What are the typical temperatures for a single, high-end GPU at different loads? I have a 7850 and I'm trying to keep it a little quieter at half and full loads; however, I'm unsure at what point cooling will become an issue. At idle, the GPU hovers around 35 w/ GPU fans running at 25% and is not audible at all. At half-load, the temps are ~50 w/ GPU fans at 35 - 40% and is barely audible. At full-load (>95% load), the temps hover around 73 w/ GPU fans running at 55% and is about as audible as a PS3 / XBox360...so not loud, but not quiet enough for me :). Should I be concerned with these temps? Can I drop the fan speeds a little more at full load? Or would I be risking an overheat?
Also, I was wondering what type of effect a CPU cooler has an on GPU's temps. I'm running a stock cooler currently, but thinking about upgrading to a Noctua NH-L12 since I took out the OD. Would a CPU heatsink / cooler upgrade help at all? Or will it only really help the CPU and RAM? How about the case fans? Would I be better off running those a little faster and bump down the GPU fan speeds? Or I should possible try and upgrade the GPU fans (if possible)? Not sure about the GPU fans since I've never replaced those before.
BTW, here is my setup: Silverstone GD05 | Core i5 3570K @ Stock | Asus P8Z77-M Pro | 8GB Corsair Vengence | MSI R7850 Twin Frozr (2GB OC) | SeaSonic M12II 620W | 256GB Corsair M4 SSD. For cooling, I have the stock Intel cooler and stock GPU fans; however, I'm running 3 x Slip Stream PWM fans as case fans.
Anyways, I know its a lot of questions, but any help would be greatly appreciated. Thanks.
Typical GPU temps at Idle, Half-Load, and Full-Load
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
-
Das_Saunamies
- *Lifetime Patron*
- Posts: 2000
- Joined: Tue May 15, 2007 1:39 am
- Location: Finland
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
1) For typical temperatures: http://www.techpowerup.com/reviews/. Pick the chip/card type you want, read as many reviews as needed. The data is there.
2) GPUs typically tolerate very high heat loads, anything up to 60-70 is no big deal for a modern chip. The rule is the same as always: cooler electronics last longer. For example my shitty cooler and very efficient GPU start at 30 °C idle and peak out at 60-ish. All my cards have been roughly in the 30-80 ballpark, and the hottest, an EVGA GTS 8800 512, fried itself after just a few hours of gameplay when a case fan failed to start (because Asus mobo control was teeeeerrible back then).
2.1) There's no "safe" high temperature. Other card components suffer from the heat, and there's a lot of those.
3) 70+ °C is on the high side and may not stabilise (run a loooong burn test). It might be typical of your chip though. Consider a better cooler or more CFM if you want silence - or just gamble it. Bigger fan may help a little - if you can even swap one in - but ultimately it is about the heatsink's ability to absorb and dissipate the GPU's heat.
4) CPU cooler can radiate a little heat on the card or block air flow in a non-standard internal layout, but that's it. No major effect; it's moreso the card that will heat up the case air and impede the CPU cooler's ability to cool the CPU. GPU > CPU when it comes to heat production in modern systems.
5) Stock to Noctua I would swap in a heartbeat. The default CPU heatsinks are adequate at best, the fan is usually bad (my guess is this is amplifying the annoyance in this case). My wager is it will not help your card, as it won't be able to blow cool air on the cooler or exhaust its hot air.
6) Case fans are easy to experiment with. Put system under high load and run the case fans as high as you can tolerate (try around 800 RPM), then determine if the coolers are good enough (parts overheating or not). If the coolers can stabilise, you can lower the fans' RPM and test again.
It's all about TESTING. Individual circumstances and rigs vary too much for there to be absolute answers, but the principles of comparing data (which you're already doing, good job) and testing your setup for optimal performance apply.
PS. Run Afterburner's OSD and put on GPU temp and FPS. You can monitor performance as well as your risk of frying an expensive piece of equipment.
PPS. For that last leg of the Silent Path - acoustic materials. Bare metal is prime real estate for amplifying and reflecting noise sources.
Edit 1: Fixed an instance of RPM instead of CFM.
2) GPUs typically tolerate very high heat loads, anything up to 60-70 is no big deal for a modern chip. The rule is the same as always: cooler electronics last longer. For example my shitty cooler and very efficient GPU start at 30 °C idle and peak out at 60-ish. All my cards have been roughly in the 30-80 ballpark, and the hottest, an EVGA GTS 8800 512, fried itself after just a few hours of gameplay when a case fan failed to start (because Asus mobo control was teeeeerrible back then).
2.1) There's no "safe" high temperature. Other card components suffer from the heat, and there's a lot of those.
