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

[ioxakoyba]

Now that is something very weird you Astelith is showing! I have never seen such low idle GPU die temps on this machine like 39C. The idle temp is always something like 65-67C on my GPU die, just after the boot.

It may really be that contact on the cooler is poor. But this would mean it's the same thing with most of the riMacs if this is true.

I wish the machine was easier to open. I would gladly repaste this with my Arctic Cooler 5 Silver paste.

 

[boast]

Use the osd on screen installed with afterburner (https://gaming.msi.com/features/afterburner)
Honestly I don't know if mine stutter, I'm not a gamer, I play just a couple of hours a week or so, I enjoy the performance but maybe it's me less demanding.

For the record, my idle temp are very low:
[url=http://s16.postimg.org/vjnwe5wd1/Screen_Shot_2015_03_18_at_9_48_35_PM.png]Image[url]

If I were you I will try to go in an apple store and ask to repaste the GPU/heatsink, we should have the same components, what can make such a difference ?

Wow, some of those temps are lower than my apartment temp!

 

[Astelith]

Now that is something very weird you Astelith is showing! I have never seen such low idle GPU die temps on this machine like 39C. The idle temp is always something like 65-67C on my GPU die, just after the boot.

It may really be that contact on the cooler is poor. But this would mean it's the same thing with most of the riMacs if this is true.

I wish the machine was easier to open. I would gladly repaste this with my Arctic Cooler 5 Silver paste.

I think the problem is manufacturing side, not design, I cannot give other explanations, the only way to verify without losing the warranty is to make Apple do the job and verify, show them this thread with my temps

----------

Wow, some of those temps are lower than my apartment temp!

And I have like 24C here, I'm in t-shirt, not in the freezer

 

[WilliamG]

Use the osd on screen installed with afterburner (https://gaming.msi.com/features/afterburner)
Honestly I don't know if mine stutter, I'm not a gamer, I play just a couple of hours a week or so, I enjoy the performance but maybe it's me less demanding.

For the record, my idle temp are very low:
[url=http://s16.postimg.org/vjnwe5wd1/Screen_Shot_2015_03_18_at_9_48_35_PM.png]Image[/url]

If I were you I will try to go in an apple store and ask to repaste the GPU/heatsink, we should have the same components, what can make such a difference ?

I think that screen shot tells everything. Every single temperature you have is lower than mine. It's NOT just the GPU. It's the CPU, the TCON, your hard drive etc. Everything.

I think either Astelith has some excellent ambient temps in his location, or there's something weird with his readings, because it's all much, much lower than my temps. Heck, even my hard drive is much higher, reading 36C right now. Every temp is lower on Astelith's system by 7-10C. That one you really can't explain away with manufacturing tolerances or thermal paste.

 

[sugawi]

I think the problem is manufacturing side, not design, I cannot give other explanations, the only way to verify without losing the warranty is to make Apple do the job and verify, show them this thread with my temps

----------



And I have like 24C here, I'm in t-shirt, not in the freezer

I live not far from this place:
http://m.yelp.com/biz/core-care-macintosh-support-specialists-sacramento-3

I can take my 5k with m295x for rattling fan repair. At that place I can talk directly with person who does repairs. I can show that firefly website where my 5k reaches 105C in seconds. When they'll have it open, what should they check in terms of manufacturing?

 

[Astelith]

I live not far from this place:
http://m.yelp.com/biz/core-care-macintosh-support-specialists-sacramento-3

I can take my 5k with m295x for rattling fan repair. At that place I can talk directly with person who does repairs. I can show that firefly website where my 5k reaches 105C in seconds. When they'll have it open, what should they check in terms of manufacturing side?

Check the connection of the gpu chip with the heat sink and the correct application of the thermal paste

----------

I think that screen shot tells everything. Every single temperature you have is lower than mine. It's NOT just the GPU. It's the CPU, the TCON, your hard drive etc. Everything.

I think either Astelith has some excellent ambient temps in his location, or there's something weird with his readings, because it's all much, much lower than my temps. Heck, even my hard drive is much higher, reading 36C right now. Every temp is lower on Astelith's system by 7-10C. That one you really can't explain away with manufacturing tolerances or thermal paste.

Maybe my sensors are crazy, who knows, but ambient temp is pretty high here, I hate low room temps, I'm in t-shirt all the time when home

But If you have constant 70C in the GPU it wouldn't surprise of an average higher components temperatures, the GPU is soldered into the main board in a narrow space

 

[sugawi]

Check the connection of the gpu chip with the heat sink and the correct application of the thermal paste

Thank you for advice, will mention that before they do repairs.
It must be something that most of 5k affected. My previous base model reached 85C in seconds, so 20C difference seems reasonable for stronger CPU and GPU.

