Page 2 of 3
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 12:27 pm
by markdotgooley
One RTX 2060 and another RTX 2060KO. They're crammed just above a power supply with a tiny WiFi card between them. When they're both running work units, one (I forget which) is always over 70C and occasionally as high as 77C. The other is at least 60C and occasionally as high as 72. Lots of case fans in a big case; I wish there were an easy way to mount the GPU cards far apart but I can't think of one. Also the Ryzen 5 1600AF usually has 10 threads running a work unit and two to tend to the GPUs, stock AMD cooler that came with it, around 60C I think but could be higher.
I'm a bit worried about all this but only a bit. If they die, they die. Still haven't seen a full month's power bill and I expect this whole thing is using 12 kWh a day. I believe I'm paying US$0.10/kWh for what's probably from the local nuke plant...
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 1:23 pm
by bikeaddict
Recently, I've noticed a relationship between GPU clock speed and a few degrees temperature change on my GTX 1650 Super.
With the fan in auto speed mode, which stays at 25-35% fan speed, the GPU temp fluctuates between 57-62C but mostly stays around 60C. The clock runs at 1905 MHz in this temp range.
When I set the fan speed manually as high as 50%, the GPU clock increases to 1920 MHz and the temp stays in the 55-60C range.
With the fan speed at 55% or higher, the clock increases to 1935 MHz and temp stays under 55C. The speed does not increase beyond 1935 MHz even with the fan at higher speeds.
This appears to be NVIDIA's GPU Boost 3.0 feature adjusting the clock in 15 MHz increments. When testing the fan speed changes, it always takes a few minutes for the temp to stabilize before it will boost the clock.
Even though this is only about a 1.5% increase, it's a safe way to boost speed without risking bad work unit errors from regular overclocking.
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 1:28 pm
by NBR
bikeaddict wrote:Even though this is only about a 1.5% increase, it's a safe way to boost speed without risking bad work unit errors from regular overclocking.
I am doing this on my Mac, I am not sure how big of a difference it makes for my CPU.
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 4:06 pm
by Joe_H
NRT_AntiKytherA wrote:Amusingly it's hotter here in North Wales today than it is in Brasilia, I only looked because of that Core i5's temperature. However, it's an iMac (all-in-one?) according to NBR's profile so no doubt has a laptop processor and cooling system.
iMac CPUs are desktop models, the GPUs have been mobile chip based. Some CPUs even have been on socketed logic boards potentially allowing for a CPU upgrade for those willing to do the disassembly needed to get there. Have not checked the GPUs for the latest high end iMac's or the iMac Pros.
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 5:04 pm
by NRT_AntiKytherA
They are still essentially a desktop replacement laptop bolted to the rear of a screen and share many of the same cooling compromises. So using an i5 or i7 from the extensive range available with a high TJunction value is to be expected.
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 6:55 pm
by NoMoreQuarantine
So long as it is below the limit specified by the manufacturer (and above the dew point technically), the processor should be fine for a long period of time. As far as GPUs are concerned, a fan will probably fail before the silicon and the motherboard, PSU, or RAM will probably fail before a CPU. The exception is if you have an amazing cooling solution and are overvolting. In that case, you could be operating within the "safe" temperatures I mentioned, but have enough current to cause electromigration within the processor. Best case scenario, this will cause increased resistance across the internal circuits, therefore requiring higher and higher voltages over time to maintain the same clock rates, until eventually it will no longer be stable at any voltage or clock rate.
Re: Comfortable Hardware Temperature
Posted: Thu May 21, 2020 10:02 pm
by NRT_AntiKytherA
There's also the issue of M.2 slots for nvme drives in close proximity to PCI-E slots on motherboards. Anything over 55C is flagged as too hot for an EVO 970 for example by Samsung Magician software. They use PCI-E bus so the circuit layout partly dictates the location just as much as the required footprint but it's really a stupid place to have the drive when a high-end GPU is spewing out hot air onto that area. Unless you have decent cooling and no cables fouling air flow, you better be prepared for a slow performing and early failing nvme drive.
Before anyone says fit a heat spreader, they are next to useless. It's actually better to have direct airflow over the nvme rather than a lump of metal covering the chips on the drive absorbing more heat from the surrounding components.
Re: Comfortable Hardware Temperature
Posted: Fri May 22, 2020 9:10 pm
by MeeLee
Depends.
I've done a test on one of the TV boxes I own, and the bare chip gets hot really quickly (80C).
The factory cooling solution was a thick thermal pad, with a steel plate doing the cooling.
Reached perhaps 2-5C lower.
