The mismatched drivers are not gauranteed to work every time, and  Monster3D II is not reference compatible,so don't expect to ever get those cards working together in SLI.  I'd advise you to aquire an identical card to one of your existing ones.  I have a Creative Lab's 3D Blaster II that won't work with reference boards in SLI.  This is commonly the case with Diamond/Orchid/Creative Voodoo2's.

You won't fry your voodoo5 with a BIOS setting.  If you have a 3.3v AGP 1.0 voodoo5 5500 AGP like most Voodoo5's, then you can only install it in AGP 1.0 and AGP Universal 2.0 motherboards.  The slot keying prevents incorrect installation.  A 2.0 Uni. board is what you have, this is basicaly an AGP 1.0 and 2.0 chipset combined.  So it can support your Voodoo5 5500 3.3V AGP.  If you set the AGP bus to 4x clocks per cycle the interface will automatically fall down to 2x when it detects that a 3.3v device is installed.  The next problem is the nature of the AGP bus, which doesn't support devices like the Voodoo5 (single board SLI), so it falls back to PCI operation, just double the speed.  The aperture is also useless as the Voodoo5 does not support Direct Memory Execute mode.  In short:  you can set 4x AGP and a huge aperture from CMOS setup, but you're wasting your time.

QuoteOriginally posted by koolsmoky

Theory, eh? That's a good one. I look forward to your contribution, perhaps to Linux? I see you are using Mandrake.

-KoolSmoky

I'm pretty certain that Linux already sets the PLL through pllCtrl1 anyway, so no fixing is required in that case.

I may work on a simple overclocker in C or Pascal if I have time.  No promises there though.

QuoteOriginally posted by koolsmoky

Want to make OC Tools, eh?

The overclocking is achived by programming the PllCtrl1 register of all VSA100 chips. Not a registry thing. There is clear improvement with fillrate. Try running GLExcess in 640x480 32bpp mode.

The reason why V.Control only overclocks the first chip is because I had assumed that the slave chips were always snooping the master, which wasn't the case with Windows Desktop mode.

-KoolSmoky

That was all I wanted to know, thanks!

When I first ran some Q3 timedemo's with pllCtrl1 set effectively to 183MHz (via a bios flash of course) I could tell that this made a difference, as compared to normal software overclock, which made no difference in fps.  This is good news indeed and it's good that this has been confirmed and solved.  Well done.

@dborca: True that Oc'ing is hardly worthwhile, I only recently coolmodded my Voodoo5 due the thermal epoxy being burnt up badly and the card overheating, (I had never OC'd it either).  The stock cooling was never made to last though.

@amp_man: Exactly what I meant, the simple theory or the register/variable used, not the full source code. Hehehe...

QuoteOriginally posted by koolsmoky

I've already sent a vsa100 oc tool to Hank. Sit tight and wait for the results.

-KoolSmoky

I, and quite a few other posting in this topic, Would be interested to know how this was achieved?  Unrevealing binaries disclose nothing of course...

Quotethis explains perfectly why a soft overclock doesnt show up in fillrate gain like expected, while a bios overclock does.

Thats sort of what I though initially...

Quotebeta, i can't confirm the edit part of your message. i have the bios modded to 178 mhz, that gives me a multitexturing fillrate of 630 mTexels in 3dmark2000. i "underclocked" back to stock 166mhz with vcontrol and got 585 mTexels.
take this as 100% and do a quick calculation, voila 108% of clock gives 108% fillrate.
if i overclock via vcontrol up to 190mhz i get 635mTexels, same as with everything else higher or equal than 181mhz, maybe my bios settings are 181mhz, dont remember excactly.

Have you tried Quake3 benchmarking?  See if you get the same sort of result.  Direct3D is a different subsystem and SLI AA is handled differently when using this API.

Quotea thought: why would they use an extra PLL for the secondary VSA100 and the other half of ram? i see no reason to do so.

Due to it being effectively another "card" rendering alternate scanlines, this device and it's memory is only actually used when Direct3D or Glide are initialised the rest of the time it is idle.

Quoteedit: checked my bios, it clocks to 178,9mhz so i dont know why the fillrate improves till 181mhz

eh? it's late...

-Edit: wow it must have been...

Anyway with a pllCtrl1 of aprox 182.8 the card shows a definate SLI mode fps gain, at 182.5: lower fps than at 166??!

Overclocking Voodoo5's with "Grxclock", like that is a waste of time: http://www.3dfxzone.it/enboard/topic.asp?TOPIC_ID=967

Try running your benchmarks with FSAA 2x enabled or use single chip mode (SLI disabled).  I guarantee you will see a small but noticable performance differential.  Under SLI you will see virtually no difference whatsoever because only one VSA-100 is overclocked.

