Checking temperature of GTX670 (6xx) Cards in OS X?

[sarthak]

I've been playing Batman: Arkham City, Dirt 2, F1 2012 and many other games for hours and hours at a time. Using a GTX 670 card but I don't seem to hear the fan spin up. I am running on Mountain Lion with a non EFI card. I hope that playing games in OS X is not causing the card to boil up to 90c or higher, enough to cause damage.

In Windows, the card used to spin up to maximum RPM after just a few minutes of gaming. Is there something going on with OS X? How can I check the temperatures of my GTX 670 non-EFI in OS X? I tried Zeus but it only shows the temp for my GT 120.

 

[Inconsequential]

I wouldn't worry.

At the end of day the BIOS controls the fan speeds.

I think the answer is no you cant currently however.

 

[sarthak]

I wouldn't worry.

At the end of day the BIOS controls the fan speeds.

I think the answer is no you cant currently however.

I see, in that case I might have to look into a IR thermometer gun (~$20 at Canadian Tire) to see if I can meter the temperature through the vents of the card. Just to be on the safe side.

Although, since I since you can't control the fan speed in OS X like you can in Windows (using EVGA Precision or other software), there is no point in worrying about it.

 

[Tesselator]

You can see how many watts it's drawing using "Hardware Monitor" that should be a good derivation on temperatures. Aside from that there are USB and stand-alone thermometer probes and sensors you can get for about $10 and up...

 

[sarthak]

You can see how many watts it's drawing using "Hardware Monitor" that should be a good derivation on temperatures. Aside from that there are USB and stand-alone thermometer probes and sensors you can get for about $10 and up...

Interesting. I use the CyberPower 1350AVR LCD and it gives me a reading of 320-400w draw when gaming or about 170w with light weight use.

 

[Tesselator]

100% of the power consumed is converted to heat at 100% efficiency. Electrical watts consumed equals heat watts produced.

Heat watts x 3.41 = Btu/hr.
Or if you like Kcal:
Heat watts x 0.86 = Kcal/hr.
Ambient gives you the differential.
And then I think there are tables on-line for the differential multiples of the Btu/hr/ft^2 or the Kcal/hr/m^2

The apendix in this doc helped me one: http://www.rittal.us/literature/download.cfm?id=622 or http://www.rittal.us/literature/download.cfm?id=622&SBU=03&action=view to view in browser. And intel has some good information in trying to figure out what that means in terms of part temperature and required air flow for maintaining spec.

At 400 watts your GPU fan should probably be racing. How fast are the MacPro fans (specifically the PCI fans) rotating?

Your fingers can be good indicators as well.

Careful... I've smoked a few GPUs in the past. As you know they NEED rapid cooling!

 

[DanielCoffey]

The 6xx series are supposed to be more energy efficient than our old 5870s and should run pretty cool.

----------

Interesting. I use the CyberPower 1350AVR LCD and it gives me a reading of 320-400w draw when gaming or about 170w with light weight use.

That will be for the entire system I suspect, not just the GPU. A 670 should be well under 225W at max load.

 

[m4v3r1ck]

You can see how many watts it's drawing using "Hardware Monitor" that should be a good derivation on temperatures. Aside from that there are USB and stand-alone thermometer probes and sensors you can get for about $10 and up...

Interesting! Could you provide a link?

 

[Tesselator]

Interesting! Could you provide a link?

Here's one - but you would need to remove the (probably copper) probe from the cap and hammer into shape in order to touch whatever part you're interested in (shrink-tube for insulation!). Ask before you buy to make sure it can record temps of 100˚C or whatever you are monitoring max operational rating is.
http://www.ebay.com/itm/Flow-scout-...132?pt=LH_DefaultDomain_0&hash=item27cc2c0794

This one tells airflow speed and temperature but only goes up to 40˚C
http://www.ebay.com/itm/LCD-Digital...475?pt=LH_DefaultDomain_0&hash=item4170d201e3

Here's one that goes to 70˚c but I don't like the probe:
http://www.ebay.com/itm/LCD-Thermom...116?pt=LH_DefaultDomain_0&hash=item46028e6ccc

Here's on in a bezel form factor with 4 probes and 4 fan pots.
http://www.overclockers.co.uk/showproduct.php?prodid=BB-003-LL

Here's a USB one with a nice range (40˚ ~ 120˚C):
http://www.ebay.com/itm/PC-Laptop-U...141?pt=LH_DefaultDomain_0&hash=item1c30403f35


etc. there's lots more!!!


