AMD Radeon R9 M295X Core Clock Throttling, Heat, and Performance

[Astelith]

Guys, even the CPU core is at 105° at the top of the graph so you're telling me that even my CPU reached 105° ?
How about the RAM ? I have 100°

And yes ! AMBIENT ! I had 100° C in my studio ! I was so lucky beign outside when it happened !

cmon... this thread was about to became a honest discussion

----------

Edit: for me the question remains - is the throttling caused by the heat?

The question can be answered, at least partially, by the test Steve can make.

Let's see

 

[andy9l]

Guys, even the CPU core is at 105° at the top of the graph so you're telling me that even my CPU reached 105° ?
How about the RAM ? I have 100°

I just re-installed iStat Menus on this iMac and all the graphs have dynamic value axis, based on the peak temperature reported by each sensor.

I'm not doing anything, so they all varied between about 25° and 50°, but they were all relevant and changed as the hardware jumped up 1°. So yes, any temperature you see at the top of your graphs = the maximum temperature reported by the relevant sensor.

We are having an honest discussion, and this is how iStat Menus works. Same as most peak/trough monitoring tools.

 

[WilliamG]

Guys, even the CPU core is at 105° at the top of the graph so you're telling me that even my CPU reached 105° ?
How about the RAM ? I have 100°

And yes ! AMBIENT ! I had 100° C in my studio ! I was so lucky beign outside when it happened !

cmon... this thread was about to became a honest discussion

----------



The question can be answered, at least partially, by the test Steve can make.

Let's see

You can believe what you want to believe, at this point. I KNOW that your GPU has hit 105C, because you've provided the proof! The i7 CPU gets VERY hot, too - just an fyi.

As much as I think Astelith MEANS well, he's providing a LOT of misguided information about how our iMacs are supposed to be running.

----------

I just re-installed iStat Menus on this iMac and all the graphs have dynamic value axis, based on the peak temperature reported by each sensor.

I'm not doing anything, so they all varied between about 25° and 50°, but they were all relevant and changed as the hardware jumped up 1°. So yes, any temperature you see at the top of your graphs = the maximum temperature reported by the relevant sensor.

We are having an honest discussion, and this is how iStat Menus works. Same as most peak/trough monitoring tools.

Yes, this is exactly how iStat works.

 

[Astelith]

Maybe theres a default scale that can be dynamically increased if the value reach the end of the scale...

Explain me the ram, all the other sensor and most important part... the ambient temp !

 

[andy9l]

Nope, starts at whatever temperature the sensor is when you first start iStat Menus.

https://www.youtube.com/watch?v=ywCpTxastAE

Post a screenshot/video of your 100°C RAM graph?

 

[Astelith]

Nope, starts at whatever temperature the sensor is when you first start iStat Menus.

https://www.youtube.com/watch?v=ywCpTxastAE

Post a screenshot/video of your 100°C RAM graph?

 

[andy9l]

Am I understanding this correctly...

If Steve runs the same benchmark, in the same environment, with the fans at 2700rpm (manual intervention) and there's NO throttling this time, then have we not proven that the M295X iMacs, as sold by Apple, cannot cool the GPU sufficiently?

It's just one data point, sure, but the rest have been very, very consistent so far...

 

[Astelith]

Am I understanding this correctly...

If Steve runs the same benchmark, in the same environment, with the fans at 2700rpm (manual intervention) and there's NO throttling this time, then have we not proven that the M295X iMacs, as sold by Apple, cannot cool the GPU sufficiently?

It's just one data point, sure, but the rest have been very, very consistent so far...

If forcing the fan to the maximum he got a lower temp (around 95°) the clock go back at 850 straight we prove that even his machine can potentially have enough thermal power to dissipate the GPU at 100% but also that the cooling system settings have an issue, fan regulator, sensor... who knows...

Tomorrow I will do the exact test

 

[steve62388]

New video. It's late and I need to go to bed, so I can discuss this tomorrow. But basically all settings are the same as my previous vid.

GPU is fairly stable at 834Mhz. It shows me increasing the fan speed to 2700rpm and there is no change to the GPU clock.

