AMD claims 95 Celsius is the optimal temperature for a GPU to perform for its lifetime.

RainMotorsports

Leadership
Leadership
Joined
Mar 6, 2011
Messages
8,649
Toms Hardware had an AMD graphics rep for an ask me anything and the thread is probably a pretty good read. Just look out for any posts by "Thrack" - Ask Me Anything - Official AMD Radeon Representatives - Radeon - Graphics & Displays

But this one interested me the most:
My favorite is the new implementation of PowerTune on the 290X and 290. There's a lot of doom and gloom around the 95C temperature, because people are used to a world where the product is designed to run as cold as possible... but that's not the world we're living in with these units. The doom and gloom is based on an old viewpoint.

95C is the optimal temperature that allows the board to convert its power consumption into meaningful performance for the user. Every single component on the board is designed to run at that temperature throughout the lifetime of the product.


If you throttle the temperature down below that threshold, then the board must in turn consume less power to respect the new temperature limit. Consuming less power means lowering vcore and engine clock, which means less performance.


You want to take full advantage of product TDP to maximize performance, and that is accomplished with a 95C ideal operating temperature for the 290 and 290X.


Even with a third-party cooling solution, like the Accelero 3 some users have started deploying, the logic of PowerTune will still try to maximize TDP by allowing temperatures to float higher until some other limit is met (voltage, clock, fan RPM, whatever).


It's so bloody smart and it kills me that more people don't fully understand it.

So rather then designing and releasing cards intended to give a consistent performance under normal temperatures they do indeed intend these cards to throttle the shit out of themselves in order to deliver the most powerful card on the market. Just with the catch "performance not garunteed". Nothing wrong with that its just the first time I have seen that as an absolute fact. There is however a difference between Turbo until 95C and heavy throttling.

But 95C? I call the life of a GPU about 3 years. Their viewpoint is if the core can withstand 95C for its lifespan then instead of holding back its clock speeds to allow for consistent performance we should burst it as much as we can. Best I remember at temperatures this high the chance of electrons propagating across 2 pathways (read, a short) increases drastically. But I am no electrical engineer so that's more of a question to something I remember.

Okay thats fine... But uh 95C? I don't think any overclocker would call that sane. I have a 30% overclock on my GPU and I am not insane enough to let the temps go that high. Between the fact that the 570 is in no way a top of the line chip and the damn good cooler it has on it I am fortune to not go over 70C. If they were not making so much sense I would say there is an excuse for why AMD runs so hot somewhere in there.
 
What are you bitching and moaning about this for? Like Nvidia doesn't do the same thing, but at 85C? They throttle the shit out of their cards so don't play like you've never heard of this before. It all comes down to cooling. If you have better than adequate cooling you get better performance.

And no, I don't give a shit about the reference coolers. Nvidia's done this for two generations now, ever since the 480 ran so hot so why keep going on about this?
 
What are you bitching and moaning about this for? Like Nvidia doesn't do the same thing, but at 85C? They throttle the shit out of their cards so don't play like you've never heard of this before. It all comes down to cooling. If you have better than adequate cooling you get better performance.
And no, I don't give a shit about the reference coolers. Nvidia's done this for two generations now, ever since the 480 ran so hot so why keep going on about this?

They have done this for Kepler, call it 2 generations if you want but the 500 series was not like this. For Kepler its pretty much the entire lineup receiving the treatment. Its not like they went out and tried to take advantage of just high end GPU's or get temperatures under control while remaining competitive.

My problem is illustrated well here:
throttle.jpg
This image directly matches the performance reported by an end user I posted about earlier. 1Ghz until the core heats up and then drop and as soon as the temp drops a degree then throttle the holy living shit out of the card.

Nvidia has done the exact opposite of AMD. AMD does not specify the base clock of a card, AMD specifies the maximum frequency. Nvidia specifies both and AMD's partners tend to do so as well since every partner card may have different clocks. The boost introduced with Kepler isn't even worth doing in the first place i mean a prescribed 58Mhz on the 780 Ti what does that even buy you? with AMD were talking a range of 200 Mhz and selling the card like audio companies sell Amps advertising peaks. Though in all fairness AMD's peaks would be relevant if you could ever reach them.

In looking at this apparently Nvidia has the same view point as me and is smear campaigning AMD as hard as I am lol.
 
So, 95C is the new "limit", but for how long? Most of us won't be maxing the cards at 95C all the time. I'm assuming the AMD rep is taking into account that the person using the card will eventually reduce the load on the GPU.
 
If you have a 290X reference cooled and are playing BF4 you will be. I mean only 5 minutes till crash tho

Sent from my SPH-D710 using Tapatalk
 
Someone made a joke of the GTX 480 I think it was with a diagram showing heatpipes connected to a kettle, but things are definitely getting up there :-) ... I think silicon or other materials have an effective maximum of 130 though, or thereabouts, so I think the closer you get the more dangerous it is.

The logic of "95C is safe, so we might as well run there 24/7" is the same as "my car outputs the highest HP at 5000 RPM, so let's run there on average." Yeah, but it comes at a cost of increased wear. Maybe it's not mechanical wear, and who knows, maybe you can really run it there for years or even decades with the current materials and production, but the risk is higher I guess...
 
yea, and im not hardware genius, but let's say you run that card at "optimal" performance temps 24/7 for a few weeks...95c...and you have a decent cooling system (air/liquid), does not that 95c put off heat to other components who live longer under lower temps??
 
I think silicon or other materials have an effective maximum of 130 though, or thereabouts, so I think the closer you get the more dangerous it is. Yeah, but it comes at a cost of increased wear. Maybe it's not mechanical wear, and who knows, maybe you can really run it there for years or even decades with the current materials and production, but the risk is higher I guess...

It was once put like this. You can put a CPU in an oven at 80C for 20 years. Put it back into the machine and it will work perfectly fine. You can not run that CPU at 80C for 5 years without expecting it to fail. 130 Celsius might have been survivable in the past but post High K Dielectric processes on the sub 45nm level likely not so well. The smaller the pathway the less current capability it has and the hotter it is even less so. Its like a thin wire the colder it is the higher the technical capacity. But heat is the enemy you take that wire and heat it up its not going to have the same performance.
 
Unless you have a hermetic seal for your case, 95C is a really dumb temperature to run at. It's probably a marketing trick so you'd buy everything from them, as any other components seeing that temp next to them would stop working.


You will essentially get a massive coat of charred dust on your motherboard which will overpower the coolers and eventually fail you. But i guess AMD will sell you a dust-filtered room just for that, and an amd NBC suit so you don't let any of the dust off from yourself.

Cool your shit, you aren't that rich.
 
Back
Top