D901C SLI Performance Issues

I finally got around to "trying" SLI on my D901C, but with a catch, i'm using a 9800m GTX and a Quadro FX3700m from an HP and even more so, I don't have another D901C heatsink, so I am using a D900F heatsink.

I copied the VBIOS from my 9800m GTX and flashed it to my FX3700m, all was good, BIOS shows 2 GPU's that match

got into windows, installed drivers, all good, enabled SLI, all good, it boots up warning me the bridge is missing(as expected) so next i wen to run games... it's really weird HIT OR MISS if the SLI works or not, which i attribute to the old driver(since they haven't updated it in like years?) so alot of my SLI profiles are out of date. In games like HL2, Rocketleauge, I have noticed a MAJOR performance increase, in other games like World of Tanks, Xblades, even 3dmark, I've noticed significantly worse performance

here's a non comprehensive benchmark(i should probably run vanatge?)

http://www.3dmark.com/compare/is/4066802/is/4066421 (it says SLI ON but i disabled SLI in that first run by switching the GPU count to single in the nvidia profile)

when running games in SLI that get worse performance, the GPU's are under about 50% load each, as if something is holding them back(SLI bridge?) but there's others who ran Unbridged SLI with fine results, better than single card, what am i doing wrong?)

Also my FX3700m with the D900F heatsink gets VERY hot, cannot tell if it's a heatsink issue, thermalpaste(used poor quality stuff) or if it's card based(extra cores?)

 

We had a customer purchase a P570WM with a SLI configuration some years back (I forget the card; I want to say a 7xxM or 8xxM) and run World of Tanks with the same results; single card = great performance. SLI = poor performance.

Multi-GPU configurations have a history of being hit or miss with older titles, as not all older titles were developed to accommodate multiple graphics cards. I.e. there may be nothing technically wrong here.

How hot? (How many degrees Celsius?)

 

9800m GTX 70c under load
Fx3700m 86c under load

 

You could try swapping the card positions to see if it's a heatsink or a card thing impacting the temps.

 

Honestly if i wasn't out of thermal paste, i'd try it right now

also i am unsure if the FX3700m cross flashed will even post to the LCD or not >.>

 

The cards are obviously old so alot of SLI problems are outdated profiles from the driver, is there anyway to import the current profiles to my system using NV inspector?

 

I though profiles were mostly card agnostic?

 

my apologies meaker, idk what that even means

update on the heat issue, the D(00F heatsink is offset 2mm so the what i assume are chokes at the top of the card collide with the heatsink and cause it not sit on the core, so i am grinding slots into it to fit the FX3700 properly, also my MXM bracket is not the normal bracket, it's requiring longer screns and does not have the studs so it can be over tightened, need spacers

 

Ack, that's annoying. You could look out for a sager x-bracket instead too.

 

If you're not using a bridge, then it's possible that the games need to transfer too much data when they perform badly. You're using PCI/e 2.0 at best, unlike later bridgeless SLI configurations which use PCI/e 3.0, etc. You can't compare one bridgeless run to another unless the configuration is the same. The games that are getting worse performance I'm almost sure are feeling the lack of a bridge.

SLI is on system-wide no matter what you do. Your firestrike may have been on one card, but since SLI is still "on", it would detect that SLI is indeed "on". You would be required to either disable SLI system-wide, or use nVidia Profile Inspector to disable SLI via NV API for a certain program. That program will likely see that SLI is totally off, and it will kill any problems that are given to a program by SLI being on and forced to single GPU (regardless of how rare such a problem may occur).

That's a trait of negative scaling. Scaling & utilizations are different. You could have 99% utilization on both cards and only get 20% scaling (60fps up from 50fps, let's say). The maximum scaling for SLI is around 95%, give or take. It averaged 85-95% up until 2013 in games (from my testing anyway) but now even 70% is considered great, and as low as 50% is still considered good. Not by me, but by most everyone else.

 

Except there's another thread with someone doing this exact same thing(Same Laptop), and they had a better score. Alot of games getting worse performance with a lack of a bridge on such old hardware would be kinda silly since people reported the GTX 280m on SLI bridge less in the m 17x r1 was 1fps difference between a bridge and no bridge, so it wouldn't make sense.

I cannot add a bridge anyways.

SLI in the second run was off, I verified this by GPU 1 having a Zero load condition and entering L1 state while GPU2 was at 99%. Switching SLI on gets 55-60% usage per GPU in ice storm and cloudgate. And a much lower score even though in other games the SLI works great.

The point again is some games went down in performance, but changing the SLi profiles fixes quite a few games. Some I cannot fix.

 

Do these people have the same setup? 9800MGTX + flashed Quadro? If that's the case then maybe try reinstalling your drivers (I don't remember seeing you say you did this).

Ah, due to quadro having no bridge... right.

You don't seem to understand. I'm saying:
Selecting "Single GPU only" in nVidia Control Panel = SLI on, one card used for game
Disable SLI system-wide = SLI off, one card used for game
Disabling SLI via NVAPI with nVidia Profile Inspector = SLI on system wide, SLI off for program in question, one card used for game.

My second card is broken, so I have SLI off, but look at the first setting that's 0x0000001:

That's a driver-level disabler without disabling SLI system-wide. It can be useful for those games who cannot get SLI working for you.

 

The bridge is quite heavily used so those games not scaling might be hitting bandwidth or even driver issues where it expects the bridge to be there.