Samir wrote:

SCSI was quirky in terms of drivers and compatibility with the various devices before ASPI, but it was almost always a slow drive to blame if the drivers and software were correct. I never saw any IDE system back in the day that could touch a properly set up SCSI system in terms of CPU usage--it was that much of a difference.

And I think you may have the bottleneck backwards as Fast SCSI could transfer 10MB/sec while the ISA bus maxed out at 8MB/sec. If anything, the SCSI drive would have to wait on the bus and cpu.

When I did alot of benchmarks on controllers (mainly VLB, but also ISA) between IDE and SCSI for this thread I noticed that with newer drives IDE wins over SCSI on an older system. Granted, most of it was on VLB, but I still think it holds up even on ISA as a PIO MODE 2 VLB controller had almost the same performance as an ISA IDE controller. A newer SCSI and IDE drive on a ISA bus will most likely max out both bus speeds, so it will come down to cahce and seek times on the drives. You also have to remember that for DOS, SCSI is hampered from the start due to the overhead on SCSI controllers since they need to translate between SCSI commands and DOS int13, while IDE doesn't need to do this. DOS doesnt take advantage of the special SCSI features like command queuing like a proper operating system does. As you say, with the right controller and a supported operating system SCSI was better and faster back in the days.

Samir wrote:

Gahhhrrrlic wrote:

Also I had a cool idea that I would make a RAM drive and run games from it, which would actually solve the other problem I have with my SCSI card being slow as shit (in my win95 machine, which I was going to swap into my 386). It still serves its purpose of offloading CPU overhead but being so slow, it's nice to have a RAM drive for those hdd-intensive games and what better way to do that than with excess super fast ram.

Ramdrive was always awesome. Ram was so unaffordable back then, but today you can easily have a 64mb ramdrive and get some killer speed. Kinda like the trick of copying a cd to the hard drive to make installs or running a game faster.

I did that for my Win95 installs because of how many damn times I needed to reformat. It saved much time.

In my case I have a nice adaptec that transfers on a quantum 4.3 drive at 5 mb/s and then I have another identical drive on an Always IN-2000 card that tranfers at a measly 1 MB/s but that's a card from 92 and only supports 2GB. I still think it's worth using in the 386 just for the CPU overhead relief. Nothing on a 386 that I'm using needs high transfer speeds.

On the topic of ram, if I were to maintain bank interleaving by keeping 4 of my existing SIMMs, wouldn't that be a problem if the SIMMs don't have the same specs as the fast ones I'm buying? How do you interleave 60nS EDO with 70nS non EDO?

I'm more concerned about the slot limits and total system limits, as it would be trivial for me to just buy 8 SIMMs but that would be an insane 32MB.. somehow I don't know the computer can handle it.

30-pin memory is FPM only. So no worries about EDO here.

Speed ratings 60 vs 70ns is only an issue when 70ns isn't fast enough. Then you have to increase memory wait-states. But this is rarely a problem.

Not every motherboard supports bank interleaving - most don't.

Most 386DX boards have no problem with 8x4MB = 32MB of memory. But you will need to increase the cache size to 256KB.

I wouldn't mind doing that if I could actually find some for sale.

Do you have to increase cache for 16 MB?

vetz wrote:

Samir wrote:

SCSI was quirky in terms of drivers and compatibility with the various devices before ASPI, but it was almost always a slow drive to blame if the drivers and software were correct. I never saw any IDE system back in the day that could touch a properly set up SCSI system in terms of CPU usage--it was that much of a difference.

And I think you may have the bottleneck backwards as Fast SCSI could transfer 10MB/sec while the ISA bus maxed out at 8MB/sec. If anything, the SCSI drive would have to wait on the bus and cpu.