 

[steve62388]

I think that screen shot tells everything. Every single temperature you have is lower than mine. It's NOT just the GPU. It's the CPU, the TCON, your hard drive etc. Everything.

I think either Astelith has some excellent ambient temps in his location, or there's something weird with his readings, because it's all much, much lower than my temps. Heck, even my hard drive is much higher, reading 36C right now. Every temp is lower on Astelith's system by 7-10C. That one you really can't explain away with manufacturing tolerances or thermal paste.

I think you have hit the nail on the head there, because everybody else's temps are in the ballpark what you are seeing.

 

[blufrog]

Those with M295X - how about doing what Astelith did and post your IDLE temps?

Idle temps are the #1 give-away of poor thermal paste application (as I'm sure everyone knows - if you try and run the system without a heat-sink attached, it will fail within seconds).

I think this will resolve any discussion immediately. At idle, you shouldn't see anything above ~40 deg. C (CPU or GPU).

 

[LVEB]

any idea for the performance between

old iMac 27
3.5GHz Quad-core Intel Core i7
1TB Fusion
16gb ram
NVIDIA GeForce GTX 780M 4GB GDDR5 support CUDA . mercury playback

new riMac 27
4.0GHz Quad-core Intel Core i7
1TB Fusion
16gb ram
AMD Radeon R9 M295X 4GB GDDR5

price different is around $700+/-

My only concern is the GPU & CPU cause most my work need
Adobe apps, PR (HD edit/MagicBullet effect../4K soon) ID / PS & LR (some AE)

Please advice which machines worth to upgrade according my usage.
anyone experience btw above setup ?

 

[ioxakoyba]

I think I am considering this theory possible: As idle temps on Astelith's machine are much more lower - and also his iMac's cooler seems to be able to cool down the GPU much more effeciently (for example keeping GPU die temp between 93-94 C on gaming when fan set to 2700rpm, comparing to mine and most of the others that still suffer temps like 106 C and throttling), I am thinking that there is something wrong in the iMac retina's cooler heatsink's fastening and/or the used thermal paste (= meaning the contact between GPU die and heatsink must be very poor in most of the riMacs).

Also as this kind of product tolerance is allowed in most of the iMac 5k's Apple outputs with outrageous prices, I believe it does not much help to contact Apple service for repairing this, because they just will test the machine and as it's within their tolerances, they will return it back to customer as it is.

So this is just my theory, but I believe now that if I open my iMac I will see a chunk of poorly applied white thermal paste between GPU and heatsink just like in these iFixit pictures:

(Maybe for some unknown reason there are some unit's that do have a different thermal paste, or applying it is done properly, like Astelith may have. Not know for sure, though, but Astelith's temps really would suggest something like this.)

So I am really considering that instead of sending this thing back to Apple service to be done nothing, I am going to order a replacement display attachment sticker and opening tools and open this thing and check the cooler's heatsink attachment and repaste the CPU, GPU and memory die contacts with proper Silver compound. (At the same time I could replace also 1 TB SSD inside this bought much inexpensive afterwards than from Apple). iFixit provides really good instructions here.

Of course this is insane that after you buy a 3500 euros computer the first thing is that you need to disasseble it to thousand pieces to make it run properly. Also this is not something that average Joe could do, but I think in my case I've got quite a good background for this considering my last building project here.

So we'll see. I will keep you posted in case I will do it. In the meantime further tests still have to be done.

 

[CrankItUp]

I made a few tests: I manually set the fan speed to the lowest setting -> 1200 RPM (with Macs FanControl), then i started the PixMark Julia FP64 (IMHO the best GPU stresstest out there, just try it - sth like the valley benchmark is peanuts compared to) just to see whats going on with the temps, clocks etc..

The first pic shows the gpu idling.

On the big screenshot you can see that Macs FanControl shows the true RPM (the fan speed in the GPU-Z screenshot shows wrong)

Real heavy throttling started at 108°C (see the 2nd GPU-Z screenshot, that was 2 sec after scratching the 108° maximum) - but even then all ran smooth

If fans are manually set to 100% (=2700 RPM or 2300 RPM auto-controlled) - the max temp is 105°C as usual - clocks are fluctuating between 784-834 Mhz

Verdict:

A) The difference between 2700 RPM/ 2300 RPM and 1200 RPM @ full load is 3°C - so what?

B) If the fan is auto-controlled it wont ever reach 108°C - so you wont notice heavy throttling anyway

 

[andy9l]

The difference between 2700 RPM & 1200 RPM @ full load is 3°C - so what?

Exactly. That *is* the problem.

You've got a GPU that can only push the fan to 2300rpm, when even 2700rpm doesn't come close to cooling it enough.

In all except Astelith's case, the cooling system is insufficient for the GPU inside. The GPU and CPU throttle themselves to avoid overheating when under significant and sustained load. Simple, as, that.