2x/4x AGP is pointless as DIME is not supported.  1.5V slot compatibility is what was discussed intitially in this topic, not 4x clock transfer mode which won't function with SLI unless an AGP bridge is onboard anyway.

This does apply to Voodoo5, changing tghe value of "Grxclock" only OC's one of the VSA-100's, this makes a difference to FSAA performance but under normal operation does not effect framerate.  In order to OC a Voodoo5 you need to hack the BIOS "pllCtrl1" register and reflash to give the desired core/memory clock speed for both VSA-100's.  This does work.

-Edit: After further testing I'm convinced that flashing the BIOS won't do it either.  The second VSA-100 and it's 32MB of SDRAM (The SLI slave) doesn't seem to have it's speed set by the firmware or software, it seems to locked at 166MHz, which is obviously the stock hardware clock output.  The BIOS must be solely for the SLI master chip.  SLI must be enabled in software, (drivers).  In other words when you're booting up, in safe mode, DOS etc, SLI is not functioning.

The files and the forums are up and running.  The forums have moved, I'm not posting a link as I respect these people and their board.  Google search if you want to find it.

Hmmm, no idea about this game, we covered it 'you know where' some time ago... but that's another story of course.

The 5000AGP is a sample.  Note the use of the 1.5V/3.3V Universal edge connection PCB.

V5 5500 & V4 4500 sys' ~ Mandrake Linux 9.1

No the 1.5V key as well as the 3.3V key indicates AGP Universal 2.0.  I have never seen this problem where you plug in the Voodoo5 and your mobo goes into meltdown.  The slot will simply not allow you to istall the card.  The only exception being the AGP Pro slots hwere people have somehow manged to install none pro cards incorrectly!?  Hence the yellow warning label on Pro slots.

Anyone exchanging a Voodoo5 for a Voodoo4 is making a terrible mistake.  I have both cards and I can tell you the Voodoo4 is a complete con.  And was never worth the money.

Only the AGP Universal Voodoo4 and 5 PCB's have soldering points for DVI or S-Video.

V5 5500 & V4 4500 sys' ~ Mandrake Linux 9.1

QuoteOriginally posted by amp_man


cool beans! in other words, the V5 (except for the MAC, forgot to include that, oops) was not available with the DVI connector, it had to be added. and, as I stated before, AGP 4x will not work, or at least if it does will have damn near no effect. i guess the only thing you corrected me about was the 1.5V support. sorry lecram, you know how much I hate to argue with you

when the sun goes down, the music turns up...

Partly correct, the 2x and 4x AGP rates don't function on the VSA-100 SLI config in general as it does not have a bridge controller, that is whatever signalling is used.

The DVI connector does not even have a location for installation on the 3.3V AGP 1.0 Dual SLI PCB, but on the one in the first post (1.5/3.3V AGP Universal 2.0 Dual SLI PCB) it does, just like the Voodoo 4 4500 AGP PCB, which also has a location for an S-Video package in addition.

V5 5500 & V4 4500 sys' ~ Mandrake Linux 9.1

The images in the first post show the Voodoo 5500 AGP 2.0 1.5/3.3V with universal edge connection.  That PCB is the direct counterpart to the standard Voodoo 4 4500 AGP.  It is The PCB that the Voodoo 5 should have had.  It will still only run as a PCI 66 device though.  It also facilitates LCD though the package is not installed unlike in the third post where the bracket has been removed and a LCD package soldered in place.

V5 5500 & V4 4500 sys' ~ Mandrake Linux 9.1

QuoteOriginally posted by Black_OutFact: Voodoo 5 - 64MB total memory - shared between both TMUs, capable of texture compression.

I believe you are confused, the single VSA-100 chip has 2 texture unit's, the Voodoo 5 is simply 2 VSA-100's in SLI on a single PCB, this is 4 TMU's in total.

And still the 1.5V/3.3V signalling confuses.  The SLI AGP 'Napalm' config will not allow true AGP operation.  This is dictated by the Intel AGP 1.0 specification.  Where the master chip cannot be assigned the card defaults into PCI 66 mode.  I will not run in 4x AGP or even 2x AGP.  And all this is largely pointless as execute mode is unsupported as has already been said.  Any signalling voltage modification would be purely for compatibility reasons, not to make the card miraculously go into 4x AGP DIME transfer.

The Voodoo 5 5500 AGP was mounted on the 3.3V AGP 1.0 spec PCB for this very reason, 3dfx shipped it without the bridge controller and knew full well it would only run as a PCI 66 device, so why bother with a 1.5V/3.3V AGP 2.0 (1x, 2x, 4x) Universal PCB like that of the Voodoo4, it would never run as 2x let alone 4x... so as a result they did not.

Also note: In 3dfx tools (or your frontend) you can set the AGP mode to ('SSTH3_AGP_XRATE') 2x rate but it does nothing.

V5 5500 & V4 4500 sys' ~ Mandrake Linux 9.1