Not an SLI expert: Why differences in single v double memory clock?

[BonzaiDuck]

Could be a trivial question, or someone will tell me I'm a Doo-fus.

Now running 2x GTX 970 SLI.

I explored the stable ranges of core and memory clock with a single card. Haven't applied this "experience" to the SLI yet, and may wait until I find a slightly beefier PSU.

After stressing or running a game with the single card, the core clock would settle down to an idle below 1,000, as did the memory clock.

Now, with properly configured SLI and even for using "GeForce Experience" to "optimize" the settings for a game, the core(s) as expected max out at 1,319Mhz, then settle back below 1,000 afterwards. But the memory clock only drops from 3,500 Mhz to 3,000. [DDR of 7,000 versus 6,000]

Is this to be expected? How and why does memory clock behavior vary between single card and SLI?

 

[cmdrdredd]

It shouldn't change that much. At least not in my experience. Are you overclocking at all? You can set the power target higher and it will not throttle as quickly. The memory should be constant and won't throttle based on the same criteria as the core clock.

 

[Kenmitch]

It shouldn't change that much. At least not in my experience. Are you overclocking at all? You can set the power target higher and it will not throttle as quickly. The memory should be constant and won't throttle based on the same criteria as the core clock.

Not sure but I think he meant they don't drop down to the low idle states once game is closed....Guess he'll have to clarify.

 

[BonzaiDuck]

Not sure but I think he meant they don't drop down to the low idle states once game is closed....Guess he'll have to clarify.

Here's the point of it, and thanks for asking.

When I installed the first card, I am DARN sure that it would idle with the core clock around 900+Mhz and the memory clock in the same neighborhood. In fact, under certain circumstances, I thought I saw the core clock drop as low as 300+. Since these were idle states, I didn't pay much attention to the precise numbers, but I'm SURE that I saw the memory clock at something below 1,000 Mhz.

Now . . . of COURSE . . . the spec loaded stock values are about 1,316 Mhz core and 3,500 Mhz memory. And there are general expectations that you can OC the core up to (even above) 1,500, and the memory to 4,000. (The memory is often explained as a double-data rate 7,000 and 8,000).

With the SLI configuration, the idle memory clock is now 3,000.

I'm SURE I didn't imagine the much lower idle memory clock, but at my age -- perhaps anything is possible.

I was reading these numbers from the Afterburner monitor. I think I updated the AB software, but it couldn't have been buggy enough to have a memory clock reporting error.

Put it another way. If my (own) memory serves me, I thought someone would be more familiar with these clock patterns, even for an older NVidia GPU. And it's also possible my (own) memory doesn't serve me so well.

But I'm (almost) sure that's what I saw: an idle memory clock below 1,000.

Or . . . put it another way. If y'all think that 912 Mhz core and 3,000 Mhz memory clocks are "normal" idle values, we can put all this to rest as "not a problem."

 

[kasakka]

Mine idle at 900 MHz core (I believe this has something to do with a 144 Hz display) and 3500 MHz memory. So I think it's normal or a possible driver bug. The second card memory clock is all the way down to 324 MHz on the desktop.

 

[BonzaiDuck]

Mine idle at 900 MHz core (I believe this has something to do with a 144 Hz display) and 3500 MHz memory. So I think it's normal or a possible driver bug. The second card memory clock is all the way down to 324 MHz on the desktop.

It's good to compare notes. It dawns on my that the choice of card for connection to the monitor has an impact, as do all the settings in the NV control panel So far, my cards are both showing the same identical memory clock -- and your own reported experience is different from mine: for the 2x SLI, my memory clocks remain identical.

As for bugs, I couldn't say. I just installed the latest driver release after reading somewhere that this particular version offered improvements for the memory usage. Personally, I'm not sure I can tell if it does. With SLI, I'm not sure that I even care about it. I didn't even care about it after the frenzy grew in threads here since late January.