I then also max out the CPU using Prime95, fan 2700rpm, GPU clock drops from 834Mhz to 784Mhz. The video was a bit long at that point but I left it running for a while without recording and there was no further change, GPU remained at 784Mhz.

https://drive.google.com/file/d/0B9vl5TopANpYOERDbGoxbWJsY1U/view?usp=sharing

 

[Astelith]

New video. It's late and I need to go to bed, so I can discuss this tomorrow. But basically all settings are the same as my previous vid.

GPU is fairly stable at 834Mhz. It shows me increasing the fan speed to 2700rpm and there is no change to the GPU clock.

I then also max out the CPU using Prime95, fan 2700rpm, GPU clock drops from 834Mhz to 784Mhz. The video was a bit long at that point but I left it running for a while without recording and there was no further change, GPU remained at 784Mhz.

https://drive.google.com/file/d/0B9vl5TopANpYOERDbGoxbWJsY1U/view?usp=sharing

Thanks for this test.

I made one very quickly yesterday night (poor quality and with the dog snoring all the time ), I still have Win 10 so I need to do it properly as soon as I format with the 8.1.

I ran the same bench as you with auto, full fan and then 1200rpm, in this test the performance, even if almost ininfluent, are correlated to the temperature so we need to do more with different workloads (and Win 8.1 in my case).

https://www.youtube.com/watch?v=CrPo_QoQ398

Ok now we have added some other question marks...

To summarize by now we have :

- Different temps and fan speed under the same workload (still not at 100% due to Win 8.1 vs Win10, but I don't think we will have differences)
- In certain cases temperature matters, a bit, with almost no performance drop but it matter (To be verified with other different tests) there's also the fact that at higher temperature the graphics board lose efficiency with the result to a higher power draw
- The clock ranging is due mainly to the workload on the GPU + CPU
- We have a huge difference between iMacs for temperature and fan speed , maybe for different internal components ?

After this I start to think that every iMac is in need of some intervention by Apple by improving the Firmware and/or hardware like thermal paste in the worst case.
Now I can understand why only some people are disappointed by the fan, in this test we have more than 300rpm of difference, that's odd.

We need to test more, this thing is very far to be clear

 

[Astelith]

And don't jump to fast conclusion, In all the test I ran this is the first time I saw a relation between clock and temperature below 95°, we really need to dig more before saying anything

p.s.: we are using the same drivers 14.301.1010.0 because of that I doubt to have different result in Win 8.1 (but I will do it again)

 

[steve62388]

And don't jump to fast conclusion, In all the test I ran this is the first time I saw a relation between clock and temperature below 95°, we really need to dig more before saying anything

p.s.: we are using the same drivers 14.301.1010.0 because of that I doubt to have different result in Win 8.1 (but I will do it again)

I'm pretty happy with the results. On my 5K I find that:-

1. Bumping up fan speed to 2700rpm does not limit overall top temperature. It does slow down how long it takes to get there, I observed this on my test but it's not in the video because it took too long. The difference is only a few minutes so I am therefore happy with Apple's choice running the fan at 2400rpm, 2700rpm is noticeably louder.

2. Hammering the GPU (100%) does not throttle it to any appreciable amount, certainly a drop from 850Mhz to 834Mhz does not bother me.

3. Loading both GPU and CPU to 100% for more than several minutes drops the GPU clock to 784Mhz. I consider this acceptable, it's not that huge. How often are you going to load your iMac like that? I would suggest not very often. If say for example a game did (unlikely though because most wouldn't load all cores of a CPU) then you will get a drop of less than 10%. Why is this a problem and surprise? Some apps tax a system more than others and run a bit slower.

4. My Retina iMac does throttle slightly at top temperature. A water cooled tower with exactly the same components may not throttle at all. But I don't think this an issue, what did everyone expect in an ultra thin all in one? And even when it does throttle it's not what I consider a large amount.

It would be an interesting exercise to figure out at what percentage and mix of load the GPU throttles eg 80% GPU and 70% CPU. But I'm not prepared to spend the time to find out.

Does everyone know the CPU throttles too? Nobody seems to be banging on about that.

 

[Astelith]

I'm pretty happy with the results. On my 5K I find that:-

1. Bumping up fan speed to 2700rpm does not limit overall top temperature. It does slow down how long it takes to get there, I observed this on my test but it's not in the video because it took too long. The difference is only a few minutes so I am therefore happy with Apple's choice running the fan at 2400rpm, 2700rpm is noticeably louder.

