RT2.20 multi-GPU support major RED FLAG

pmrdij · *Last edited by pmrdij; 11-26-2008 at 15:32. Reason: edited to appease Unwinder*

i just posted this in another thread but wanted to start a topic specific to the problem to gain attention to it. read below:

Quote:

OK i just traced this down to RivaTuner 2.20. this past weekend i tried out the 180.60 drivers and RT 2.20. after attempting to adjust my second GPU's clocks with RT and rebooting my system my second GPU was permanently locked to 198/1188/396. GPU-Z reported everything as 0MHz and the performance in SLI in any game or benchmark showed a huge decline from before and after.

as for steps to solve this i did everything i could think of. i tried dropping back to 180.48, trying 180.60 again and falling back to 178.24 without solution (and yes i used the current Driver Sweeper every time). even went as far to flashing the bios.

i formatted my machine last night and installed 180.60 and RT 2.20. in and out of SLI both RT and GPU-Z saw the same and correct clocks for my second GPU.

i just tried the same changes in RT with my second GPU and rebooted. the same problem has returned and my second GPU is once again running at 198/1188/396.. now i get to reinstall a second time but i am going to post this same block of info in a thread to alert Unwinder.

oh and Hybrid --> my 3DMark06 score in SLI for Mother Nature @ 1680x1050 with 2AA/16AF went from 170+FPS on avg to 70 in both instances of this problem occuring...

FYI i also, on the first occurrence, went through my Registry and stripped everything to do with RivaTuner from version 2.09 to 2.20 after about the fourth driver/RT uninstall/reinstall attempt without resolution.

- Robert (PmR)DeathInJune

Unwinder · *Last edited by Unwinder; 11-26-2008 at 08:49.*

Multi-GPU overclocking core of 2.20 has not changed comparing to the previous versions. And it is in use for months in EVGA Precision too with zero problems. So before giving major RED FLAGS please ensure that you're completely understaning the things you're doing. 198/1188/396 are standard safe/boot clocks which are used as the lowest throttle speed in case of hardware failures (including too agressive overclocking!!!). Also, as I've said these are the clocks your system is booting at, if you decide to disable 2D/3D clock separation in RT during enabling hardware overclocking, display adapter will constantly run at these clocks until you re-enable overclocking with 2D/3D clock separation enable. Finally, I don't know if you ever used SLI before, but you DON'T need to overclock the secondary GPU separately. All tools including RT and EVGA Precision sync secondary GPU clocks with it while adjusting master GPU clocks.

P.S. Seen your thread at hardforum about it. Was too lazy to register and reply right there, sorry.

Unwinder · 11-26-2008, 08:52 | posts: 10,312 | Location: Taganrog, Russia

Also, risking to see another wolf2009-like thread at hardforum, let me still recommend you to migrate to EVGA Precision. That tool is much more simple for beginners.

Dcrawler · 11-26-2008, 10:00 | posts: 114 | Location: Somewhere in America

Hey Unwinder. Not trying to derail this thread but my previous thead seems to have gotten lost in the mix and thought I'd ask your opinion here. I'm trying to use force constant performance level. You had told me to disable sli and then enable fcpl for both cards. However, when I disable sli the second card does not have this option available (cards listed as gpu 0 and
gpu 1). Your infinite wisdom is needed

. Thank you sir.

Dc

Unwinder · 11-26-2008, 14:18 | posts: 10,312 | Location: Taganrog, Russia

Remember that in case of failing accessing any driver's option via GUI you can always edit it directly via PowerUser tab. This tab was initially developed in 1998 as a tool for exploring the driver's settings directly. Go to PowerUser tab, open ForceWare database, expand <System> node, then copy GPU0 OverrideMaxPerf value to GPU1.

pmrdij · *Last edited by pmrdij; 11-26-2008 at 17:38.*

Unwinder i have been using SLI now back to the 6600GT in various systems and have never encountered this problem while using RT throughout. also have been OC'ing for 13 years and have used RT with several nVIDIA cards over such time. a few things to note:

