perhaps it's a RAM issue. But in any case, the system is nowhere near the performance of a 386DX-33. When I get home tonight, I'll try removing every ISA card except for the VGA card and the multi I/O card. If it still behaves the same way, I'll try putting an ISA multi I/O card in there. If that still doesn't work, I will be completely out of ideas.

3) Re: BIOS -- yes, I've tried everything in the BIOS settings. Nothing helped. I thought that changing it from CLK/4 to CLK/5 would do the trick, since 40/5 = 8 ~= 8.33, but no such luck.

4) Re: Turbo button -- yes, I know. I've dealt with that issue many times. I do appreciate the suggestions!

konc wrote:

After seeing these numbers I'd guess memory + cache timing.

Wrong kind of RAM, or what? There aren't any jumper settings that pertain to the RAM other than the amount of it.

Re: cache timing -- I tried 20ns and 15ns cache, and saw no difference in performance whatsoever.

aquishix wrote:

3) Re: BIOS -- yes, I've tried everything in the BIOS settings. Nothing helped. I thought that changing it from CLK/4 to CLK/5 would do the trick, since 40/5 = 8 ~= 8.33, but no such luck.

aquishix wrote:

Wrong kind of RAM, or what? There aren't any jumper settings that pertain to the RAM other than the amount of it.
Re: cache timing -- I tried 20ns and 15ns cache, and saw no difference in performance whatsoever.

BIOS settings. Judging by what you wrote for the bus speed, better post (the already requested earlier) pictures with your settings.

konc wrote:

aquishix wrote:

3) Re: BIOS -- yes, I've tried everything in the BIOS settings. Nothing helped. I thought that changing it from CLK/4 to CLK/5 would do the trick, since 40/5 = 8 ~= 8.33, but no such luck.

aquishix wrote:

Wrong kind of RAM, or what? There aren't any jumper settings that pertain to the RAM other than the amount of it.
Re: cache timing -- I tried 20ns and 15ns cache, and saw no difference in performance whatsoever.

BIOS settings. Judging by what you wrote for the bus speed, better post (the already requested earlier) pictures with your settings.

I will certainly do that when I get home this afternoon, but I think you're barking up the wrong tree. This motherboard doesn't currently have a CMOS battery installed, so the settings are cleared every time I power cycle the system. It performs like complete crap with the default settings. When I try changing the BIOS settings, I change them carefully one at a time, and only after reading up on them to fully understand what they mean. The shadowing settings, and the CLK/X settings are the only ones that I've ever seen alter performance significantly on any of these 80s-90s systems, and neither had any significant impact in this case.

kixs wrote:

You said performance of your board is like 386SX-25. By the benchmark results it looks more like DX-33 or even better. Speedsys […]
Show full quote

You said performance of your board is like 386SX-25. By the benchmark results it looks more like DX-33 or even better. Speedsys result of 8.11 is in range of 386DX-40.

This is your board:
https://stason.org/TULARC/pc/motherboards/D/D … OPTI-495SX.html

My guess there could be some setting in the BIOS that drops the performance. Play around with it or post screenshots of all BIOS settings.

PS:
Turbo button is the most common mistake when performance difference is too big - like 386sx levels.

Here is the benchmark that I was using prior to your suggestions. (Attached.)

I think you would agree that based on THIS benchmark, my math wasn't far off: If a 386DX-33 measures 35.9, then 30.2/35.9 ~= .84123, and .84123 * 33 ~= 27.76. I've also seen it come in at 30.0 plenty of times, which would result in a MHz equivalent estimate of 27.58 -- which is only 10.32% more than 25. So I was within ~10% based on the benchmark I had been using prior to your suggestion to use NSSI. This is one of the benchmark programs that comes with Phil's Computer Label DOS benchmark pack, but I can't remember the specific benchmark name at the moment.