When I did alot of benchmarks on controllers (mainly VLB, but also ISA) between IDE and SCSI for this thread I noticed that with newer drives IDE wins over SCSI on an older system. Granted, most of it was on VLB, but I still think it holds up even on ISA as a PIO MODE 2 VLB controller had almost the same performance as an ISA IDE controller. A newer SCSI and IDE drive on a ISA bus will most likely max out both bus speeds, so it will come down to cahce and seek times on the drives. You also have to remember that for DOS, SCSI is hampered from the start due to the overhead on SCSI controllers since they need to translate between SCSI commands and DOS int13, while IDE doesn't need to do this. DOS doesnt take advantage of the special SCSI features like command queuing like a proper operating system does. As you say, with the right controller and a supported operating system SCSI was better and faster back in the days.

Fascinating set of experiments. From what I remember of our own experiments of IDE vs SCSI back in the day, top speed was always one or the other, but the real-world usage speed when the cpu was having to be interrupted by the ide is where things really came into play. Now ide controllers with their own bios were kind of in their own league on par with scsi controllers having the same, so this wasn't a common comparison. It was usually the on-board ide vs an add-in scsi card. But the battle did heat up in the pio and dma days of IDE where the bus speeds were approaching the 20MB/sec of wide scsi. Then ultra-wide came about the speed war continued. The main difference I remember was that if you played an mp3 in winplay from an ide drive, you'd get stuttering because of the cpu interrupts whereas the scsi stayed smooth. The mouse pointer was also a good indicator, freezing during interrupts on an ide system. Really fun going down this memory road for me. 😊

Gahhhrrrlic wrote:

I did that for my Win95 installs because of how many damn times I needed to reformat. It saved much time. […]
Show full quote

Samir wrote:

Gahhhrrrlic wrote:

Also I had a cool idea that I would make a RAM drive and run games from it, which would actually solve the other problem I have with my SCSI card being slow as shit (in my win95 machine, which I was going to swap into my 386). It still serves its purpose of offloading CPU overhead but being so slow, it's nice to have a RAM drive for those hdd-intensive games and what better way to do that than with excess super fast ram.

Ramdrive was always awesome. Ram was so unaffordable back then, but today you can easily have a 64mb ramdrive and get some killer speed. Kinda like the trick of copying a cd to the hard drive to make installs or running a game faster.

I did that for my Win95 installs because of how many damn times I needed to reformat. It saved much time.

In my case I have a nice adaptec that transfers on a quantum 4.3 drive at 5 mb/s and then I have another identical drive on an Always IN-2000 card that tranfers at a measly 1 MB/s but that's a card from 92 and only supports 2GB. I still think it's worth using in the 386 just for the CPU overhead relief. Nothing on a 386 that I'm using needs high transfer speeds.

On the topic of ram, if I were to maintain bank interleaving by keeping 4 of my existing SIMMs, wouldn't that be a problem if the SIMMs don't have the same specs as the fast ones I'm buying? How do you interleave 60nS EDO with 70nS non EDO?

I'm more concerned about the slot limits and total system limits, as it would be trivial for me to just buy 8 SIMMs but that would be an insane 32MB.. somehow I don't know the computer can handle it.

haha, I kept those same win95 install disks on the hd as well--so much faster!

Adaptec always made some great cards, although I recall the Always brand somehow...probably as one of our initial spec'd cards for our 486 build back in the early 1990s.

As far as interleaving, yes, you'll probably need the exact same memory modules as even modern day systems can't deal well with major differences (ecc vs non-ecc).

I wouldn't worry about the upper limits on memory. Unlike today, I think the limits on memory were on what type of modules the system could recognize vs any bios limitations like today. We have 128MB in our Cyrix P166+. I still gotta get that system up and running again as it was a BEAST (just need to get the motherboard repaired as the original varta battery must be dead--it would boot with error msgs but then one day just didn't anymore.)

kixs wrote:

But you will need to increase the cache size to 256KB.

Hmmm...I don't remember any of this back in the day. You got whatever size cache you wanted (ie could afford) from what I recall. 😕

I did that for my Win95 installs because of how many damn times I needed to reformat. It saved much time.

