Yepp, I can confirm this settings as working. It is a great universal board.
I think it's the only 386 / 486 hybrid motherboard with a good vesa local performance. A 486DLC-40 with an ARK VLB card achieves outstanding 25MB/s vesa performance according to SpeedSys 4.78 .

A sidenote: Despite its name the board has the OPTi 495XLC Singlechip.

When I had my 495SX VLB hybrid board I don't think I ever tried the ARK in it. I remember the Mach64 not being all that fast though. 25MB/sec is pretty damn fast for a 386. Too bad DRAM throughput isn't all that great on the opti 495 chips, otherwise it would be perfect.

Do you have a speedsys snapshot of this system in action?

I use this board in my "ultimate ultima 7" machine and really like it; the turbo function actually lowers the CPU clock instead of doing the wait state trick, etc.

I've had some compatibility issues with it though - my Tseng ET4000AX VLB card doesn't like it (the ISA version of that card is OK though), and it won't even POST with a Roland MPU-401 card installed.

Anonymous Coward,

I seem to notice our hardware collection is closely aligned. I remember seeing somewhere you had an ALI board (1429 chipset) with VLB which was pretty fast and worked fine until you discovered mold or something under the onboad processor which you removed?

You also speak of an Opti-495SX board with VLB? Could I ask how many VLB slots it had? The reason I ask is, i'm trying to ID my version of this board and i'm curious to see if it's in any way similar to yours. Thanks

I did have an ALI 1429, but sadly it didn't have a VLB slot. I did have an OPTi 495SX board with VLB, and that was a pretty solid 386 motherboard. It had 2 VLB slots.

https://picasaweb.google.com/misterzeropage/F … 750894097353090

I traded it for something better though.
At the moment I only have a single 386DX motherboard. 495SX without VLB. I didn't really feel that VLB was proper on a 386.

Really? Interesting to hear you say... I was under the impression that the 386 can utilize VESA hardware however it's effectiveness depends very much on chipset and board settings, also the VLB card itself. Can I ask if this is the only experience you've had with a 386/VLB system, or have you also tried with different systems?

I'm looking to try build a couple of 386/VLB systems using various chipsets, with a view to finding a stable system where the 386 does utilize the VESA slots effectively. I'm aware that motherboards containing a single VESA slot do not communicate at the desired 33Mhz and instead only a little faster than ISA (Maybe this could be the same for 2 slot boards?).

I have a few hybrid motherboards, some with Opti and ALI chipsets... and have seen but do not own a UMC hybrid. The reason I rememberd you had an ALI motherboard was because you said it was fast... must have been confused about the VLB sorry. This makes me look forward to testing my ALI hybrid as not only can it take a 386... but is has 2 x 72-pin simm slots and can accomodate 512K cache, though I can only hope the system will be stable with all of this.

You have owned some really good gear... i'm surprized you let some of it go, especially the #9's!!! I'm curious though, why do you not like the AWE32's?

386 VLB?

Usually I don't ask this, but why? Stick with a good ISA board.

Since the dawn of the PC / networking, we have always craved / demanded extra bandwidth. The 386'DX' is a 32-bit processor, whilst the regular old ISA is 16-bit and can prove to exentually become a bottleneck depending on the application. To upgrade from this bottleneck there can only be one of two solutions... either by widening your throughput channel or upgrade the system entirely. I agree with anyone who argues toward upgrading the system entirely (which I've done), but then again... that will be neglecting a valid challenge of the former, expanding the capability of the 386... which "If" it can be done and proven to make a difference, then why not?

I already have two very good 386 ISA only boards, and they pretty much serve to just about everything that can comfortably run on a 386. Though as a personal project, and for my own curiousity / satisfaction, I have aquired the various hardware to test the performance of each 386 / VLB system using various chipsets to find a combination that does (hopefully!) give a notable (ISA = 8Mhz, VLB = 33Mhz) performance increase, whilst being a fluid and stable 32-bit system (as 32-bit as it can be).

While you might get an increase in measurable percentage, anything that struggled on your ISA 386 will still struggle on a VLB 386. I find that 386 VLB boards are really more 486 board with the capability of running a 386 as well.

I believe there are VLB benchmarks floating around.

I agree entirely... but I have an ALI 1429G board waiting for me at home... which has 3 x VLB slots / 2 x 72-pin simm banks (EDO) w/ 512k L2 cache, and a 132-pin PGA socket crying out... begging to be plugged by a 386, it must be done.

VLB benchmarks? Are they for different VLB graphics cards or for different systems supporting VLB?

Sounds interesting... will have a look.

3DBENCH CPU Benchmark Database

The fastest 386 is VLB I believe.

This thread is awesome!

It's interesting for me to see which chipsets, cache and video mem perform best as this will prove useful as I start my new builds.

I can only see one other ALI 1429 there, w/128k L2 cache and a 1Mb Tseng ISA... with a score on the list of 16.6

I will be looking to compare various DRAM/VRAM VLB cards i've picked up with the two hybrid boards I have so far (Opti and the ALI): -

ALI; 15ns 512K L2 cache, 50ns 128Mb EDO 72-pin simms

Opti; 10ns 256K L2 cache, 60ns 32Mb FPM 30-pin simms

I suspect that DRAM is more suited to dos, with VRAM being more for GUI and 3d related program.

You have owned some really good gear... i'm surprized you let some of it go, especially the #9's!!! I'm curious though, why do you not like the AWE32's?