1. if the "wolf2009" thread you speak of is what i've heard about leading to your banishing of nvnews from posting RT believe me that's not anything i wish to bring about. RT is in the top echelon of quality utilities in my books
2. one GPU is water cooled and the other has a ZALMAN VF1000 and at 625/1450/975 they're stable in everything
3. according to the Hardware monitoring/GPU-Z the same OC above was only taking on GPU0 in both SLI mode and out of SLI mode
4. this problem has occurred in both instances just by my checking the box to "Enable driver -level hardware overclocking" for the second GPU when not in SLI mode and rebooting. choosing GPU1 and checking off the box and opting to detect now just simply unchecks the box

i've got to get off to work but thanks for the heads up on the OverrideMaxPerf. haven't touched the PowerUser tab in a long time but i'll revisit it later tonight.

- Robert (PmR)DIJ

Unwinder · 11-26-2008, 14:49 | posts: 10,312 | Location: Taganrog, Russia

pmrdij

Don't touch OverrideMaxPerf please, it was the answer to Dcrawler and it is not related to your issues.

pmrdij · 11-26-2008, 15:29 | posts: 24 | Location: Portland, OR

Quote:

Originally Posted by Unwinder

pmrdij

Don't touch OverrideMaxPerf please, it was the answer to Dcrawler and it is not related to your issues.

ah, good to know. i started to go into it for the first time since the GeForce 2 days before leaving for work but didn't do anything.

by the way on my way into work i thought about getting you access into my system tomorrow morning via logmein so you can see what i am talking about if you'd care to (i say tomorrow morning because i know what time it is there at 6PM tonight here). i swear to you that this is a first for myself as well but any OC clock changes i make are not being applied to GPU1 according to RT and GPU-Z in SLI mode. in SLI mode setting the OC (which as you and i know should sync over both cards) to 625/1450/975 results in both GPU-Z and the built in RT Hardware monitoring reporting my second card as unchanged and still running along at its default values of 576/1350/900.

course now it is perma stuck at 198/1188/396.

also i have been operating on four or less hours of sleep since Saturday while trying to reverse this so if i came off sounding like an irate ranter regarding RT this morning in my posts that's my excuse.

- Robert (PmR)DeathInJune

boogieman · 11-26-2008, 16:39 | posts: 1,560 | Location: Pa

Quote:

course now it is perma stuck at 198/1188/396

I had this happen to me the other day on GPU1. Uninstalling and reinstalling the Nvidia drivers fixed the issue.

pmrdij · 11-26-2008, 17:29 | posts: 24 | Location: Portland, OR

Quote:

Originally Posted by boogieman

I had this happen to me the other day on GPU1. Uninstalling and reinstalling the Nvidia drivers fixed the issue.

sadly i've had the opposite result when it comes to uninstalling/reinstalling the Forceware drivers (and several versions of since encountering this problem). going to see if EVGA's tool does anything with this tonight but i am probably correct in it not supplying half the functionality of RT (i do more than just OC with RT).

- Robert (pMr)dEATHiNjUNE

pmrdij · 11-27-2008, 01:45 | posts: 24 | Location: Portland, OR

Unwinder --> OK so checking the box to allow "2D/3D clock frequency adjustment" and rebooting regained myself the stock speeds on the second GPU. thanks for that heads up on using that to get out of the woods. i apolize for the lack of clarity out of the gate this morning. has been a long four days trying to get rid of this problem. just an FYI --> i have been able to duplicate it a third time with RivaTuner 2.20, however, not with 2.11.

firstly let me clarify (i've had seven cups of coffee in the last five hours so this should come out better) that what i have been after all along has nothing to do with OCing while in SLI mode but rather increasing my second second GPU's speed with SLI disabled as i fold with both GPU's. with SLI disabled i am able to OC my primary card to the same speed i use in SLI mode via RT (625/1450/975). i can do this with any version of RT.

the second GPU available in the target adapter list is a different story. with RT 2.20 checking the box to "Enable driver-level hardware overclocking" and rebooting without checking the box to enable 2D/3D clock adjustments results in the second GPU being set to 198/1188/396 after reboot. not checking the box to enable 2D/3D clock adjustments in RT 2.11 does not produce the 198/1188/396 speeds after reboot (remains at stock speeds - 576/1350/900).

hope that all makes more sense than earlier today and if you'd like to delete this thread feel free. oh and EVGA's Precision Tool as you recommended is something that i wouldn't use even if being held at gunpoint vs. RT.