For me it was Win98. I reinstalled so many times that I memorized the CD key. And it's been a few years but... I can recall the last 20 digits but I'm having trouble with the first 5, 🤣

Anyway, I would expect SCSI cards to be a bit faster on the ISA bus because I thought they can use DMA, whereas ISA IDE cards always use port IO and the 600ns cycle time on the IDE cable.

Gahhhrrrlic wrote:

I wouldn't mind doing that if I could actually find some for sale.

Do you have to increase cache for 16 MB?

I'd increase cache size to the maximum in any case. It improves performance from 0 to 10% - depends on the app/game used. Just read feipoa's comment about stability... you won't know till you try 😉

What can't you find for sale?

Samir wrote:

kixs wrote:

But you will need to increase the cache size to 256KB.

Hmmm...I don't remember any of this back in the day. You got whatever size cache you wanted (ie could afford) from what I recall. 😕

Of course it will work with current 64KB. But going to 32MB it would be recommended to have more then that.

Please let me know how more than 64K of cache works out for you. More than 64K wouldn't work properly on my same board.

What symptoms did you have when you tried it?

Several years have passed now and I cannot say for certain, but I think the symptoms were HIMEM memory errors or failure to boot. Others have tried 256K with this chipset and have also failed. I did have another motherboard with the same chipset work with 128K though.

feipoa wrote:

Several years have passed now and I cannot say for certain, but I think the symptoms were HIMEM memory errors or failure to boot. Others have tried 256K with this chipset and have also failed. I did have another motherboard with the same chipset work with 128K though.

Ah, this makes sense now. Yeah, there were quirks like this all over the place in the early motherboards so this doesn't surprise me. And this is why it's so hard to work on all this stuff today, moreless back then.

Can you put 128 in with the existing module capacity or is there some rule that you always need exactly a certain number of modules installed, so the individual capacity needs to change? I could simply install fewer of them if that would work. Seeing as motherboards were certainly designed to be able to handle more than 8MB, there must be a correct way of supporting 16MB. Maybe 256 is simply overkill unless you have maxed out the main memory?

The best ISA VGA card for both speed, compatibility and price I have found is by far the Western Digital WDC90C30 or 31, they are as quick as the Tseng but without the compatibility problems it has with the Keen games, also a lot cheaper to buy than ET4000 is these days. The main difference between the 30 and 31 is some additional windows acceleration functions on the later, but for DOS both are equally brilliant. The western digital often seems to be the forgotten card when it comes to ISA performance so they are still pretty cheap to buy 😀

What about the 90C33?

Anonymous Coward wrote:

What about the 90C33?

As far as I'm aware the WDC90C33 was only on a VLB card, not ISA, at least that is all I can find. Do you know what that chip added to the party?

The 90C33 is basically a 90C31 with a VLB / local-bus interface and support for up to 2 MB of memory. IIRC its acceleration engine is very similar to the 90C31's; about on par with a CL-GD5428/5429 performance wise. I'll agree with BSA Starfire about the 90C3x's generally very good DOS speed and compatibility.

feipoa wrote on 2018-05-17, 23:45:

Several years have passed now and I cannot say for certain, but I think the symptoms were HIMEM memory errors or failure to boot. Others have tried 256K with this chipset and have also failed. I did have another motherboard with the same chipset work with 128K though.

Interesting, I have the same issue with my 386 board that has the Symphony Labs SL82C461 chipset: http://www.win3x.org/uh19/motherboard/show/5265

With 32kb of cache there are no problems, but when I upgraded to 128kb I got the same HIMEM problems. Interestingly if I left the new chips in the socket but just jumpered the board back to 32kb the problem were gone. Are there any theories out there as to why this problem occurs?

Maybe some TTLs that interface the chipset to the cache are marginal/bad?
I'd like to think it wasn't a problem with the cache controller.