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

[tmorterlaing]

I've been following this thread with some interest now.
I'm a video editor, and without meaning to sound all cynical - Astelith, please can you do the test again but this time film the whole screen. When you cmd+tab I want to see the app change bar thing appear when you go to Finder. Without being super distrusting; and PLEASE note this is NOT an accusation because I'm very grateful for you input - but it's interesting that both the 'fade to black's appear to be 10 frames long, which is the default time for that transition in FCPX and other NLEs. Thanks man

If it is the case, to be honest Astelith's machine might be throttling a bit, but the point is that the game is still running smoothly - so it sorta doesn't really matter? The main point is that if this is proof, then Asteliths machine is running cooler than all of ours, which means there's a HUG amount of iMac owners out there with faulty GPUs.

Contrary to my previous post btw, I'm getting a replacement one soon - I'll run the same tests and let you know if this one is the same. Should be with me by Wednesday.

 

[steve62388]

Perfect, at this point we don't have much options, I have a Win 8.1 full license so I will use it to have the same base.
Let's write down the drivers we are going to use, I suggest the last, normal or beta ? But not a driver that need system tweaking, too time consuming.

For the software i suggest Geek3D GPU Test GUI 0.7.0 as it contains all the stress test we need (TessMark, PixMark, FurMark etc.) and they behave very different to each other on the GPU boost clock, TessMark for example can push max temp and max frequency (850) but it uses only a medium part of the GPU so the TDP is below range, FurMark is heavier and the GPU average is 834.

Look here: https://forums.macrumors.com/posts/20584267/
And let's pick one or more of the tests.

To measure with extreme accuracy let's use MSI Afterburner, it's free and it can provide us the graph and the OSD.

There's nothing actual capable of logging the fan speed, we need to make a video with Macs Fan control or iStat menu to see the fan behavior

Okay, my specs. M295X, 32GB RAM, 4Ghz i7, Windows 8.1 fully patched, Windows O/S installed on internal SSD (apps on external mechanical Thunderbolt drive though). Driver version 14.301.1010-141101a-177531C, this is the most recent official AMD drivers but I don't know if these have been rolled into the Boot Camp driver download yet so they may have to be installed manually.

I have Macs Fan Control 1.1.12.3, Geeks3D GPU Test GUI 0.7.0, MSI Afterburner 4.1.0.

Afterburner has been hit with the ugly stick hasn't it? Yuck.

 

[steve62388]

Perfect, at this point we don't have much options, I have a Win 8.1 full license so I will use it to have the same base.
Let's write down the drivers we are going to use, I suggest the last, normal or beta ? But not a driver that need system tweaking, too time consuming.

For the software i suggest Geek3D GPU Test GUI 0.7.0 as it contains all the stress test we need (TessMark, PixMark, FurMark etc.) and they behave very different to each other on the GPU boost clock, TessMark for example can push max temp and max frequency (850) but it uses only a medium part of the GPU so the TDP is below range, FurMark is heavier and the GPU average is 834.

Look here: https://forums.macrumors.com/posts/20584267/
And let's pick one or more of the tests.

To measure with extreme accuracy let's use MSI Afterburner, it's free and it can provide us the graph and the OSD.

There's nothing actual capable of logging the fan speed, we need to make a video with Macs Fan control or iStat menu to see the fan behavior

I have been playing around with everything. What stress test do you want to run, what resolution and AA? I can see the throttling and that's fine it's doing what it's supposed to, but I can't decide which test best shows the results.

 

[thundersteele]

105° and 2300rpm doesn't make sense, one of the two are wrong, a functional machine will never and never let the GPU go above 100°C (or 101° if you consider some peak) and at 2300 there's other 400 rpm to use to keep it cool, something is not wright in your machine

This is how Apple programmed the machines, in OSX. I sometimes switch my games to windowed mode, so I can check GPU temps and fan speeds with the game still running. Usually the fan will be at 2300 rpm and the GPU temperature around 104 C (TESO 2560x1440 everything high).

I use iStat menus for temperature and fan speeds. I prefer not to manually set the fans higher, since the noise level goes up quite a bit.
I believe that Apple chose a default behaviour that will give a better experience to the average user, namely moderate noise and some throttling under heavy load. This is very much in line with how Apple makes MBPs behave, except that they chose a slightly higher temperature band for the M295X.