 

[Astelith]

I made a few tests: I manually set the fan speed to the lowest setting -> 1200 RPM (with Macs FanControl), then i started the PixMark Julia FP64 (IMHO the best GPU stresstest out there, just try it - sth like the valley benchmark is peanuts compared to) just to see whats going on with the temps, clocks etc..

The first pic shows the gpu idling

On the big screenshot you can see that Macs FanControl shows the true RPM (the fan speed in the gpu-z screenshot shows wrong)

Real heavy throttling starts at 108°C - but even then all runs smooth

If fans are set to 100% (2700 RPM) - the max temp is 105°C as usual - clocks are fluctuating between 784-834 Mhz

Verdict:

The difference between 2700 RPM & 1200 RPM @ full load is 3°C - so what?

As expected, the heatsink/fan is clearly not doing his job, but not because of his design, it seems that the GPU is not transferring the heat to the heat sink

----------

Exactly. That *is* the problem.

You've got a GPU that can only push the fan to 2300rpm, when even 2700rpm doesn't come close to cooling it enough.

In all except Astelith's case, the cooling system is insufficient for the GPU inside. The GPU and CPU throttle themselves to avoid overheating when under significant and sustained load. Simple, as, that.

And assuming (and it would be absurd the opposite) that we have the same component, there's something wrong with the manufacturing, we should ask to the iFixit guys how did they found the contact point between GPU and heatsink

 

[ioxakoyba]

By the way, I must correct myself about the idle temps. The reported 67-69C idle I said before was in Windows 10 and it wasn't actually idling, but it had browser plus some other tasks in background going on.

But I did further tests, that also gave me some more info about idle temps. First I ran Metro Last Light and I didn't even get to play, but instead set the settings plus saw the intro for couple of minutes and clocked nice 105 C top temp for the GPU in GPU-Z as you can see in the attached picture (temp_win10_after_metro_last_light.jpg). The fan was also spinning 2700 rpm set by Macs Fan Control.

But then I exited and something in like ten minutes the idle temps were quite low, ca. 38 C for CPU and ca. 45 C for GPU. And these should be core die temps, not any proximity or heat spreader values. (See temp_win10_idle.jpg).

I also screenshotted OS X HWMonitor as attached (temp_osx_idle.png) and it seems that GPU die is 51 C on idling and CPU around 40 C. So I cant see much difference on idle temps after all compared to Astelith.

Still it's very weird that nearly all of the users except Astelith has very similar heat profile than me (105-106C on full load in just a minutes). It's more like that these are the normal iMac retina machines and Astelith had some defect manufacture (that actually cools down properly!). Did they maybe forgot the put that horrible heat paste on the CPU and GPU dies on your case? It should cool down better then even with the dry contact Just joking...

 

[Astelith]

By the way, I must correct myself about the idle temps. The reported 67-69C idle I said before was in Windows 10 and it wasn't actually idling, but it had browser plus some other tasks in background going on.

But I did further tests, that also gave me some more info about idle temps. First I ran Metro Last Light and I didn't even get to play, but instead set the settings plus saw the intro for couple of minutes and clocked nice 105 C top temp for the GPU in GPU-Z as you can see in the attached picture (temp_win10_after_metro_last_light.jpg). The fan was also spinning 2700 rpm set by Macs Fan Control.

But then I exited and something in like ten minutes the idle temps were quite low, ca. 38 C for CPU and ca. 45 C for GPU. And these should be core die temps, not any proximity or heat spreader values. (See temp_win10_idle.jpg).

I also screenshotted OS X HWMonitor as attached (temp_osx_idle.png) and it seems that GPU die is 51 C on idling and CPU around 40 C. So I cant see much difference on idle temps after all compared to Astelith.

Still it's very weird that nearly all of the users except Astelith has very similar heat profile than me (105-106C on full load in just a minutes). It's more like that these are the normal iMac retina machines and Astelith had some defect manufacture (that actually cools down properly!). Did they maybe forgot the put that horrible heat paste on the CPU and GPU dies on your case? It should cool down better then even with the dry contact Just joking...

You are joking, but who know what happened to my 5K, maybe is that

For instance, I've wrote a comment in the iFixit section of the Heatsink asking a feedback about the connection parts, maybe they will help us understand.

If it's only a paste problem or a wrong assembly position of the heatsink is something fixable

 

[blufrog]

54 deg. C ***AT IDLE*** (just sitting on the desktop with nothing running I presume?) strongly suggests a problem with the thermal paste.

It is not surprising that under load it is hitting 106 deg. C. It's not simply that it is starting from a higher idle temp, but the efficiency is greatly reduced, so it takes less for it to get hotter.