 

[2is]

I've had my 680's downclock on certain titles for no apparent reason. Power factor and temps are all below the point that throttling should occur. What I did to mitigate this issue is go into the Nvidia control panel > manage 3D settings > program settings and under "power management mode" I set it to "prefer maximum performance" for the games that I saw this behavior in.

 

[BonzaiDuck]

I've had my 680's downclock on certain titles for no apparent reason. Power factor and temps are all below the point that throttling should occur. What I did to mitigate this issue is go into the Nvidia control panel > manage 3D settings > program settings and under "power management mode" I set it to "prefer maximum performance" for the games that I saw this behavior in.

Yeah -- I found a forum thread at Tom's just hours ago, looking for insight about these "phenomena." The focus was on that very setting, and the difference between "adaptive" and "prefer maximum performance." In that case, the poster WANTED the card to down-clock, but it wouldn't. And "Adaptive" was the solution.

Right now, I'm not even over-clocking these cards. But I was pleased with OC'ing potential on the initial card I put in the system. None of the games or bench programs caused the card to throttle unless it went over the power limit, and I had the power limit set to the max of 110%.

Between the drivers, the utilities (AfterBurner, Kombustor, MSI Gaming App), and just the hardware configuration with two cards and the SLI bridge -- I find it amazing that I haven't found any conflicts between programs bundled with the cards.

I DID have a crash with Stop Code 124 with the overclock excursion. One incident seemed to derive from pushing the core and memory just a tad too high; the other one was most likely a conflict between two monitoring programs running simultaneously. That was the "first half" of the excursion. Everything was under control with more modest clock settings.

But as I said -- it's all stock settings now . . . for the time being.

 

[Kenmitch]

Strange. My 970's will drop to the lowest idle states when not under any loads. Using maximum performance mode. Only using a single 1080p 60Hz monitor currently.

On my initial overclocking adventure I did run into the issue you describe. Seems the NVIDIA drivers aren't as solid as the devoted say.

Think I had to wipe the drivers and reinstall to fix it.

 

[BonzaiDuck]

Strange. My 970's will drop to the lowest idle states when not under any loads. Using maximum performance mode. Only using a single 1080p 60Hz monitor currently.

On my initial overclocking adventure I did run into the issue you describe. Seems the NVIDIA drivers aren't as solid as the devoted say.

Think I had to wipe the drivers and reinstall to fix it.

Same monitor spec as mine, except that I have my 1080p HDTV connected as a second monitor.

Did you mean that your memory clocks also drop below 3,000? You could be right about the drivers.

 

[Kenmitch]

Same monitor spec as mine, except that I have my 1080p HDTV connected as a second monitor.

Did you mean that your memory clocks also drop below 3,000? You could be right about the drivers.

Yep. Some ridiculous clocks like 135c/300m I think it is.

What dies it do with TV unhooked?

 

[2is]

My cards have no problem clocking down when not playing. I only set the "prefer maximum performace" option as a game specific one, not in my global settings.

 

[cmdrdredd]

Here is what my Cards idle at. Left side shows the core and memory clocks. When looking at the desktop and no web browser open it's 135Mhz Core and 324Mhz memory.

 

[BonzaiDuck]

Thanks, everyone, for the input so far.

I put in a support ticket with MSI. I'm a little peeved at myself because the monitored readings I gave them of the single GTX 970 before I added the second card were incorrect. The correct values I had observed were the same shown by cmdrdredd.

What I observe now as idle values are core 912 and memory 3005.

But I should ask this: "Who has any kind of multi-monitor configuration which shows those figures?"

I found this today at Tom's HW, dated from September, 2014 up to sometime last week:

http://www.tomshardware.com/answers/id-2311911/high-idle-temperature-msi-gtx-970.html

I think somewhere in that discussion there is mention of "multiple monitors" related to the heat issue, which I actually anticipated, but my temperatures at idle have the master card at 42C and the bottom card at 38C -- give or take, this or that. But I think the link also deals with f***y [heh- "funny"] idle clocks, and either the OP in the Tom's HW thread or someone else noted having a multi-monitor setup.