- Robert (PmR)DIJ

mattymaxima · 11-27-2008, 02:47 | posts: 100 | Location: Florida, USA

Quote:

Originally Posted by pmrdij

Unwinder --> OK so checking the box to allow "2D/3D clock frequency adjustment" and rebooting regained myself the stock speeds on the second GPU. thanks for that heads up on using that to get out of the woods. i apolize for the lack of clarity out of the gate this morning. has been a long four days trying to get rid of this problem. just an FYI --> i have been able to duplicate it a third time with RivaTuner 2.20, however, not with 2.11.

firstly let me clarify (i've had seven cups of coffee in the last five hours so this should come out better) that what i have been after all along has nothing to do with OCing while in SLI mode but rather increasing my second second GPU's speed with SLI disabled as i fold with both GPU's. with SLI disabled i am able to OC my primary card to the same speed i use in SLI mode via RT (625/1450/975). i can do this with any version of RT.

the second GPU available in the target adapter list is a different story. with RT 2.20 checking the box to "Enable driver-level hardware overclocking" and rebooting without checking the box to enable 2D/3D clock adjustments results in the second GPU being set to 198/1188/396 after reboot. not checking the box to enable 2D/3D clock adjustments in RT 2.11 does not produce the 198/1188/396 speeds after reboot (remains at stock speeds - 576/1350/900).

hope that all makes more sense than earlier today and if you'd like to delete this thread feel free. oh and EVGA's Precision Tool as you recommended is something that i wouldn't use even if being held at gunpoint vs. RT.

- Robert (PmR)DIJ

I mean did it EVER occur to you to check that box before freaking out?! You started the post about how you were manually setting the speed of GPU1, Dont you think that "enable 2d/3d clock adjustments" would have to be checked in ordcer to"adjust" them. BTW Unwinder RT and Precision rule!

pmrdij · 11-27-2008, 03:53 | posts: 24 | Location: Portland, OR

Quote:

Originally Posted by mattymaxima

I mean did it EVER occur to you to check that box before freaking out?! You started the post about how you were manually setting the speed of GPU1, Dont you think that "enable 2d/3d clock adjustments" would have to be checked in ordcer to"adjust" them.

obviously as you know --> no it did not occur to me so yes my fault there. the box is checked by default on the primary GPU but not the secondary (i should have caught that obviously). still this problem does not exist whether that box is checked or not after rebooting the system with RT 2.11. only with RT 2.20 does this occur when that box is left unchecked and the system is rebooted. with it checked neither 2.11 or 2.20 has any ability to adjust the clocks of GPU1 (all options are grayed out).

anyhow as mentioned prior i've had less than 24 hours of sleep while trying everything else that came to mind outside of that check box (doh) since Saturday so my "freaking out" had that involved this morning. i think i've also apologized (of which the eighth cup of coffee has helped me realize i spelled the word wrong as "apolize") for my "freaking out" to Unwinder so you can have that apology extended to yourself as well. even suggested deleting this thread...

- Robert (pMr)dEATHiNjUNE

Species1571 · 11-29-2008, 20:00 | posts: 11 | Location: Scotland, UK

So is there any way of getting back full control of both cards clocks in RT 2.20? I am having the same problem with 2 GTS 640MB, I can get the stock speed on the second card by allowing the separate 2D/3D clocks, but any clock profile is only being applied to the first card. I thought 2.20 was going to introduce independent clock speeds for each card. Did i read that wrong or is it for a future update?

Unwinder · 11-29-2008, 21:05 | posts: 10,312 | Location: Taganrog, Russia

RT supports SLI overclocking for YEARS. 2.20 declared improved monitoring and scheduling features for SLI, it was not supposed to change anything in SLI overclocking area, it didn't change anything and new versions won't change anything too. So I ask once again: guys, if cannot configure RT or if it is too complex for you - please just migrate to EVGA Precision tool. It was initially designed as deadly simple overclocking tool.