Thermal paste, if too much is applied, acts as an insulator. It must be applied as a very thin layer (it's purpose is to fill the voids in the heat sink material). It should be applied as thinly as possible prior to fitting it to the device.

The amount shown in the iFixIt photo is excessive and will create serious problems.

If people can show high idle temps (CPU and/or GPU) then IMHO this constitutes a manufacturing defect, and requires rectification.

 

[Astelith]

54 deg. C ***AT IDLE*** (just sitting on the desktop with nothing running I presume?) strongly suggests a problem with the thermal paste.

It is not surprising that under load it is hitting 106 deg. C. It's not simply that it is starting from a higher idle temp, but the efficiency is greatly reduced, so it takes less for it to get hotter.

Thermal paste, if too much is applied, acts as an insulator. It must be applied as a very thin layer (it's purpose is to fill the voids in the heat sink material). It should be applied as thinly as possible prior to fitting it to the device.

The amount shown in the iFixIt photo is excessive and will create serious problems.

If people can show high idle temps (CPU and/or GPU) then IMHO this constitutes a manufacturing defect, and requires rectification.

100% agree with you, 3°C delta from 1200 to 2700 fan speed it really seems the GPU is insulated, you point out the right word to explain this.

 

[CrankItUp]

The whole thread about this issue again begs the following questions:

Is that really a problem? Because everyone is affected (except Astelith)

Are ten thousands ore more sold rImacs in the last past months incorrectly produced?

If the engineers reckoned on an early failure they wouldnt have greenlighted it, would they?

or is this a masterpiece of planned obsolescence?

Give the answer to yourself - choose what sounds reasonable for you

 

[Astelith]

The whole thread about this issue again begs the following questions:

Is that really a problem? Because everyone is affected (except Astelith)

Are ten thousands ore more sold rImacs in the last past months incorrectly produced?

If the engineers reckoned on an early failure they wouldnt have greenlighted it, would they?

or is this a masterpiece of planned obsolescence?

Give the answer to yourself - choose what u prefer

This machine will unlikely broke in suspicious numbers and by the end of the year it will be old compared to the best Retina iMac ever
It's a mix of the two plus the fact that very few will notice a higher fan noise or the GPU temp, most of the users don't even install monitoring tools and they are living in peace

 

[blufrog]

It doesn't matter what the engineers think - it's a manufacturing/quality assurance problem.

Excess/incorrectly applied thermal paste should be taken seriously by Apple. Not only will it result in poor performance under load, but it will result in premature failure of the components affected.

 

[CrankItUp]

It doesn't matter what the engineers think - it's a manufacturing/quality assurance problem.

Excess/incorrectly applied thermal paste should be taken seriously by Apple. Not only will it result in poor performance under load, but it will result in premature failure of the components affected.

All products are manufactured under certain guidelines - if they aren`t assembled according to these the workers will be fired. Period. There is a quality control before the final acceptance.

I dont think that neither every 5k mac is wrong assembled nor that its wrong designed. Maybe it has not the best cooling solution, but its not faulty by design..

 

[blufrog]

Did you actually look at the iFixIt photo, showing the hugely excessive thermal paste that was applied?

The high idle temps people are observing are confirming this.

 

[CrankItUp]

Did you actually look at the iFixIt photo, showing the hugely excessive thermal paste that was applied?

And? Do you think thats the reason?

I assembled many windows computers in the past and had many overclocking sessions. And i can say in all conscience: The difference between excessive and hardly visible applied thermal paste is not as much as you believe..

They could have set the max temps at 90°C with lower stock clock speeds.. so why the heck have they set the limit to 108°C ? Because apparently it doesnt matter...

 

[Astelith]

And? Do you think thats the reason?

I assembled many windows computers in the past and had many overclocking sessions. And i can say in all conscience: The difference between excessive and hardly visible thermal paste is not as much as you believe..

They could have set the max temps at 90°C with lower stock clock speeds.. so why have they set the limit to 108°C ? Because apparently it doesnt matter...

The target temp is 100°C with a spin up fan point at 96°C, in the hot units (so except mine and others) the fan reaches maximum speed at 100° and for this unknown reason, the heatsink can't cool the GPU enough and so reaching the 108°C triggering the frequency reduction of AMD Power Tune to keep the temperature stays below this point (fortunately, with a very minimal loss of performance).
In my 5K, the heat system can handle to keep the GPU at 100°C by playing only with the fan ranging from 1200 to 2300 and helped by the TDP prediction of Power Tune that uses lower frequencies for high demanding tasks to keep the power draw into the limits (and so the temperatures).

The cooling system is obviously not the best on the market but it can be sufficient, and is by default the same for everybody, to summarize :

- we have the same system specs
- we have the same heatsink

why different temperatures (and different fan speed/noise) ?