I'm beginning to suspect the that the FPU is actually the problem, based on the FPU Performance screen in NSSI. If these benchmarks are using integer math but the FPU is hung up for some reason(mysterious jumper settings?), that would explain it. I can re-compile my Mandelbrot benchmark and target the i386 architecture this afternoon, and then run it on my 486DX2-66 system and my XT Turbo 10MHz system for comparison. Both of those systems are performing exactly as they should, so that should help box in the numbers and convince me that I'm measuring what I should be measuring. It uses so few variables that it never hits main RAM; everything is allocated on the stack and fits in less than 64 bytes, IIRC. I could write an ASM program that would fit in the registers entirely and not even bring cache into the picture, just to be sure.

Fun times.

My DX/40 with the optimum vs the default safe settings and a crapton of wait states. Does the number achieved using the the safe settings look familiar?

konc wrote:

My DX/40 with the optimum vs the default safe settings and a crapton of wait states. Does the number achieved using the the safe settings look familiar?

SYSINFO.jpg

Holy Hell!

What are your wait state / timing settings in the BIOS?

Lowest possible of course, WS set to 0 and timing to the tightest. But I thought you "changed them carefully one at a time, and only after reading up on them to fully understand what they mean" and that "I'm barking at the wrong tree" so it can't possibly be that 🤣

(lower everything, memtest and run stuff, don't just assume your ram/cache can handle them just because it boots. Oh, and leave that poor CLK setting to /5 -unless you intentionally overclock the bus)

konc wrote:

Lowest possible of course, WS set to 0 and timing to the tightest. But I thought you "changed them carefully one at a time, and only after reading up on them to fully understand what they mean" and that "I'm barking at the wrong tree" so it can't possibly be that 🤣

(lower everything, memtest and run stuff, don't just assume your ram/cache can handle them just because it boots. Oh, and leave that poor CLK setting to /5 -unless you intentionally overclock the bus)

As far as I can tell, there's no contradiction here. I didn't claim that I traversed the *entire* set of possible settings in the BIOS. Just that none of the changes I'd ever made made any positive improvement. I seem to recall that when I tried WS 0 on this setup, it performed even worse or crashed. Perhaps I'm just unlucky with this motherboard. And yes -- leaving it at CLK/5 seems to be *correct*, so I had no intention of changing that again. I guess I'll just map out all the combinations and try them one at a time until I hit a good one. In my defense, it doesn't make any sense that the defaults on this BIOS result in terrible system performance! Maybe someone upgraded the BIOS chip at some point and the new one was intended for a different motherboard.

Anyway, shouldn't take any longer than doing the same thing with the jumpers on my multi I/O card. Heh. I need to post the results of that on VOGONS and stason.org, because that card has been discussed on here before and no one could come up with the manual. (Off-topic at this point.)

Could it be that mixing RAM types is the problem? I have two different 4MiB kits installed in this system, and I don't know what their specs are.

You don't need to try all the combinations. Just lower a value and test. If stable, lower more and re-test. Then move to the next setting. There is no magic combination, just a matter of using the lowest value your system is stable with for all related settings.

konc wrote:

You don't need to try all the combinations. Just lower a value and test. If stable, lower more and re-test. Then move to the next setting. There is no magic combination, just a matter of using the lowest value your system is stable with for all related settings.

Interesting. I was under the distinct impression that RAM timing settings were not that simple to figure out because of interactions between the clock speed and latencies for a given RAM kit. Because of that, I've had a tendency not to screw around with them unless I knew what the correct settings were.

2) The benchmark results are not reflecting reality, though. I distinctly remember playing two video games back in ~1994 on my 386SX40: One Must Fall 2097, and Warcraft I. Both games ran very well, even with sound. Neither game runs anywhere close to acceptably on this Am386DX-40, even with sound disabled.

Taking this into consideration... the problem is your VGA card. If you have any other (even Trident 8900D/CL will be better) test it.

Download Phil's DOS benchmark suit and run 3DBENCH, PCPBENCH, DOOM and post results.