Sometime ago someone posted a nice tool in the MBP forums:
https://forums.macrumors.com/threads/1731178/
Maybe it would be a good idea to start a new thread here using that tool.

 

[Astelith]

This is how Apple programmed the machines, in OSX. I sometimes switch my games to windowed mode, so I can check GPU temps and fan speeds with the game still running. Usually the fan will be at 2300 rpm and the GPU temperature around 104 C (TESO 2560x1440 everything high).

I use iStat menus for temperature and fan speeds. I prefer not to manually set the fans higher, since the noise level goes up quite a bit.
I believe that Apple chose a default behaviour that will give a better experience to the average user, namely moderate noise and some throttling under heavy load. This is very much in line with how Apple makes MBPs behave, except that they chose a slightly higher temperature band for the M295X.


Sometime ago someone posted a nice tool in the MBP forums:
https://forums.macrumors.com/threads/1731178/
Maybe it would be a good idea to start a new thread here using that tool.

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

But first read carefully this document from AMD :
https://www.amd.com/Documents/PowerTune_whitepaper_WEB.pdf
Before testing we need to clarify once for all the difference between Throttling and the clock ranging because of the "AMD PowerTune" technology, also called "Boost".
TDP means Watt as power draw in current, not C°/Thermal and you get to really understand this point, Nvidia cards work with Temperature limit, here is different, AMD works with power draw, even if you chill the GPU with LN2 to -100°C and you are asking more than the TDP (should be 125W) you will notice a clock reduction.

p.s.: In Teso I got 100°C maximum with the fan around 2300, because of that I think your 104° is my 100°, and the crazy one is the sensor (mine, yours or both)

p.s. II: Even with 108° there's no throttling but an adjustment that PowerTune made to prevent the Throttling, lowering to 6xx/7xx the clock, to engage a real throttling I had to put at 100% GPU + CPU + Fan forced at minimum to trigger it

 

[tmorterlaing]

This is all well and good - but I don't see why on earth Apple made the decision to drastically increase noise levels. This card does NOT need to work so hard to play a lot of games (my previous example being borderlands) i if one imac can do it silently at max spec, why on earth can't a NEWER imac?

I really hope this is all a big understanding and that the sensors are just being oversensitive or something - 2300rpm is the loudest I've heard an imac fan go doing anything, ever - and I've messed with every iteration.

Also, to the guy a few posts back who suggested theyd happily run it at 2700rpm - have you HEARD 2700rpm? that is loud - I couldn't play a game like that. Unless you intend to where headphones, that is...

 

[cynics]

Diablo 3, 5K, High settings, Fullscreen, Os X



https://www.youtube.com/watch?v=MhO24PCbD84&feature=youtu.be



But considering that in Os X is not that easy to track the temp I made a better test under Win 10 with better tools.



I played the chapter "Separation" in Metro Last Light, 4K, High Details, first with the fan at max (2700) and then in "Auto", so controlled by the system (Apple default), pay attention to the fan noise and look at the frame per second, exactly the same, 93-94° vs 99-100° = no difference of performance (look at 1:17 and 7:50), and the temp is not passing 100°C, even with the GPU constantly at 100%



https://www.youtube.com/watch?v=EWLuCQpt5Kc&feature=youtu.be



This is for sure the most demanding game I tried on this machine but the fan doesn't go over 2300 as you can hear, yes, a bit loud, but hey, we are talking about 4K on High detail and 25+ Fps !

If I lower the resolution and/or the detail I have an average clock higher (800-850) and temps around 95-96° due to the GPU usage mostly below 75% and so the fan at lower rpms.



I can keep testing and making videos but I hardly think nobody here will change their minds.



You (or who with high temps) should do a similar test to compare and understand better this issue...

I actually feel this helps illustrate the problem. You are seeing fairly high temps at a lower GPU usage.

In the PC world if I built a machine 100% GPU and 75% GPU usage should have a very wide margin between them. If they didn't I'd know I had insufficient cooling.

Or basically it would operate like a non retina iMac.