First I moved from the US back to Canada, and then from Canada to China. Things had to go. I'm lucky I still have what I do, which in my opinion is still quite a lot. My picasa web album is not a complete inventory of everything I have either. I traded the hybrid 386 board away with a number 9 2meg ISA card that I never used for a Roland MPU 401AT. A good trade in my opinion. I still own 4 or 5 good ISA cards with 2MB, and about a dozen VLB cards with 2MB or more, only so many of them can be used at a time.

The VLB implementation on 386s is probably both dependent on chipset (and revision) as well as PCB layout. The 495SX board I had didn't really seem to work that well with a 386 CPU and VLB cards. It was barely faster than ISA. With Cyrix DLC things improved slightly. With the IBM DLC it was almost as fast as with a real 486 cpu. I can't remember exactly how many chipsets suppported VLB in a hybrid setup, but there may have been as many as a dozen. BTW, 512kb is unlikely to work in a 386. All the documents I have looked at for hybrid chipsets explicitly say that 512kb cache only works with the 486 CPU installed. Not sure why.

On all of my systems, I try to stay away from late hardware from any given generation of CPU. I just find it doesn't deliver an authentic feeling experience, and sometimes they cut corners too. For example, I'm not a fan of PCI 486s or 20MHz 286s. I like to focus on quality motherboards from AMI, Micron, Tyan, Supermicro, Mylex etc. My ideal 386 by the way would be MCA, EISA or proprietary 32-bit expansion slots, and would have very fast memory interleaving.

386_junkie wrote:

ALI; 15ns 512K L2 cache, 50ns 128Mb EDO 72-pin simms

Opti; 10ns 256K L2 cache, 60ns 32Mb FPM 30-pin simms

I suspect that DRAM is more suited to dos, with VRAM being more for GUI and 3d related program.

10ns cache is going to be tricky. It seems to have a pretty high failure rate, so you might need a memory tester to sort out the bad ones. We're still not even 100% sure that what we have is not counterfeit. To be honest, I don't think you need faster than 20ns cache (with 15ns tag) on a 386 unless you're trying to run a 50MHz fsb, and even then the board/chipset isn't really designed to handle that.

50ns RAM would be pretty sweet. Do you know what would be even better though? If somebody could fabricate 30 and 72pin SIMMs that used SRAM instead of DRAM, then in theory we could run 0 wait state and wouldn't even need a cache. It's totally possible, it just takes somebody with enough money, time and motivation to do it.

If you can set all the timings / weight states to the lowest settings and your machine works you know you have fast enough cache and memory. Getting faster chips beyond that point doesn't do anything.

To improve VGA performance most benchmarks results that are "out of the ordinary" use overclocked ISA bus.

This is great... thanks for your posts and sharing your thoughts.

Parrallel to what much of you guys are saying, i'm not holding up too much expectations in terms of stability for the high-frequency spec components.

The 10ns SRAM I acquired only for the Opti board as I know that the FSB there can go to 50MHz, whether or not it will be stable with VLB... I will need to wait and see. The ALI board I know for sure is limited to a FSB of 40MHz... so of the two this will (more than likely) have the greater stablity. Much of this is going to be trial and error and I already suspect I will end up using some 15ns SRAM I have stored instead for the Opti... keeping a lid on the FSB at either 33 or 40Mhz if I can get away with it. BTW, the 10ns SRAM was bought from HK so (now you mention) more than likely it will be counterfeit!

For me though it's more the enjoyment of doing it, I know i'll find the time and spend a few solid days at it when I return home from work, hopefully coming out with two working hybrid systems as stabily optimized as they can be. Whether or not either one of them will excel much more than an ISA based system remains to be seen but it will be fun trying... that and seeing a couple of my DX40's taking a nice hot bath in a near complete 32 bit system, aside from the Soundcard of course. To me... the 386 was ahead of it's time, and that's probably why it lasted so long in the fast moving tech industry of that era... aside from not having L1 / pipeline cache, as a 32 bit processor... it deserves to be used as one.

Both your inputs are fantastic, I appreciate thanks. I will no doubt take some pictures and start a thread of my progress, should anyone be interested.

Anonymous Coward wrote:

386_junkie wrote:

ALI; 15ns 512K L2 cache, 50ns 128Mb EDO 72-pin simms

Opti; 10ns 256K L2 cache, 60ns 32Mb FPM 30-pin simms

I suspect that DRAM is more suited to dos, with VRAM being more for GUI and 3d related program.

10ns cache is going to be tricky. It seems to have a pretty high failure rate, so you might need a memory tester to sort out the bad ones. We're still not even 100% sure that what we have is not counterfeit. To be honest, I don't think you need faster than 20ns cache (with 15ns tag) on a 386 unless you're trying to run a 50MHz fsb, and even then the board/chipset isn't really designed to handle that.

50ns RAM would be pretty sweet. Do you know what would be even better though? If somebody could fabricate 30 and 72pin SIMMs that used SRAM instead of DRAM, then in theory we could run 0 wait state and wouldn't even need a cache. It's totally possible, it just takes somebody with enough money, time and motivation to do it.

The SRAM would have similar pin-out to DRAM chips to accomodate? Surely it wouldn't be as simple as de-soldering the DRAM chips and replacing with the SRAM though? Bad day for simms if it didn't work. 😢

No, you absolutely wouldn't be able to desolder DRAMs from old SIMMs and put SRAM in its place. I'm talking about designing an entirely new PCB. I know this is possible, because over on VCF they have built SRAM memory cards for 8088s and PCjrs.