FX5900XT Vmod, anyone?

[XanderF]

So, hey, anyone get pics of the vMem mod yet?

And, making sure I've got this, I can read the voltages at the following points:


Right? Can I just measure from the cap itself to ground, or do I have to match through one of the pins going through the PCB to measure from?

(As you can see, I did get the heatspreader off without major catastrophe. That thing is a pain to get off - much harder than a Radeon shim.)

4bit colors etc at some point it gave some 'no enough resources' error.

That's exactly what I've seen every time when using a 5950 BIOS on my 5900XT - and I've tried nearly *all* the BIOSes available. Including the BIOSes for the MSI 5950 VIVO. You'd think, if any of them where to work on my MSI 5900XT VIVO, it would be that one.

In any case, reading about a couple of the other 5900XT BIOSes here - Leadtek? Point-of-View? - has me wondering about using one of them. Lower timings for higher ram clocks? Sounds interesting!

Where could I find a PoV BIOS?

[Garrett]

Here you can find lots of bios files
http://www.mvktech.net/download.php

[XanderF]

Right, found that site.

But they don't list a 5900XT BIOS for Leadtek - just the LX (not the same thing, according to Leadtek's site) and the regular 5900 non-Ultra (3 BIOSes for that!)

[XanderF]

As to the caps, I'm guessing I measure from the pin sticking through the PCB for the '+' side. Only thing is....how can I tell which is the '+' pin?

If I'm looking at the board side-on (it's mounted in my mobo right now), which pin do I measure from? The left one, or the right one?

[XanderF]

Hmmmm....further research would seem to indicate that the 5900LX is near-identical to the 5900XT, but with a 10mhz higher *core* clock - ram is the same!

Is this the BIOS Macci was using to get such higher memory clocks?

I got my Albatron mem from 803MHz (minor artifacts) to 911MHz (100% clean, will bench at 936 easily) by removing the small ramsinks and flashing Leadtek 5900XT BIOS in.

Or, when he said 'XT BIOS', he *really meant* the XT BIOS, not the LX. In which case, I'm still searching. I'm about the same right now, no vMod (well, I have them on, just haven't adjusted them yet until I figure out how to read the voltage), improved the ramsinks, but hitting a brick wall at just over 800mhz ram.

[Garrett]

Here's the XT bios from Leadtek that I got from an acquaintance:
http://home.planet.nl/~dikke332/NV35332.ROM

[XanderF]

Hmmm....did that BIOS help you with your clocks any?

I did it, card booted into Windows okay - driver actually worked this time - but no improvement at all. Wasn't any *worse*, just didn't improve any.

Dunno.

Also tried flashing to the 5950 again (I had heard on some forums people who couldn't go directly from their brand of 5900XT->5950 had luck flashing to another type of 5900XT, THEN to a 5950). Didn't have any luck with that, either. Still got that 4-bit color display and 'could not allocate enough resources' or whatever in device manager.

Trying the Point-Of-View BIOS now.

[HaLDoL]

Does changing the VR affect your vcore much? I mean between 4.8 and 4 Kohm my vcore barely changes (+0.2V).

[XanderF]

Still haven't quite figured out where *exactly* I'm supposed to measure the vCore *from*.

Left pin from the cap, or right pin?

In any case, the P-o-V BIOS didn't help at all. I'm quite stuck at 465/800 until I turn up my vMods.

Odd that my max core overclock dropped 10mhz by removing the heatspreader. Not sure what's going on there. Wish there was some way I could get a thermal reading of the core. It seems stable and artifact-free, it's just thermal throttling lower now.

[HaLDoL]

That's what happened to me too, I removed the heatspreader and performance was worse.
Before I could do decent benches with 606 on the core, but now that the heatspreader is removed I can only do 580 without the core kicking back to lower frequencies.

Still the max overclock has increased with the heatspreader removed, it can pass the test at 630 core, but at this speed it immediatly throttles back to lower frequencies in 3dmark.

The thermal throttling thing is really annoying, benches run without artifacts and still it throttles back. Is there a way to avoid them? Other drivers didn't work.

The voltage reading point is here:
http://members.chello.nl/~h.prijt/59...adingpoint.jpg

[XanderF]

I got that that's the cap to read, I just don't know how to read it. Which pin from it going through the PCB do I read? Left one or right one (looking down on card when installed in PC)? Does it matter?

I would think the '+' side....but which one is that?

Or, do you mean measure from the tiny pad visible from under the cap? That'd be damn hard to get at while running.....

[Garrett]

Ok, look at the cap above the one the arrows points to...
See the small metal square at the left of the cap?


Look at the small white arrow, that's the measuring point

In my case I could measure all 3 little squares near all 3 caps, didn't matter which one I chose

[HaLDoL]

it's pad where the code (c189 on mine) is next to. It's not under the cap it's next to it. Like in the picture.