Species1571 · 11-29-2008, 21:36 | posts: 11 | Location: Scotland, UK

I must have misunderstood, in the What's new thread, you said:
- New multi-GPU oriented monitoring module and launcher allow using the scheduler to program independent dynamic gamma, fan and clock control algorithms for each physical display device installed in the system.

I know profiles applied to both cards in previous versions, I have set RT up many times and now it only takes minutes to set up after a format, but this problem of not applying to the second card is a new one in 2.20. I think I speak for myself and the other guy that we do NOT want to go to Precision. If anything, we are trying to help you get to the bottom of this problem as it may only be on certain system specs, or it may well just be a lack of knowledge on the new ways of doing things in 2.20, in which case a point in the right direction from you or anyone would be appreciated.

Thanks

Unwinder · 11-29-2008, 21:48 | posts: 10,312 | Location: Taganrog, Russia

There are no changes in overclocking layer of 2.20. Are you sure that the problem is new to 2.20 and it was working for you before on the same configuration (i.e. the same ForceWare version etc)?

Species1571 · 11-29-2008, 22:20 | posts: 11 | Location: Scotland, UK

Well I installed 2.20 and 180.48 on the same day. I will try going back to 180.43 and see if that fixes anything.

Unwinder · 11-29-2008, 22:26 | posts: 10,312 | Location: Taganrog, Russia

What OS do you use BTW?

Species1571 · 11-29-2008, 22:50 | posts: 11 | Location: Scotland, UK

I went back to 180.43 and it's still the same. This is on Vista 64. When activating the overclock profile, GPU0 goes to 612/1566/972, GPU1 stays at 198/1188/396

Unwinder · 11-29-2008, 22:55 | posts: 10,312 | Location: Taganrog, Russia

Quote:

GPU1 stays at 198/1188/396

That's NOT called "GPU1 doesn't overclock", that's called "GPU1 stalls at boot clock". Reread my first posting in this thread, it gives you all info you need to know about it. I clearly see that it is not RT issue and leaving the trhead, sorry.

Species1571 · 11-29-2008, 23:13 | posts: 11 | Location: Scotland, UK

I guess the problem is how to stop it stalling, I have allowed separate 2D/3D and rebooted with the same results.

Unwinder · 11-30-2008, 09:12 | posts: 10,312 | Location: Taganrog, Russia

When SLI mode is disabled this option is available independently for both GPUs. With SLI enabled you can only control it for the primary one.

Species1571 · 11-30-2008, 13:49 | posts: 11 | Location: Scotland, UK

OK problem solved, I didn't realise that the last driver install disabled SLI by default. So what I did is reinstall 180.48, reboot with only TV connected (for some reason after installing drivers my system defaults to TV-out only), so boot windows with only TV connected, then plug in the monitor, exit rivatuner, go to NVIDIA panel, switch display back to monitor only, enable SLI, start rivatuner, enable overclocking and separate 2D/3D and click reboot and now it works, both cards underclock to 114.75/229.5/198 at bootup and both overclock to 612/1566/972

I would still love to see fully independent clock control for both cards if such a thing is possible as I know one of my cards can clock higher than the other.

One more question, when I select GPU1 from the main screen is it normal for it to say no supported drivers detected for this display adapter?

Anyway I'm happy now, thanks again.

Unwinder · 11-30-2008, 15:19 | posts: 10,312 | Location: Taganrog, Russia

Quote:

Originally Posted by Species1571

One more question, when I select GPU1 from the main screen is it normal for it to say no supported drivers detected for this display adapter?

Yes, that's normal for Vista. Unlike Windows XP where all GPUs are always (i.e. in both SLI and multimonitor modes) represented by independent logical Windows display adapters, Vista's multi-GPU architecture is different and it assumes that slave GPUs are completely invisible to OS when multi-GPU mode is enabled. It applies to both SLI and Crossfire systems because as I've said above it is OS architecture specific. RivaTuner is still able to detect such orphan GPUs (i.e. GPUs not attached to any logical Windows dispaly device) and provide low-level functionality for them, but attributes of normal windows display device (i.e. display driver name/info, independend driver settings, attached monitor name etc) are missing for them.