I replaced it with sticking on a cheap heat sink you can find on Amazon.
Results between the 'thermal tape sticky heat sinks', and a 'bare heat sink, using dual-sided tape' where identical. Temps dropped by over 10C.
And of course, a small fan added a lot more cooling.
A lot of people stick heat sinks on their SSDs for best cooling. Especially on the controller chip. The memory modules don't really get that hot.
Re: Comfortable Hardware Temperature
Posted: Sun May 24, 2020 12:22 am
by Paragon
I run my GPUs up to 85, no problems in my multi GPU rig (4 blower style cards in one tower). CPU is typically 50-60, but up to 95 overclocked. Still no problems...the constant load of F@h is better than things that thermally cycle the chips. The cycling causes more stress.
Re: Comfortable Hardware Temperature
Posted: Sun May 24, 2020 10:19 am
by FireFox-89
Paragon wrote:I run my GPUs up to 85, no problems in my multi GPU rig (4 blower style cards in one tower). CPU is typically 50-60, but up to 95 overclocked. Still no problems...the constant load of F@h is better than things that thermally cycle the chips. The cycling causes more stress.
Nice, got any pictures also what kind of PPD are you getting with that setup?
Re: Comfortable Hardware Temperature
Posted: Sun May 24, 2020 12:40 pm
by Paragon
FireFox-89 wrote:Paragon wrote:I run my GPUs up to 85, no problems in my multi GPU rig (4 blower style cards in one tower). CPU is typically 50-60, but up to 95 overclocked. Still no problems...the constant load of F@h is better than things that thermally cycle the chips. The cycling causes more stress.
Nice, got any pictures also what kind of PPD are you getting with that setup?
Ha! you would laugh if you saw this build. It's intent wasn't so much for PPD but for making heat for my house in the winter (we have solar panels so I wanted to build a solar powered folding at home computer space heater). I am working on an article on this for my blog.
Basically, this machine is an old nForce 680i EVGA board that has 3 PCI-E X-16 slots and one PCI-E X1 slot. I am running two old GTX 480 (fermi based) cards and a GTX580 in the X16 slots, and am using a riser extension to mount a GTX460 off the X1 slot (installed it with a custom bracket I made out of velcro and an old CD drive). So three of these cards dump heat out the back, one dumps heat out the drive bay. This is powered by a Core2Quad Q6600 to have one dedicated CPU core per GPU. It's got 4GB of DDR2 ram and is running Ubuntu. It's got a 1300 watt Seasonic power supply in it.
Average power consumption when folding full-tilt is 980 watts, measured at the wall. This makes it a pretty good space heater...I control the heat output by switching on and off the GPUs remotely using Teamviewer. Since this is running really old hardware, the efficiency is terrible...total PPD on the system is about 50K, so 50,000 PPD / 1000 watts = 50 PPD/Watt. Not so good.
The cool thing is, this entire machine and these GPUs were obtained for less than $150. This is stuff that was otherwise destined for the land fill. Instead of me using a conventional 1500 watt resisitive heat space heater in the winter, I can run this thing, and convert solar energy from my roof to heat for the room, while doing a small amount of science at the same time.
I have some pictures, as well as thermal images from my FLIR camera that are really awesome. I'm writing up that blog post so I'll link it up when it's finished.
Re: Comfortable Hardware Temperature
Posted: Sun May 24, 2020 6:13 pm
by FireFox-89
Haha that is sketchy and awesome
Re: Comfortable Hardware Temperature
Posted: Mon May 25, 2020 3:11 am
by Paragon
Here you go...I'll throw a thread in the hardware forums on this as well
https://greenfoldingathome.com/2020/05/ ... u-want-to/
Re: Comfortable Hardware Temperature
Posted: Mon May 25, 2020 9:41 am
by FireFox-89
Haha that is awesome
Re: Comfortable Hardware Temperature
Posted: Mon May 25, 2020 11:06 am
by NRT_AntiKytherA
that dust cake
, my main folding system (also primarily my gaming rig) is a nice space heater even at 380W power consumption which can get annoying in the Summer but is fantastic over Winter. I clean out my systems every 12 months even though I have an air purifier plus they have both the standard washable mesh and/or expanded polyurethane vacuum motor dust filters on all openings because finer lint still finds a way in eventually. I tried using HEPA style but it restricted the airflow too much, so I use vacuum cleaner motor filter material instead. Without all that I'd clean it out every 6 months to avoid the evil dust bunnies leaving their droppings everywhere and baking those GPU and CPU heat sink cakes.
Nice write up though and good to see kit avoiding landfill being put to a useful purpose other than just heating your room