3) 70+ °C is on the high side and may not stabilise (run a loooong burn test). It might be typical of your chip though. Consider a better cooler or more CFM if you want silence - or just gamble it. Bigger fan may help a little - if you can even swap one in - but ultimately it is about the heatsink's ability to absorb and dissipate the GPU's heat.
4) CPU cooler can radiate a little heat on the card or block air flow in a non-standard internal layout, but that's it. No major effect; it's moreso the card that will heat up the case air and impede the CPU cooler's ability to cool the CPU. GPU > CPU when it comes to heat production in modern systems.
5) Stock to Noctua I would swap in a heartbeat. The default CPU heatsinks are adequate at best, the fan is usually bad (my guess is this is amplifying the annoyance in this case). My wager is it will not help your card, as it won't be able to blow cool air on the cooler or exhaust its hot air.
6) Case fans are easy to experiment with. Put system under high load and run the case fans as high as you can tolerate (try around 800 RPM), then determine if the coolers are good enough (parts overheating or not). If the coolers can stabilise, you can lower the fans' RPM and test again.
It's all about TESTING. Individual circumstances and rigs vary too much for there to be absolute answers, but the principles of comparing data (which you're already doing, good job) and testing your setup for optimal performance apply.
PS. Run Afterburner's OSD and put on GPU temp and FPS. You can monitor performance as well as your risk of frying an expensive piece of equipment.
PPS. For that last leg of the Silent Path - acoustic materials. Bare metal is prime real estate for amplifying and reflecting noise sources.
Edit 1: Fixed an instance of RPM instead of CFM.
Last edited by Das_Saunamies on Sat Dec 22, 2012 10:20 am, edited 1 time in total.
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
+1 on the above.
FYI, newer cards will throttle the clocks when they get too warm. This threshold varies by GPU, but will typically be 90C or higher.
FYI, newer cards will throttle the clocks when they get too warm. This threshold varies by GPU, but will typically be 90C or higher.
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
The highest I've ever had is 117C. A Geforce FX5900 with the shim coming unstuck after a heatsink swap. No damage done, it just throttled back. After the loose shim was removed the card has been running fine for the last 6 years.
Realistically get up to 150-200C and something WILL burn.
Realistically get up to 150-200C and something WILL burn.
-
Das_Saunamies
- *Lifetime Patron*
- Posts: 2000
- Joined: Tue May 15, 2007 1:39 am
- Location: Finland
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
If we're talking instant-fry then yes. Neglecting prolonged heat exposure is not a smart move - maybe the card will take it, maybe it won't. My EVGA card fried with a fully-working, monitored and clean cooler, and it's not the only story of that kind.edh wrote:Realistically get up to 150-200C and something WILL burn.
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
What fried on it? Did you complain about the failure? The card has built in clock throttling and fan boost control so really should be protected from such a failure. If this has not been triggered then I would think that it is either a bad design of some part or a part was faulty anyway.Das_Saunamies wrote:My EVGA card fried with a fully-working, monitored and clean cooler, and it's not the only story of that kind.
-
Das_Saunamies
- *Lifetime Patron*
- Posts: 2000
- Joined: Tue May 15, 2007 1:39 am
- Location: Finland
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
Couldn't find any clear signs on the PCB. Nothing had split, no spills or scorch marks or discolouration of any kind. It just gave up the ghost and never started up again in any machine. RIP
I didn't RMA since I thought it was my fault. The fan was as loud as always (oh those horrible days), so I didn't even realise the case fan was off - until I noticed it wasn't spinning down as usual when I jolted to look at the case rear for rising smoke. Rage was had.
PS. I bought the card for its extended (lifetime?) warranty thinking that would surely guarantee quality parts. It fried up after a year of use, so I wouldn't say there were faulty parts per se, it had been through enough hard use to weed those out. It just couldn't take the heat, was my conclusion.
I didn't RMA since I thought it was my fault. The fan was as loud as always (oh those horrible days), so I didn't even realise the case fan was off - until I noticed it wasn't spinning down as usual when I jolted to look at the case rear for rising smoke. Rage was had.
PS. I bought the card for its extended (lifetime?) warranty thinking that would surely guarantee quality parts. It fried up after a year of use, so I wouldn't say there were faulty parts per se, it had been through enough hard use to weed those out. It just couldn't take the heat, was my conclusion.
Re: Typical GPU temps at Idle, Half-Load, and Full-Load
Thanks for all the great information. As soon as the holidays are over, I plan on picking up a NH-L12. Once I do that, I go through the testing to see how quiet I can get my machine while maintaining efficient cooling.