[XanderF]

Cool, thanks.

As to the removed-heatspreader....

Was playing around with it more, and, sure enough, it does let me test higher in the control panel now. I can test up to 525mhz. I can even run at 500mhz with thermal throttling not really kicking in.

Previously, I had been testing with my memory at 800mhz - one test failed, moved setting down. My memory is not *quite* stable at 800mhz. Works fine in-game with no artifacts, but it won't test okay all the time. Taking it down to 780 for testing allows me to really play with the core speed a lot, then I can always use RivaTuner to kick the mem back up to 800.

It's too damn bad we can't disable thermal throttling altogether - that's a VERY annoying function.

Looking at Futuremark's ORB....it seems *all* the high scores on 5900XTs are using the Point-of-View cards. Maybe that card's BIOS has more lax thermal throttling? Supposedly, it also has more lax memory timings, but trying both PoV 5900XT BIOSes from this site yielded no memory increase at all.

Will have to try this evening RE: thermal throttling with other BIOSes. Looks like I have an opening from 4-7pm PST where my PC won't be recording anything, so I'll try then.

If anyone else wanted to experiment some and report back....

[XanderF]

Okay, got the vCore, thanks.

Still need to find the vMem location. The cap previously identified as the vMem obviously isn't:


I can't probe around on the card in my PC (too tight a fit), and soldering onto those pads is a total . Any assistance in finding the vMem read point would be appreciated!

[HaLDoL]

Weird stuff ...
I can bench without artifacts with 575-6xx Mhz on the core. If i try 570 or below, i get artifacts.
A possible explanation is that 570 is the real limit of the card and if you set it to something above it, it can pass the test but due to thermal throttling it probably goes below 570.

[ALT-F13]

HaLDoL
no, your theory of card limit is wrong. LaikrodiZ benches on his Sparkle @ 605/965 without problems... Vgpu = 1.8, no Vmem mod still - can't find any CLEAR guide...
Clocks are "real", without thermal throttling - he got 23200 yesterday on it. He didn't upload because that was still on watercooled CPU @ 3700... Just need to put it back to Mach II GT, that's ll be 4.3GHz

[HaLDoL]

Sure he can bench at 605, I can bench at 615 but I got the best scores at lower frequencies, 606 (3dmark03) and 580 (3dmark 2001).

I read somewhere that due to thermal throttling, the core frequency falls back to 2D speeds. A possible fix would be setting your 2D speeds as high as your 3D speeds but as your 2D setting uses 0.2V less than in 3D, that would be very hard to accomplish.

[ALT-F13]

Well... Now he can bench @ 610 with 1.9V. And scores go up until test freezes.
That's 137.3 nature and 23522 overall - http://service.futuremark.com/compare?2k1=7692483, current #1 with FX5900XT.

Thermal throttling could ber nicely monitored on-the-fly by voltmeter.

[XanderF]

Dunno how useful this will be, as I'm the only one who didn't seem to get it, but...

Some voltage points on a 5900XT:


The 3.00v is controlled by the second ISL chip, the 1.30v is controlled by the first. IE., 3.00v is vMem, 1.30v is vCore. These are the '2d' voltages. vMem is not changed going from 'safe' mode to '2d' mode to '3d' mode. vCore is -0.10v in 'safe' mode and +0.20v in '3d' mode.

[HaLDoL]

2.92 ohm gives me 2.73 vcore in 3D, it helps getting more consistent benchmarks and less (thermal) throttling.

XanderF: nice pic, you're gonna try a vmem mod?

ALT-F13: how about 174.2 in nature @ 580 core?

[TodB]

Originally posted by HaLDoL
2.92 ohm gives me 2.73 vcore in 3D,

uuummmm... WHAT?!?

I really hope, that you ment 1.73V...

[XanderF]

Tried vMem mod. Running it up to 3.2 didn't help me 1mhz on mem. Leaving it at 3.1. I suppose it did help a little - now my memory *always* tests okay at 800mhz, whereas at stock it would sometimes fail the test.

No artifacts in either case, so...dunno.

Card as it looks now:

[Garrett]

Hmmm... after vmodding my GPU I put a small Tweakmonster heatsink on the voltage regulator chip just to be sure... the chip you use to solder the GPU vmod on that is...
It normally runs quite hot you see, can't cool that sucker enough

[XanderF]

It's doing okay with a 1U HSF I put on it now:



Dropped the OC a *little* from a waterblock. Stable at 530/800, 3dMark03 score of 6809.

I'm done playing with it now, though - better things to do. I'd *like* to post a sale of it on these forums, but I don't have a 100-post-count, and I don't feel like spamming the forums up some, so....feh.