2. Hammering the GPU (100%) does not throttle it to any appreciable amount, certainly a drop from 850Mhz to 834Mhz does not bother me.

3. Loading both GPU and CPU to 100% for more than several minutes drops the GPU clock to 784Mhz. I consider this acceptable, it's not that huge. How often are you going to load your iMac like that? I would suggest not very often. If say for example a game did (unlikely though because most wouldn't load all cores of a CPU) then you will get a drop of less than 10%. Why is this a problem and surprise? Some apps tax a system more than others and run a bit slower.

4. My Retina iMac does throttle slightly at top temperature. A water cooled tower with exactly the same components may not throttle at all. But I don't think this an issue, what did everyone expect in an ultra thin all in one? And even when it does throttle it's not what I consider a large amount.

It would be an interesting exercise to figure out at what percentage and mix of load the GPU throttles eg 80% GPU and 70% CPU. But I'm not prepared to spend the time to find out.

Does everyone know the CPU throttles too? Nobody seems to be banging on about that.

Well, overall there's no performance issues (or very very minimal), the only real downside is the fan speed, we have 400rpms of difference doing the same thing, that is not small thing...

 

[tmorterlaing]

I'm pretty happy with the results. On my 5K I find that:-
The difference is only a few minutes so I am therefore happy with Apple's choice running the fan at 2400rpm, 2700rpm is noticeably louder.

Well, overall there's no performance issues (or very very minimal), the only real downside is the fan speed, we have 400rpms of difference doing the same thing, that is not small thing...

So Astelith, am I write in saying yours runs at those temps at 1900rpm? In which case why don't we all run a test - lock our fans off at about 1950rpms and see if our GPUs stay at below 106 degrees. If they DO then I guess that means Apple has just amped the fans up a bit too much? After all, we know from Steves test above that even on 2700rpm the GPU doesnt run cooler. I wonder what the cut off point is before the GPU starts to exceed that 105mark...
If it's about 2000 for example and Apple were just overcompensating, then I'll be a happy camper - because there;s quite a big noise difference between 2000 and 2300

Cheers for the testing guys, much more insightful. Also quite pleasing to notice that the high temperature is looking like its expected. I was most interested by watching the steve video that the 'throttling' really isnt throttling at all, its just lowering a touch and then going back to maximum. So, i take back what Ive said with two caveats -
1: we still don't know for certain this thing is supposed to run this hot, especially as quite a few people who know their stuff have expressed well backed worries. But we probably won' know for months...
2: the fan is still noisy, and whilst I understand why now, its a shame that Apple have managed to take a step back in that department. You'd have thought they would try and find a better balance before releasing a new one.

 

[steve62388]

Well, overall there's no performance issues (or very very minimal), the only real downside is the fan speed, we have 400rpms of difference doing the same thing, that is not small thing...

I did notice that and consider it. But on reflection I don't feel like I can state anything one way or another, my fan speed sensor might be faulty, your fan sensor might be faulty (although I think 2400rpm is more typical from what I have seen from other posters), you might have better thermal paste, it could be the difference in Windows 8.1 vs 10, I might have more dust in my iMac, and it really could be ambient temperature in the room. This last point might be particularly relevant because we do keep quite a warm house.

That's why I thought it was more interesting to state that the fan kicked in at certain temperatures etc because that appears to be set by the software and both did it at the same time.

 

[steve62388]

So Astelith, am I write in saying yours runs at those temps at 1900rpm? In which case why don't we all run a test - lock our fans off at about 1950rpms and see if our GPUs stay at below 106 degrees. If they DO then I guess that means Apple has just amped the fans up a bit too much? After all, we know from Steves test above that even on 2700rpm the GPU doesnt run cooler. I wonder what the cut off point is before the GPU starts to exceed that 105mark...
If it's about 2000 for example and Apple were just overcompensating, then I'll be a happy camper - because there;s quite a big noise difference between 2000 and 2300

I don't think anybody knows for sure what the max temp of the R9 M295X is supposed to be, but it looks around 105 degrees, with a few reaching 108 or so. There is a poll here that will let you know. Once it hits the rated max then it should throttle to prevent damage.

But your 2000rpm vs 2400rpm comment got me thinking whether my iMac compared to Astelith's is aggressive with the fan speed and it seems to be. (Room temperature issues again?).