Any GPU can get insanely hot the m295 is no exception. And all are designed to throttle and all have maximum safe operating limits. No issues there. It all comes down to the system that is designed to cool it.

In an all in one there should never ever be throttling as that indicates insufficient cooling design. Fan shouldn't reach maximum on a new machine either because that leaves no margin of error namely dust build up over a couple months.

It seems to me people are seeing one or the other or both. Although I find the 2300 rpm talk a little baffling if temps exceed 100c.

 

[tmorterlaing]

Why is no one investigating this properly? Seriously, this is so obviously an issue - there is not a SINGLE GPU out there which is 'supposed' to operate at over 100 degrees for more than a short period of time. This will happily sit at 105 for HOURS if gaming. Apple has messed up, clearly, and we're all sitting around wondering whether 100 degrees is acceptable etc - it clearly isn't! It's too loud, too hot and not acceptable for such an expensive machine. There should be no compromise of having to use 3rd party apps or putting up with distracting noise when gaming -again, its an EXPENSIVE machine, it should be able to do things at least as well as previous versions.

 

[steve62388]

Why is no one investigating this properly? Seriously, this is so obviously an issue - there is not a SINGLE GPU out there which is 'supposed' to operate at over 100 degrees for more than a short period of time. This will happily sit at 105 for HOURS if gaming. Apple has messed up, clearly, and we're all sitting around wondering whether 100 degrees is acceptable etc - it clearly isn't! It's too loud, too hot and not acceptable for such an expensive machine. There should be no compromise of having to use 3rd party apps or putting up with distracting noise when gaming -again, its an EXPENSIVE machine, it should be able to do things at least as well as previous versions.

Let's throw in as much hyperbole as possible shall we?

Maybe you should read my post here:- https://forums.macrumors.com/showthread.php?p=20682691#post20682691

It looks like this applies to you.

 

[steve62388]

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

But first read carefully this document from AMD :
https://www.amd.com/Documents/PowerTune_whitepaper_WEB.pdf
Before testing we need to clarify once for all the difference between Throttling and the clock ranging because of the "AMD PowerTune" technology, also called "Boost".
TDP means Watt as power draw in current, not C°/Thermal and you get to really understand this point, Nvidia cards work with Temperature limit, here is different, AMD works with power draw, even if you chill the GPU with LN2 to -100°C and you are asking more than the TDP (should be 125W) you will notice a clock reduction.

p.s.: In Teso I got 100°C maximum with the fan around 2300, because of that I think your 104° is my 100°, and the crazy one is the sensor (mine, yours or both)

p.s. II: Even with 108° there's no throttling but an adjustment that PowerTune made to prevent the Throttling, lowering to 6xx/7xx the clock, to engage a real throttling I had to put at 100% GPU + CPU + Fan forced at minimum to trigger it

Here is a little video I made, excuse the quality and stuff. Ignore the sound my iPhone mic seems a bit sensitive, you can even hear me breathing and I didn't hear myself.

https://drive.google.com/file/d/0B9vl5TopANpYQnNUTkktaTVXN28/view?usp=sharing

So what does this tell me about my Retina iMac?

- My fan sits at 1200rpm and doesn't start ramping up until the GPU hits 100C.
- It starts to lower GPU clock speed at around 95C. If you can see it clearly enough you will notice this is ever so slight (850/834) and it does it on the on the fly both up and down.
- Fan hits max rpm about 102C. 2300rpm.
- At about 104C my iMac has hit max TDP and the GPU dynamically alters clock speed between a low of of late 700's (very seldom) and usually sits at 834Mhz.
- The fan does not start to scale back from 2300rpm after unloading the GPU until it hits about 70C. A longer test changing the GPU load from 100 to a lower percentage may show something different.

Overall I'm fairly impressed with this. The lowering of the GPU clock under load is what I consider very minor, I mean 850 to 834 (most of the time) is not a biggie is it? It's got to be done and all seems to be working as it should, the GPU gets a little hot and scales back, cools down and kicks back up with all this happening every few seconds.

I mentioned earlier everyone seems to be fretting about whether 105C is too high and I'm saying... well read my previous post.

 

[tmorterlaing]

Let's throw in as much hyperbole as possible shall we?