[And I should've told that to the MSI folks, too . . . ]

At this point, I could take a cue from KenMitch and reinstall all the drivers and software. Maybe it would be more productive if I could just do it with the NVidia driver. But I need to be convinced that this isn't something related to the HDTV (monitor #2) connection, use of the master card for both monitor connections, the 3D [NV Ctrl Panel] configuration, the Windows Power Options "Advanced" settings or something f***y with AfterBurner -- or maybe even the MSI Gaming App. [But I've only toyed with the App, and I make sure to reinstitute a default "Afterburner" profile after so doing.]

I guess the real question to answer is that pertaining to multiple-monitor setups. And I suppose I could simply disable the HDTV temporarily through the NV Control Panel.

I guess that's what I do next, while waiting for tech support, even more insightful comments (keep 'em coming), or some . . . personal inspiration . . .

 

[96Firebird]

If I recall correctly, the latest Nvidia cards can handle multi-monitor displays with different resolutions without having to keep the idle clocks higher than normal. On older cards (ie my 460), the clocks would be higher than normal.

Unplug your HDTV and see what the clocks do. Also, what is the resolution and refresh rate of your main monitor? I'm assuming the HDTV is 1080p @ 60Hz...

Also, you can try out the suggestion in post #2 here.

 

[cmdrdredd]

I have the ability to run a monitor and an hdmi cable to a tv to try that but I am not home at the moment.

 

[BonzaiDuck]

If I recall correctly, the latest Nvidia cards can handle multi-monitor displays with different resolutions without having to keep the idle clocks higher than normal. On older cards (ie my 460), the clocks would be higher than normal.

Unplug your HDTV and see what the clocks do. Also, what is the resolution and refresh rate of your main monitor? I'm assuming the HDTV is 1080p @ 60Hz...

Also, you can try out the suggestion in post #2 here.

Well -- here's what's happened so far. Instead of "unplugging" the TV, I simply disabled its use within NV Control Panel.

This had no effect. To be more succinct, I even changed the "multiple display mode" to "single display mode" in NV Ctrl Panel ->Manage 3D Settings. Again, no change. I also rebooted after these changes, but I don't think it was necessary.

You can continue to follow this as I answer our fellow member below . . .

I have the ability to run a monitor and an hdmi cable to a tv to try that but I am not home at the moment.

Assume you read the above. The plot thickens! I began to ponder what to do just in preparation for driver removal. So I:

-- went into AfterBurner and uncoupled the synching of the gfx cards. [No change]
-- then I went into NV Ctrl Panel and disabled SLI.

Suddenly, the core clock is down to 135, and the memory clock is ~ 324.

I'm assuming that you 2x-GTX-970 folks are running your cards in SLI, because anything else wouldn't make sense. I can also imagine it most likely that those with single cards would probably report these numbers.

But I think we're having a discussion among multi-GPU GTX 970 users who are using SLI?

If I seem slow to pick up the nuances, I thank you for your patience and please let me know.

I'm also beginning to wonder: should I perhaps uninstall Afterburner and Gaming App, reset the system to SLI, and then reinstall AfterBurner? Here I'm grasping at intuitive straws . . . .

[I also seem my temperatures have dropped from 42/37 to 32/30 (Celsius).

By the way and one more thing. My motherboard is PCI-E 3.0 capable, but I'm running a 2700K -- which isn't.

 

[cmdrdredd]

Well the pcie type shouldn't affect it. I would try uninstalling drivers and then installing them again. I am running sli as you expected and I have MSI cards as well.

 

[96Firebird]

Is your main monitor 144Hz?

Edit - Nevermind, seems you said above you have the same as Kenmitch which was 1080p @ 60Hz.

 

[BonzaiDuck]

Is your main monitor 144Hz?