I ran another test this time with a fixed fan of 2000rpm. MSI reported the exact same GPU clock speed as at 2400rpm, so the slower and quieter fan speed does not result in extra throttling. But to really push it I also maxed out the CPU and I got the exact same result. My 5K does not appear to suffer any extra throttling or temperature on the GPU when CPU/GPU are maxed out and the fan is fixed at 2000rpm compared to 2400rpm. I do not know if my system was throttling the CPU more at 2000rpm than 2400rpm to stay within spec.

It would appear my iMac is more aggressive ramping up the fan to 2400rpm than it needs to be. Am I going to change it? No, I would rather leave it in the hands of Apple's engineers to decide how to run it.

Video:- https://drive.google.com/file/d/0B9vl5TopANpYRndlTWFkQTEwd3c/view?usp=sharing

 

[tmorterlaing]

Still, an interesting point - I wonder how far that line goes? I'll try when my replacent arrives, taking it down in iterations of 50rpm.

I suspect Apple is probably compensating for something else like ambient temperature, but I might bring it down by 100 rpm or so if I notice no difference in max temp with longer testing- for quiet 's sake if nothing else!

 

[andy9l]

So, what happened in my video where the clock speed dropped to the 700s?

http://vimeo.com/116208647

I mean, the correlation between FPS drop, clock speed throttle and GPU heat seems to be 1:1...?

 

[steve62388]

So, what happened in my video where the clock speed dropped to the 700s?

http://vimeo.com/116208647

I mean, the correlation between FPS drop, clock speed throttle and GPU heat seems to be 1:1...?

Your results seem pretty consistent with mine. You will note that your video shows it hovers around 784Mhz most of the time which is exactly like mine when I stress the GPU and CPU together. Slight variation is no surprise to me, a number of things could contribute to that. If you're really curious try the exact same test with the same drivers, software and O/S.

There could also be differences in the build quality of your iMac and not all GPUs and CPUs are created equal, that's why some overclock and some don't.

 

[thundersteele]

Let me think about the test and I let you know, I have also to reinstall Windows so I need time.

OK so I found some time to run the tests from https://forums.macrumors.com/threads/1731178/
They are only marginally useful, since only the CPU temperature and frequency is monitored. Yet I was able to note some interesting facts about the behaviour of the fans. I ran in three configurations:
1. Prime95
2. 3D GpuTest
3. Prime95 + 3D GpuTest

1. The test ran through with constant 3.5 GHz (i5), no throttling. Max fan speed 2300rpm, max CPU temp was 97C
2. Max fan speed 2300rpm max GPU temp 105C. No fps drops, no visible evidence of throttling
3. Max fan speed 2700rpm, max CPU temp 100C, max GPU temp 106C. The CPU throttled to 3.2GHz-3.3GHz once it hit around 98C. No fps drops in the GPU test, no visible evidence of throttling on the GPU. After Prime95 ended, the GPU test still continued running. Yet the fan speed quickly dropped to 2300rpm, keeping the GPU stable at 105C

It is not clear what to learn from this, but it is interesting that the GPU alone doesn't push the fans above 2300rpm. One possible interpretation is that 105C is indeed considered an OK temperature for the GPU, and no thermal throttling happens there.

Plots attached, but are not super useful as there is limited GPU information

Prime95 alone:

GPU test alone (this is of limited use):

Prime95 + GPU test at the same time - CPU throttles:

 

[andy9l]

Your results seem pretty consistent with mine. You will note that your video shows it hovers around 784Mhz most of the time which is exactly like mine when I stress the GPU and CPU together. Slight variation is no surprise to me, a number of things could contribute to that. If you're really curious try the exact same test with the same drivers, software and O/S.

But if we're not calling 850-834 down to 720-780MHz (nearly 10% drop) "throttling", then what is? The GPU is lowering it's core clock speed to lower the heat. It's throttling, no? The performance is clearly impacted in my video, minimal as it may be - but I'm just staring at a blank wall.

My 5K iMac has long gone back to Apple - no more tests from me I'm afraid

 

[steve62388]

But if we're not calling 850-834 down to 720-780MHz (nearly 10% drop) "throttling", then what is? The GPU is lowering it's core clock speed to lower the heat. It's throttling, no? The performance is clearly impacted in my video, minimal as it may be - but I'm just staring at a blank wall.