Maybe you should read my post here:- https://forums.macrumors.com/showthread.php?p=20682691#post20682691

It looks like this applies to you.

Ok, fair enough - a point well made. The fact still stands though, Apple have chosen to use a very hot ant GPU that sits close to other components in an all in one design. this means the fan speed goes up to 2300 (which is not the maximum btw - that's 2700) whilst gaming. This means Apple has basically said ' if you want to game, fine, but expect a fair bit of noise' - which is frustrating, and unnecessary - as proven by previous models.

 

[Astelith]

Here is a little video I made, excuse the quality and stuff. Ignore the sound my iPhone mic seems a bit sensitive, you can even hear me breathing and I didn't hear myself.

https://drive.google.com/file/d/0B9vl5TopANpYQnNUTkktaTVXN28/view?usp=sharing

So what does this tell me about my Retina iMac?

- My fan sits at 1200rpm and doesn't start ramping up until the GPU hits 100C.
- It starts to lower GPU clock speed at around 95C. If you can see it clearly enough you will notice this is ever so slight (850/834) and it does it on the on the fly both up and down.
- Fan hits max rpm about 102C. 2300rpm.
- At about 104C my iMac has hit max TDP and the GPU dynamically alters clock speed between a low of of late 700's (very seldom) and usually sits at 834Mhz.
- The fan does not start to scale back from 2300rpm after unloading the GPU until it hits about 70C. A longer test changing the GPU load from 100 to a lower percentage may show something different.

Overall I'm fairly impressed with this. The lowering of the GPU clock under load is what I consider very minor, I mean 850 to 834 (most of the time) is not a biggie is it? It's got to be done and all seems to be working as it should, the GPU gets a little hot and scales back, cools down and kicks back up with all this happening every few seconds.

I mentioned earlier everyone seems to be fretting about whether 105C is too high and I'm saying... well read my previous post.

Interesting, before I say anything could you please do the exact test but after reaching the a stable point forcing the fan at maximum and check if the core frequency go back to 850.
This is very very important to understand if is related to the Temp or to the TDP in Watt due to the heavy load on the GPU, AMD PowerTune is designed to prevent the throttling, scaling a bit down the frequency in order to avoid a possible overheating with almost no impact on the performance, 15/20/30 MHz are nothing compared to put the GPU in a lower state (3xx MHz)

 

[nj-mac-user]

Let's throw in as much hyperbole as possible shall we?

Maybe you should read my post here:- https://forums.macrumors.com/showthread.php?p=20682691#post20682691

It looks like this applies to you.

Amen

 

[Astelith]

I forgot to say, if with lower temp the core is still down to 834 it means there's no thermal problem, and is PowerTune scaling down the clock for the predictive power consumption algorithm

 

[tmorterlaing]

I forgot to say, if with lower temp the core is still down to 834 it means there's no thermal problem, and is PowerTune scaling down the clock for the predictive power consumption algorithm

Let's throw in as much hyperbole as possible shall we?

Maybe you should read my post here:- https://forums.macrumors.com/showthread.php?p=20682691#post20682691

It looks like this applies to you.

So what you're both saying is; heat is fine and normal in this case, but expect a bit more noise than previous models?

Amen

Thanks for your input. I love a bit of bandwagon jumping, really productive.

 

[cynics]

Let's throw in as much hyperbole as possible shall we?



Maybe you should read my post here:- https://forums.macrumors.com/showthread.php?p=20682691#post20682691



It looks like this applies to you.

I hate to point out the obvious but wouldn't throttling (albeit only a small amount) indicate it's too hot?

Maybe I'm misread your post but if performance has to drop when you want it the most I'd say, yes it's too hot.

EDIT : Let me rephrase that it's not too hot it's just not being kept cool enough. Same but different.

 

[Astelith]

I hate to point out the obvious but wouldn't throttling (albeit only a small amount) indicate it's too hot?

Maybe I'm misread your post but if performance has to drop when you want it the most I'd say, yes it's too hot.

EDIT : Let me rephrase that it's not too hot it's just not being kept cool enough. Same but different.