Edit - Nevermind, seems you said above you have the same as Kenmitch which was 1080p @ 60Hz.

Yup!! Abso-tively, Posi-lutely! From what CmdrDredd just said, we could start our own little sub-forum club!

Well, I have some dis-installation or de-installation to do . . . like the Governator Terminator said: "Ah'll be back!"

 

[BonzaiDuck]

I suppose it could have been "multi-causation." I just say that because I had also uninstalled then reinstalled Afterburner. The thing I noticed though -- with the 347.52 driver version, I'd failed to do a "clean install" from the original driver-disc installation -- maybe version 347.25.

So when I uninstalled the latest version, it pops up showing the earlier driver after the reboot. This time around, I got it right.

135/135 core, 324/324 memory-clock -- at idle.

I reset the "Manage 3D" for Kombustor to load up both cores with "force alternate frame rendering 1."

The cards are synched in Afterburner, but I notice under load that the master card GPU1 is showing a core clock of 1316 with GPU2 @ 1328. Is that sort of discrepancy to be expected?

On another thread in the PSU forum, I think I've discovered that this 650W Seasonic PSU still has maybe 65W of extra potential above what would arise in wattage with the CPU and GPU's loaded simultaneously. Even at extreme conditions, the maximum wattage at the PSU would be 585. So I think I can push up the clocks a tad. Well . . . just a little bit -- just to "see." The temperatures seem lower and less inconsistent (top card versus bottom card, etc.)

I just . . . didn't . . . like the new "skin" that Afterburner gave me on reinstallation, but there's a helluva-lot of 'em to choose from, so no problem there!

Mmm. . . . the core clocks have suddenly stabilized at 1316 identically. . . . Probably will replace the PSU. It's just . . . too close . . . even for some "extra potential."

Thanks, guys!!

 

[Kenmitch]

Dang...The superior,bullet proof,ultimately mastered NVIDIA drivers look to be more of a myth than reality! Lotta years of FUD

Glad you got it sorted out.

 

[BonzaiDuck]

Dang...The superior,bullet proof,ultimately mastered NVIDIA drivers look to be more of a myth than reality! Lotta years of FUD

Glad you got it sorted out.

It always seemed a bit dicey with driver update installations. On my other rig, it would regularly foul up the HDCP configuration and temporarily deprive me of my TV via HDHomerun Prime. So I prevent it from updating the driver when everything works the way it is.

Also makes me nervous that you (apparently) can't just go to the NVidia site and download a zipped or *.exe driver file anymore. The "Experience" thing is like a gatekeeper.

 

[Kenmitch]

You can download drivers from NVIDIA still.

http://www.nvidia.com/Download/index.aspx?lang=en-us

 

[BonzaiDuck]

You can download drivers from NVIDIA still.

http://www.nvidia.com/Download/index.aspx?lang=en-us

I'll try the link, but when I chose "option 1" to specify "GeForce" and "900 series" etc., the download wouldn't work. Maybe it had something to do with the "GeForce Experience" installation -- I can't say.

[30 seconds later . . ] Huh. Fancy that. Suddenly, the download link doesn't seem broken.

So I'm turning this over in my mind. The UPS software shows about 85W at the "wall" (or UPS) when the system idles. Running Kombustor to load up both GPU's can push the wattage as high as 560W measured the same way, of which about 60W is CPU wattage. Supposin' I decided to run Linpack side by side with Kombustor, that would increase the package watts to 140, so add 80W to 560. If my PSU is 90% efficient, than I'm really using 640W X 0.9 = ~ 576W under that hypothetical, less likely scenario. And if I just push the core speed of the 970's up another 100Mhz, it adds another 25W.

I really think I want a bigger PSU for this, but I could also take my time. This is a 650W Seasonic. Moral of the story: If you're going to even THINK about 2x SLI, you should pick your PSU at the beginning with that intention. This one? I'd purchased for Mom's "new computer," and at 89, folks aren't eager for "change."

I guess it all keeps me busy.