My 5K iMac has long gone back to Apple - no more tests from me I'm afraid

Well I never implied there is no throttling, in fact I said my iMac does throttle. I just don't consider it a lot when it takes 100% CPU and GPU to drop the M295X to 784Mhz on mine when that's a worse case scenario. There may be other combinations of load which cause a different throttle effect but as I say I'm not prepared to spend the time to find out.

 

[andy9l]

Yeah that's fair enough. Thanks for doing the tests Steve.

For me, the findings haven't persuaded me to give this generation another shot. I'm keeping fingers crossed for a cooler, quieter 2nd-gen - more inline with previous Macs.

 

[ioxakoyba]

I'm pretty happy with the results. On my1. Bumping up fan speed to 2700rpm does not limit overall top temperature. It does slow down how long it takes to get there, I observed this on my test but it's not in the video because it took too long. The difference is only a few minutes so I am therefore happy with Apple's choice running the fan at 2400rpm, 2700rpm is noticeably louder.

Well, I was the guy who said that it would be better with 2700rpm than 2300rpm. And yes, I am still thinking the same way.

Also this proves, in opposite what Astelith claims, and the same I on my earlier posts said, that the cooling of this iMac is just insufficient. When on 100% CPU load + 100% GPU load it cannot keep temperature in designed limits even with the max fan speed 2700rpm and ultimatelly CPU plus GPU needs to throttle to lower the temps.

But I want to point out also that this very same thing applies also with many laptops, for example my 13" Macbook Pro Retina (mid 2014). The cooling is not built to sufficient for such cases where both CPU and GPU goes to full load for several minutes. Also this is very rarely the case in practise, even with hard core gaming, so the case is only theoretical.

However, steve23094, you should also take in account that in practical use (like gaming where GPU load may be 100% nearly constantly, but CPU load varies) the result may be much different if you compare fan speeds 2300rpm and 2700rpm. It may be even possible that in such loads the actual gaming causes the GPU temperatures may top on different maximums, just like my Macbook Pro Retina case shows (100 C on default cooling, 93-96 C putting the fans to maximum).

More noise? Yes, but as far I am concerned, I game only with bluetooth headsets, so it really doesn't matter is the fan outputting 25 or 30 dB, as long as the fan keeps down on desktop applications and such (when I am not wearing the headsets). The most important is to prevent GPU from throttling complitely, and it *might just be possible* by setting max fan speed to 2700rpm with 3rd party application and on some actual workload a modern game causes.

We'll see. My full loaded riMac should arrive within 3-4 weeks.

NOTE! My point however is that I am not happy with the insufficiently built cooling system of riMac, but at this time I am so keen on the 5k display so I may be able to live with such way to tinker fan speeds with 3rd party apps.

PS. Steve23094 (or anybody) would you be so kind and make a practical test: set fan manually to run 2300rpm (or whatever the Apple default maximum is) and game about one hour (some modern game, whatever you like to game). Log the maximum GPU temp with GPU-Z. Then repeat the test by setting fan to maximum 2700rpm and check the same GPU temp (of course restarting the GPU-Z first to reset the logged maximum temp value). Is there any difference? I really liked to see couple of celcius lower temps with max 2700rpm.

 

[andy9l]

The fan at 2300rpm is 38dB, according to the Apple discussions board. Either way, it's pretty darn loud, which is what caused my initial concern and led me here. It's also loud enough for hundreds/thousands of others to question it online too.

I also found the fan spinning up on desktop applications - even just exporting and playing 1080p 60FPS video in iMovie or QuickTime. A few tasks that my 2010 iMac breezes through, caused heat and fan noise on the 5K iMac. Even a few times watching YouTube. Obviously it's a lot faster, but at a heat/noise compromise. There's quite a few posts about this in particular as it's more common than gaming.

Those who "never hear the fan" simply don't do any activities that tax the system. We've proved that here.

I don't think we should compare a fully-loaded iMac to a low-end laptop with no discrete GPU. Instead, compare against previous fully-loaded iMacs. The best would be 2012/2013 models as they are identical in terms of internal/external design. I haven't see any throttling on those, but maybe they do...?

You'll be able to make your own decision when it arrives. My opinion of another current gen iMac would be forever tainted by the findings in this thread.