We have 2 possible scenario:
1- the gpu is too hot and Powertune trigger the throttling prevention
2- we have reached the TDP limit (current absorption) and Powertune think that this workload is too heavy and it cut some frequency (regardless the temp)

 

[andy9l]

This just got interesting.

@Astelith, does the fairly constant load of this benchmark suggest it probably is heat causing the throttle? If it does turn out to be heat related, that's a hugely disappointing finding. Your 2700rpm test on Steve's iMac should give the answer.

Thanks for doing this Steve and Astelith. I'm sure there are other ways you'd rather spend your time.

 

[MacVidCards]

Ok, fair enough - a point well made. The fact still stands though, Apple have chosen to use a very hot ant GPU that sits close to other components in an all in one design. this means the fan speed goes up to 2300 (which is not the maximum btw - that's 2700) whilst gaming. This means Apple has basically said ' if you want to game, fine, but expect a fair bit of noise' - which is frustrating, and unnecessary - as proven by previous models.

Good luck.

I tried to point out that 100C and/or 105C are BOTH suicidal temperatures for a GPU and the items near it on a logic board and had my motives for pointing out obvious things questioned.

The "Reality distortion field" has been enlarged and now changes the laws of physics.

 

[Astelith]

Good luck.

I tried to point out that 100C and/or 105C are BOTH suicidal temperatures for a GPU and the items near it on a logic board and had my motives for pointing out obvious things questioned.

The "Reality distortion field" has been enlarged and now changes the laws of physics.

Analyze this readings...

subefotos

Why only the GPU is hot ? Look the other components, GPU Proximity at 54...
My Macbook air gets a way hotter than this (Cpu at more than 105°) and It's working all day since October 2012
Later I'll post a screenshot of it

 

[WilliamG]

Analyze this readings...
[url=http://s28.postimg.org/mubj8a5a5/Screen_Shot_2015_02_09_at_10_19_42_PM.png]Image[/url]
subefotos

Why only the GPU is hot ? Look the other components, GPU Proximity at 54...
My Macbook air gets a way hotter than this (Cpu at more than 105°) and It's working all day since October 2012
Later I'll post a screenshot of it

In that image, you can see your GPU hit 105C, just like everyone's does.

So, now that we've established that everyone's M295X GPU goes above 100C, the question still remains - how much damage, if any, this will do to the component/surrounding components?

The answer is - nobody knows at this point.

And Astelith, if your fan never goes over 1900rpm, something is wrong with your system. Every M295X system should hit 2300rpm under GPU/CPU load.

 

[Astelith]

In that image, you can see your GPU hit 105C, just like everyone's does.

105 is the top of the graph, look carefully, it never reach that high

----------

In that image, you can see your GPU hit 105C, just like everyone's does.

So, now that we've established that everyone's M295X GPU goes above 100C, the question still remains - how much damage, if any, this will do to the component/surrounding components?

The answer is - nobody knows at this point.

And Astelith, if your fan never goes over 1900rpm, something is wrong with your system. Every M295X system should hit 2300rpm under GPU/CPU load.

With 54 degrees of GPU proximity I see no burning apocalypse yet

My fan can spin even at 2400 when needed, if it is at 1900 it means that there's no need for more air

 

[andy9l]

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

105 is the top of the graph, look carefully, it never reach that high

The top of the graph is defined by the maximum temperature your GPU has reached in that session - it's a dynamic scale.

According to the iStat, your GPU did hit 105°C. Still lower than my 108°C record!

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

 

[Rockadile]

As a person that only owns the 2011 MBP, I'm curious how loud the 5k iMac fans are at max rpm because mine sounds like a plane engine

It looks like iMac tops out at 2300 or 2700 and my MBP is 6200. Most likely because of fan size.

 

[WilliamG]

The top of the graph is defined by the maximum temperature your GPU has reached in that session - it's a dynamic scale.

According to the iStat, your GPU did hit 105°C. Still lower than my 108°C record!

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

Exactly. 105 is the highest Astelith's GPU has actually hit in that session. And THAT is the conclusive proof that Astelith does not have some *special* "working" iMac that no one else has. That number has been highest 106C for me, that I've ever seen (I don't check completely obsessively). It's usually 105C, with an occasional spike to 106C.