Also I dunno if there are any aux shunt on the MacPro motherboard or not but if there are then just a a $2 flat probe from your chosen sensing point to the that shunt and Hardware Monitor will likely pick it up and monitor it.

The IR thermometer gun MSeth mentioned will work for this kinda thing too - but usually needs to be held.

The 6xx series are supposed to be more energy efficient than our old 5870s and should run pretty cool.

----------



That will be for the entire system I suspect, not just the GPU. A 670 should be well under 225W at max load.

Oh he didn't do his subtraction?

But if he's concerned about the fan speeds based on a comparison of it's sound in another system then there might be a failing fan in the mix. And that's a potential disaster if left unchecked etc. That's how mine were fried on the smoked ones I referenced above.

 

[m4v3r1ck]

Here's one - but you would need to remove the (probably copper) probe from the cap and hammer into shape in order to touch whatever part you're interested in (shrink-tube for insulation!). Ask before you buy to make sure it can record temps of 100˚C or whatever you are monitoring max operational rating is.
http://www.ebay.com/itm/Flow-scout-...132?pt=LH_DefaultDomain_0&hash=item27cc2c0794

This one tells airflow speed and temperature but only goes up to 40˚C
http://www.ebay.com/itm/LCD-Digital...475?pt=LH_DefaultDomain_0&hash=item4170d201e3

Here's one that goes to 70˚c but I don't like the probe:
http://www.ebay.com/itm/LCD-Thermom...116?pt=LH_DefaultDomain_0&hash=item46028e6ccc

Here's on in a bezel form factor with 4 probes and 4 fan pots.
http://www.overclockers.co.uk/showproduct.php?prodid=BB-003-LL

Here's a USB one with a nice range:
http://www.ebay.com/itm/PC-Laptop-U...141?pt=LH_DefaultDomain_0&hash=item1c30403f35


Also I dunno if there are any aux shunt on the MacPro motherboard or not but if there are then just a a $2 flat probe from your chosen sensing point to the that shunt and Hardware Monitor will likely pick it up and monitor it.

The IR thermometer gun MSeth mentioned will work for this kinda thing too - but usually needs to be held.

Great info's THANKS!!!

 

[sarthak]

The 6xx series are supposed to be more energy efficient than our old 5870s and should run pretty cool.

----------



That will be for the entire system I suspect, not just the GPU. A 670 should be well under 225W at max load.

In my experience with gaming in Boot Camp, running the card with heavy load for about 20 minutes would cause it to reset. The video signal would cut off and the fans would spin up to max until the card was cool enough to let the system continue. Hence, I used Precision to add a fan curve so the GPU temp never reaches above 60c.

Oh he didn't do his subtraction?

But if he's concerned about the fan speeds based on a comparison of it's sound in another system then there might be a failing fan in the mix. And that's a potential disaster if left unchecked etc. That's how mine were fried on the smoked ones I referenced above.

I can control the fan just fine in Windows.

At 400 watts your GPU fan should probably be racing. How fast are the MacPro fans (specifically the PCI fans) rotating?

The fans inside the Mac Pro are almost at idle speed.

 

[m4v3r1ck]

Is there a way to messure powerloads of the onboard PCIe connectors connected to PC graphics cards e.g. GTX 670 FTW?

 

[sarthak]

Update: After doing some testing with Unigine Valley, it does appear that the GPU fan kicks in immediately. I am unsure if it is load driven or temperature driven but running Valley does cause the GPU fans to kick in (audible).

When valley is run, the following occurs:
THERE IS AN AUDIBLE FAN NOISE.
PCI fan is ~1500RPM
BOOSTA is ~1250RPM
Exhaust and Intake ~1000RPM

When SMCFanControl is run with the same 1500/1250/1000/1000 RPM settings the AUDIBLE FAN NOISE IS LESS and DIFFERENT. I typically use Fan Control by Concorde Rules but for testing all fans I used SMC.

Thus, when running Valley, the system fans are ramped up but the GPU fan also kicks in.

Now the question is why do the fans not kick in when playing games for hours at a time?

 

[Asgorath]

Now the question is why do the fans not kick in when playing games for hours at a time?

Because the games you are playing are not stressing the GPU? What games are we talking about? If the GPU is fast enough that the CPU can't keep up, then the GPU will be running at less than 100% utilization and thus the fans won't necessarily need to be running at full speed.

