VOGONS


AMD 5x86 X5-133 (now with POD)

Topic actions

Reply 20 of 92, by cxm717

User metadata
Rank Member
Rank
Member
derSammler wrote:

While I have a spare Voodoo2, I don't see much point in testing that. The 486 can not fully utilize the Voodoo1 because it has to do T&L in software, for which it is not fast enough to push the Voodoo1 to its limits. The second TMU on the Voodoo2 won't help here, specially since it would only improve rendering speed if a game uses multi-texturing.

The Voodoo Rush is the only 3dfx chipset that works better with slow CPUs compared to the Voodoo1.

I've seen benchmarks showing the voodoo2 being faster than a voodoo1 even on a P100. I'm not really sure why. I have heard that it's because the voodoo2 does full triangle setup in hardware.

Reply 21 of 92, by derSammler

User metadata
Rank l33t
Rank
l33t

On a P100, it's possible. Depending on the game, a P100 may no longer be the bottleneck for software T&L. But by no means the second TMU helps any when using a slow CPU.

btw, the Rush does triangle setup partly in hardware, too. That's why it is better for slower CPUs.

Reply 22 of 92, by derSammler

User metadata
Rank l33t
Rank
l33t

New results. 😀

Sadly, PCI runs with 20 MHz now, since the M919 does not allow overclocking PCI. It does FSB/2 automatically for PCI if FSB is higher than 33 MHz. Overall speed is still much higher in all games, but according to Dr. Hardware, video speed lost more than 10 mb/sec. I may *have* to switch over to a good VLB card now to fully utilize the 40 MHz.

Last edited by derSammler on 2018-07-08, 13:53. Edited 1 time in total.

Reply 23 of 92, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I also think it would be neat to see if there is any benefit, or lack of, when using a Voodoo2 on a fast 486. I haven't got around to this yet, but seem to recall reading that the Voodoo2 was faster on a 486 compared to the Voodoo1. It would be nice to have some benchmark numbers.

The automatic 2/3 PCI multiplier is one of the main downfalls of the M919. I recall someone using the turbo switch to boot the system up with 33 MHz FSB, then switching to 40 MHz in DOS. Did you confirm that the M919 doesn't divide the VLB freq when the FSB is set to 40 MHz?

Plan your life wisely, you'll be dead before you know it.

Reply 24 of 92, by derSammler

User metadata
Rank l33t
Rank
l33t

I don't see it as a downfall. It's actually quite surprising to see such a feature in a (cheap) PC Chips board. It makes sure that the system stays within specs, which is quite cool.

VLB is allowed to run at 40 MHz. While I have not tested the speed of VLB again (don't own a card that would show the difference anyway), I know that VLB always runs at FSB speed. You can't apply a divider to it, since it is not bridged but directly connected to the CPU - so it can not run anything but synchronous to the CPU.

The PCI divider could be avoided using an Arduino. Logic would be simple: on power-on or when reset signal is detected -> open JP3A. Speaker beep after post detected -> close JP3A until reset signal is detected again. But for now, I'm building a slot bracket with a switch for 33/40 MHz FSB.

Last edited by derSammler on 2018-07-07, 06:52. Edited 1 time in total.

Reply 25 of 92, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I'd like to see a master override for the automatic 2/3 divisor, at which point, I'd not call the feature a drawback.

I did some testing which showed VLB+PCI boards having generally inferior PCI speeds compared to straight PCI/ISA implementations. For this reason, I tend to avoid VLB+PCI. However, the M919 is interesting with its fake cache history and COAST-like cache socket. Performance comparison of 486 motherboards with VLB-only, PCI-only, and PCI+VLB

I tried the 33/40 switch trick on the M919 at one point and don't recall the outcome. I sorta recall it working, but having trouble with something else and getting fed up. I also thought of a small micro-controller switch idea at some point, but decided there are far better motherboards than to put more time into this one. The easiest implementation is to use a freq. to voltage IC, but can also use an Arduino w/comparitor+interrupt counter.

Plan your life wisely, you'll be dead before you know it.

Reply 26 of 92, by derSammler

User metadata
Rank l33t
Rank
l33t

Problem with the switch trick is that the IDE controller is PCI as well and doesn't run stable at 40 MHz. You can find some reports of people having done that and got data corruption from time to time. I'm not going to hot-switch the FSB. I built a slot bracket today with a switch that allows me to manually select 33 or 40 MHz FSB without opening the case. I'm going to use 33 MHz to not stress the hardware. Only if more speed is really needed, I switch to 40 MHz, but not while the system is running. In those rare cases, I can live with the performance loss of the PCI bus.

Reply 27 of 92, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I too have run into the 33 MHz limitation of the onboard IDE controller on these boards and always run a PCI SCSI or PCI Ultra DMA controller.

Plan your life wisely, you'll be dead before you know it.

Reply 29 of 92, by BinaryDemon

User metadata
Rank Oldbie
Rank
Oldbie

That switch is great!

Check out DOSBox Distro:

https://sites.google.com/site/dosboxdistro/ [*]

a lightweight Linux distro (tinycore) which boots off a usb flash drive and goes straight to DOSBox.

Make your dos retrogaming experience portable!

Reply 33 of 92, by feipoa

User metadata
Rank l33t++
Rank
l33t++

No VLB graphics yet? It would be interesting to benchmark PCI vs. VLB on this board to see if it suffers the same performance degredation fate as SiS 496/497 boards when, both, VLB and PCI ports are included.

I'm curious about the decision to add a heatsink on the chipset?

Also, is that a memory upgrade module on the Mystique?

Plan your life wisely, you'll be dead before you know it.

Reply 34 of 92, by melbar

User metadata
Rank Oldbie
Rank
Oldbie
derSammler wrote:

S3 Virge 325 instead of a Virge/DX.

derSammler wrote:

S3 Virge replaced with a Matrox Mystique

What was the reason to go this way?

#1 K6-2/500, #2 Athlon1200, #3 Celeron1000A, #4 A64-3700, #5 P4HT-3200, #6 P4-2800, #7 Am486DX2-66

Reply 35 of 92, by derSammler

User metadata
Rank l33t
Rank
l33t
feipoa wrote:

No VLB graphics yet? It would be interesting to benchmark PCI vs. VLB on this board to see if it suffers the same performance degredation fate as SiS 496/497 boards when, both, VLB and PCI ports are included.

I'm curious about the decision to add a heatsink on the chipset?

Also, is that a memory upgrade module on the Mystique?

The VLB slot stays reserved for a 3D Blaster, which I'm currently trying to obtain.

As for the heatsink, I read from various sources that the chipset can get hot when FSB is >33 MHz. And since I had a fitting heatsink lying around, why not? 😁

Yes, that's a memory upgrade on the Mystique. It's an original one with 2 MB on-board + 2 MB on the upgrade.

melbar wrote:

What was the reason to go this way?

I want to turn that 486 into a demonstration system for early 3D cards and I think the Mystique is a better fit. Too bad I can't use the Virge and the Mystique at the same time... The last free PCI slot will get a PowerVR card.

Reply 36 of 92, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Oh. Have you ensured that the VLB slot is reliable with VLB graphic cards under stress in Windows? I personally had issues with the VLB slot on my M919 v3.4. My Diamond Stealth 64 Video VRAM (4 MB S3 Vision 968) would overheat in that slot, but not on other motherboards.

Plan your life wisely, you'll be dead before you know it.