Those scores are awfully fast. I take it this was not done on a PGA132-based motherboard? I wonder if the ISA versions on PGA-132 follow the trend you provided. What is screen -1 and why are the results less than fullscreen?

my vote goes to cirrus5434 aswell, but it also has its drawback: no ega mode fonts. simcity is one of the famous games that screw up, although there is a tsr program to fix it.

386_junkie wrote:

Both cards are VLB. The 805i for Doom is shit.

There are two possibilities. The first is that ISA is holding back the performance of both chips so much that it doesn't make a difference, or perhaps your 805i card was not optimally configured by the manufacturer. In the old issues of PC magazine, they often noted how two cards with the same chips could return different results. I will do another comparison benchmark again in the future.

Anonymous Coward wrote:

386_junkie wrote:

Both cards are VLB. The 805i for Doom is shit.

There are two possibilities. The first is that ISA is holding back the performance of both chips so much that it doesn't make a difference, or perhaps your 805i card was not optimally configured by the manufacturer. In the old issues of PC magazine, they often noted how two cards with the same chips could return different results. I will do another comparison benchmark again in the future.

Negative, I had two separate VLB 805's and tested them both... not in Windows though.

I now no longer own any VLB - S3 805's... though I do still have the VLB 5434! 😁

feipoa wrote:

Those scores are awfully fast. I take it this was not done on a PGA132-based motherboard? I wonder if the ISA versions on PGA-132 follow the trend you provided. What is screen -1 and why are the results less than fullscreen?

This thread has got so subjective and way off topic already... so yea, I was using a IBM 5x86C to go through and bench each of the VLB cards.

"?less than fullscreen?"... I always take one fullscreen and another -1 to cover all bases. I've seen some folk do one, and others do the other. Besides... it's good to see the notable difference between them.

This is wonderful and all, but this is now a school project, and i'm gonna need to start getting parts faster. Another shortlist of parts i need to grab:
CPU (working on it)
FPU (Are Cyrix FasMath really better?)
Video card (ET4000ax, Mach32, or GD5420/21 (the 5429 is 200 dollars and that's outrageous))
I/O and drive interface cards
Network card (preferably with 10BASE2, 10BASE5, AND 10BASET)
Hard drive (IDE almost certainly, as MFM, RLL, SASI, and ESDI are all old by the early 1990s and SCSI seems like too much)
Case

I figured "-1" was some size less than full-screen, but the confusion relates to why your results reduced when you run the doom benchmark at "less than full-sreen"? When I take DOOM measurements at less than full-screen, the score increases, not decreases. I do not understand how your score decreased at less than full-sreen.

its an old post but i would update my ideas a bit:
1. either integrated with amd386dx-40, or pga socket with 40mhz clockgen. i am uninterested in 386/486 hybrid boards or rapidcad/486dlc.
2. socket for x87 fpu, preferably cyrix fastmath 387-40.
3. cache socket with at least 128kb support, 256kb is even better.
4. 8* 30pin simm sockets.
5. good build quality.
6. battery not leaking.

All of those have been addressed in my board. It has each of those features.
But i really do need whatever advice you can give to me here about the above topics, as this now has a deadline!

feipoa wrote:

I figured "-1" was some size less than full-screen, but the confusion relates to why your results reduced when you run the doom benchmark at "less than full-sreen"? When I take DOOM measurements at less than full-screen, the score increases, not decreases. I do not understand how your score decreased at less than full-sreen.

My better half assured me that I wasn't coming down with the fatal "baby brain!"... must be the wake up calls at night that does it! Something tells me you can relate.

Thanks for pointing it out... I have since corrected/switched them round.

A good example of us being stronger together.

Jupiter-18 wrote:

This is wonderful and all, but this is now a school project, and i'm gonna need to start getting parts faster.

Two fish in a tank... one says to the other... "who's driving this thing!?"

.......

As said previously, I thought this thread was quite subjective without any clear cut action/direction... there is something to be said for "deadlines" and how they help to motivate. 😁

Thinking out loud, I would think that you could throw together a 386 system fairly quickly... but if you are aiming to build the best, high end, top quality component 386... it requires time, sometimes years depending on part availability. I have yet to finish my best 386 system which has been work-in-progress for a long time, having found the best parts for it over a number of years... testing, recording results, comparing, modifying etc... the hang up now? I need to finish fabricating 40ns SIMMs so DRAM can take advantage of fewer clock cycles and higher FSB's. 😎

I think the best approach to fabricating new SIMMs is to use KiCAD to whip up some brand new SIMM PCBs that can take advantage of modern low voltage components (using VRMS). If you're using vintage parts, you're never going to find enough 40ns chips to fabricate higher density SIMMs. Using modern components we can make 16MB 30-pin SIMMs possibly even faster than 40ns.