 

[sarthak]

Because the games you are playing are not stressing the GPU? What games are we talking about? If the GPU is fast enough that the CPU can't keep up, then the GPU will be running at less than 100% utilization and thus the fans won't necessarily need to be running at full speed.

Batman Arkham City, Dirt 2, F1 2012, X-Plane 10, plenty other OS X games. All games (except XP10) have all (every single) setting at maximum at 2560x1440. The CPU is not doing much work in most games, it's typically around 30-50% utilization.

The same games maxed out in Windows would cause the GPU fans to kick in.

 

[Asgorath]

Batman Arkham City, Dirt 2, F1 2012, X-Plane 10, plenty other OS X games. All games (except XP10) have all (every single) setting at maximum at 2560x1440. The CPU is not doing much work in most games, it's typically around 30-50% utilization.

The same games maxed out in Windows would cause the GPU fans to kick in.

Do you get comparable framerates between Windows and OS X? You can check your GPU utilization using Apple's "OpenGL Driver Monitor" tool, which is part of the Xcode graphics tools package. NVIDIA exports this as "GPU Core Utilization". When I'm testing the perf of my OpenGL apps I typically enable that statistic and the "CPU Wait For GPU" to get a sense of how busy the GPU is, and how much the CPU is stalling.

 

[Tesselator]

The fans inside the Mac Pro are almost at idle speed.

Hmm, well it's hard to say from sitting here in Japan but this really doesn't sound good to me.

• Mac's fans almost idle (500 ~ 900 RPM),
• The Card is at full draw according to the UPS (approximately),
• The Card's fan doesn't rev to the same as it does in Win7, and stays very quiet comparatively - give similar loads,

I suggest downloading Mac Unigine-Heaven-3.0 or Mac Furmark, removing your case's side panel, and letting it run through the demo for 15 to 30min and see how hot the card is getting to the touch. It should get hot enough to be painful when you momentary touch it (like a large pot of boiling water 30s after you take it off the stove) but not hot enough to melt stuff like wires and so on.

I think you should be able to tell by common sense. One will seem like gosh that's very hot and the other will be like holy cow that might start a fire.

If it's the later you should hurry up and figure out how to solve it. The Mac's PCI fans under SMCFanControl can do quite a lot. Maybe rex those to 1300 or 1400 and don't work the card hard till you figure it out.



===================================================================
EDIT:
Sorry, I read this later:

Update: After doing some testing with Unigine Valley, it does appear that the GPU fan kicks in immediately.

Oh, OK, you're fine then...

Now the question is why do the fans not kick in when playing games for hours at a time?

It's just not heating it up is all. AFAIK it all works from thermostatic readings. It doesn't care about load or rates or anything... just temperature. So if the fans rev up in one circumstance then this shows the fans and sensors are working and you can go back to not worrying - even marveling at how much quieter OS X is than Windoze...

BTW, I play some games that look awfully heavy yet the 8800GT's fans barely increase at all... QuakeLive at 1920x1080 with everything set to high or ultra high, and all post processing turned up to max. CPUs system wide are less than 20% (usually around 10%), and the 8800XT's fans are not audible over my SMCFanControl setting of 1300. The card draws 90 to 95% of it's max power specification according to "Hardware Monitor". Furmark on the other hand revs the GPU's fans to max - or ver near max...


.

 

[sarthak]

It's just not heating it up is all. AFAIK it all works from thermostatic readings. It doesn't care about load or rates or anything... just temperature. So if the fans rev up in one circumstance then this shows the fans and sensors are working and you can go back to not worrying - even marveling at how much quieter OS X is than Windoze...

BTW, I play some games that look awfully heavy yet the 8800GT's fans barely increase at all... QuakeLive at 1920x1080 with everything set to high or ultra high, and all post processing turned up to max. CPUs system wide are less than 20% (usually around 10%), and the 8800XT's fans are not audible over my SMCFanControl setting of 1300. The card draws 90 to 95% of it's max power specification according to "Hardware Monitor". Furmark on the other hand revs the GPU's fans to max - or ver near max...

I guess OS X games running in OpenGL are much more efficient than DX games in Windows. F1 2012 looks amazing in 2560x1440 with all settings turned up to the absolute maximum. I'd assume it would burn the GPU like it's Windows equivalent.

 

[Asgorath]

I guess OS X games running in OpenGL are much more efficient than DX games in Windows. F1 2012 looks amazing in 2560x1440 with all settings turned up to the absolute maximum. I'd assume it would burn the GPU like it's Windows equivalent.