I'd sell a kidney for a run of 30-pin simms for my IBM model 30-286.

These are four that were work in progress about a year ago using the older method I started with...

I used a dremel to take off the orignal 80ns IC's and soldered on the replacement 40ns IC's. To make the four I must have wasted at least another six sticks and IC's (they didn't POST!)... I think the soldering iron was either too hot or it was on the leg of the IC for too long... the chips are quite sensitive and it is east to blow a transistor internally.

I've started to now just de-solder the IC's and solder directly on to the pads the same 40ns IC's.... but takes a very long time to do and isn't the easiest.... plus I need donor SIMM's for the pcb design and traces etc.

Anonymous Coward wrote:

I think the best approach to fabricating new SIMMs is to use KiCAD to whip up some brand new SIMM PCBs that can take advantage of modern low voltage components (using VRMS). If you're using vintage parts, you're never going to find enough 40ns chips to fabricate higher density SIMMs. Using modern components we can make 16MB 30-pin SIMMs possibly even faster than 40ns.

I have stock piled hundreds of 40ns IC's, they are all quite low power consumption as is. There is a catch when trying to use modern equivalent IC's as each SIMM stick is only 8-bits long... more modern IC's are much wider! i.e. 16 or 32 bits etc so it is physically impossible to put them on a 8-bit bus that each SIMM stick has or divide and split them up over the 1, 2, 3, 4... SIMM's. As long as the IC's themselves are 8-bits wide... they can go on a 30 pin SIMM and it's game on!

Each stick i've made is 1MB... to get around the density issue I plan to use 4 of these (4 x 4MB = 16MB of 40ns): -

386_junkie wrote:

I plan to use 4 of these (4 x 4MB = 16MB of 40ns): -

Will you be able to use 4 of these on the same motherboard? Will four of these modules placed in adjacent or adjacent +1 slots fit parallel together?

feipoa wrote:

386_junkie wrote:

I plan to use 4 of these (4 x 4MB = 16MB of 40ns): -

Will you be able to use 4 of these on the same motherboard? Will four of these modules placed in adjacent or adjacent +1 slots fit parallel together?

... says the forward thinker.

You're right, four of these identical SIMM banks will impede the next one adjacent to it.... typically these come in sets of four (labelled A, B, C, and D) with two of the units (A and B) having a higher designed PCB to rise above the SIMM bank next to it. Though unfortunately while on a trip in the US (during my SIMM project too!) I could only find a bunch of (C and D's) NOS... they are mirror opposites to each other, once inserted... they face away from each other.

To get around the missing (A and B) I have found a couple of 30 pin pass-through risers with no components on them that lift only the SIMM banks up above the unit installed next to it.

I've tested two of them each on a 16-bit 286 and they work fine, getting four to go together (i.e. 32-bit) is still WIP... my work and a young family take all my time at the moment and have put all my projects on hold, but I aim to complete this project before all others.

That's interesting. So if you have 8 SIMM slots on the motherboard, numbered 1, 2, 3, 4, 5, 6, 7, & 8, where would you install A, B, C, and D modules? Do you have a photo of A/B? Do you know what is the function of the IC on the PCB?

That solution wont work in an IBM. Well, as long as you want to use more than 2 simms. Also the board only supports 1MB simms. If I want to use the 5-16mb range, I have to fill one of these: http://minuszerodegrees.net/5162/cards/5162_cards.htm#emea

feipoa wrote:

That's interesting. So if you have 8 SIMM slots on the motherboard, numbered 1, 2, 3, 4, 5, 6, 7, & 8, where would you install A, B, C, and D modules? Do you have a photo of A/B? Do you know what is the function of the IC on the PCB?

I don't have A & B modules... I will just be extending the height of spare C & D's.

I would be limited to only using SIMM slots 1, 2, 3, and 4. There would be no room to fit a module in slot 5! In theory... to go higher than 16Mb which I don't think is necessary... I would have to replace the 9 x IC's on a 4MB SIMM to make 4MB @ 40ns... then install 8 of these to give 32Mb. I say in theory because I am no way going to attempt SMD on the 9 chip 4MB SIMM's... the IC's are just too close together.

No idea on the function as yet... though they are simple TTL logic... no PAL's/GAL's.