The unfortunate reality is that while the game looks and plays just fine, it's probably running at a much lower framerate overall than it does under Windows. F1 2012 uses DX11 under Windows, which enables things like tessellation and DirectCompute, while Apple is basically only exposing DX9-level functionality (which tends to be less efficient overall, and thus easier to max out on the software/CPU side of things, which leaves the GPU under-utilized). DX11 games in general tend to be much more efficient, due to changes at the API level, and thus they are much more likely to be running the GPU at 100%, which results in the fans kicking in and higher framerates overall. Either that, or the DX11 version has even more advanced lighting effects and thus looks a little better at similar framerates to the DX9 version.

At the end of the day however, you have a great-looking game that runs well on your computer. Hard to complain too much about that.

 

[sarthak]

At the end of the day however, you have a great-looking game that runs well on your computer. Hard to complain too much about that.

Yes, I noticed the lack of tessellation in Batman: Arkham City in OS X. The games are still very nice looking considering they're running on OS X. We have never seen gaming at this level on a Mac before.

At least PC users can no longer say "you can't play games on a mac."

 

[Asgorath]

Yes, I noticed the lack of tessellation in Batman: Arkham City in OS X. The games are still very nice looking considering they're running on OS X. We have never seen gaming at this level on a Mac before.

At least PC users can no longer say "you can't play games on a mac."

Here's some information about how you can download the graphics tools:

http://developer.apple.com/library/...eptual/HighResolutionOSX/Testing/Testing.html

Note that it requires you to have an account on developer.apple.com, but getting one of those is free. Would be interesting to see what the GPU utilization is when running those games, my guess is that the combination of a 2.66GHz CPU and a GTX 670 means that the GPU is at 50% utilization or less.

 

[sarthak]

Here's some information about how you can download the graphics tools:

http://developer.apple.com/library/...eptual/HighResolutionOSX/Testing/Testing.html

Note that it requires you to have an account on developer.apple.com, but getting one of those is free. Would be interesting to see what the GPU utilization is when running those games, my guess is that the combination of a 2.66GHz CPU and a GTX 670 means that the GPU is at 50% utilization or less.

These are the results I got, good or bad? DOWNLOAD EXCEL SPREADSHEET WITH GRAPHS HERE or view the Attached in text format and screenshots below:

Unigine Valley Benchmark:

F1 2012

OR LINEAR IF THIS HELPS

There may be loading screens as per the drops.

 

[Tesselator]

I would assume similar as what Asgorath is suggesting. AFAIK OS X doesn't even support the 4 or 5 versions of OpenGL. And those missing extensions are likely to really put the GPU to work. More information here:

http://www.geeks3d.com/20121109/overview-of-opengl-support-on-os-x/

and here:

http://www.geeks3d.com/20121113/gpu...gl-benchmark-furmark-lands-on-linux-and-os-x/

And the games being rendered might look quite similar if you were unaware of what to check for.



PS: For the above graphs to be meaningful in this particular context we would need to comparative graphs from windows running the same code.

 

[sarthak]

PS: For the above graphs to be meaningful in this particular context we would need to comparative graphs from windows running the same code.

Thanks for the links. I wiped my Boot Camp partition because Windows 8 would not boot after I updated to the latest Nvidia driver. Using a non-EFI card made the problem even worse.

I still use Windows 7 in Boot Camp and Parallels (and W8 in Parallels) however, that's on a completely different Mac.

Update: Perhaps the card was not being utilized enough to cause it to heat up. I installed the NVIDIA GeForce and Quadro driver for mac and ran Valley for a couple hours. I can hear the fan revving up while gaming now. I assume it revs up when the card gets hot enough (like it does when you melt it with valley).

 

[Asgorath]

Yeah, I forgot to mention, setting the display to "Linear" makes the graphs much easier to read.

As you can see, the GPU core is barely above 50% utilization, memory is more like 20%, and the CPU is not really stalling a lot. This usually points to a software inefficiency, mostly due to the fact that you're running the DX9 equivalent version of the Unigine benchmarks. Under Windows, you'll be using the DX11 version, which will be more efficient (both on the app side, and on the driver side). The relative speeds of your 2.66GHz CPU and the GTX 670 are probably a factor, i.e. your CPU is from about 4 years ago while your GPU is a modern high-end one. Without a super-efficient API like DX11, the CPU is having a hard time feeding enough work to the GPU to keep it busy.

So, in summary, the reason the fans aren't kicking in is because the GPU is sitting idle doing absolutely nothing for 50% of the time.