About 2 x GPUs and heat problems

[markdotgooley]

How are you controlling the temps on the cards? I really recommend power limiting as a first step and looking at what the fans are really doing on an like MSI Afterburner.

I'm on Ubuntu Linux but I guess I could use nvidia-smi to force lower power consumption. But consumption is a little over 150W on each 2060 and I'm not sure how much I can lower it. 125W maybe? Lower? Fans around 85%: maybe I can force 100%? (Not much experience with this tool yet.)

Here's one card, excerpt of data from nvidia-smi -q :

Power Readings
Power Management : Supported
Power Draw : 152.06 W
Power Limit : 160.00 W
Default Power Limit : 160.00 W
Enforced Power Limit : 160.00 W
Min Power Limit : 125.00 W
Max Power Limit : 160.00 W

So is 125W the minimum? Will dropping power to 125W on each card reduce the temperature much? Will givve it a try later...

[HaloJones]

It also depends a lot on which units you get. I've seen GPU utilisation sometimes as low as 60% and as high as 99% just down to the science it's being asked to do.

[Mxyzptlk]

My 2 GPU’s are both twin fan EVGA’s. The upper is a 2060 Super and the lower a 2060. The GPU temps have not really altered with or without the CPU folding / the CPU air cooler running (a Noctua D15). The upper 2060 fans are running at 83%, is power limited to 75% and is right now 63c. The lower GPU fan is at 70%, is power limited to 65% and the temp is 51c.
I have always heard that the upper GPU gets hotter from the lower GPU fans blowing its hot air towards the upper GPU... Which makes sense with what i hav seen.

[bruce]

Having GPUs that exhaust the hot air out the back helps, too. Those that just circulate the hot air in the case should be avoided.

[HaloJones]

A GTX1070 and a TitanX powered by a G4400.

Liquid really is the only way with dual gpu

With single, you can make any card into a hybrid liquid cooled card with an NZXT G10/G12 and a used AIO.

Just blew up the motherboard

Oh well, chance to upgrade again. This was a B150, just won an auction for a Z170 so hopefully a little bit better and easier to debug as it has a debug LED.

[rickoic]

I've pretty much stopped using enclosures for my computers. Gone completely open air computing. Just stumbled on to open air mining frames which now lets me stagger my gpu's from on board to ribbon risers and that keeps several inches in-between gpus and helps cooling them much easier.

[markdotgooley]

I've pretty much stopped using enclosures for my computers. Gone completely open air computing. Just stumbled on to open air mining frames which now lets me stagger my gpu's from on board to ribbon risers and that keeps several inches in-between gpus and helps cooling them much easier.

Tempting. Need a motherboard with 3 or 4 nicely-spaced PCIe 3.0 or 4.0 x16 slots. Maybe an old dual-Xeon although those eat power. Put it in a rack-style setup or something makeshift.

[KtC]

I went back to Folding with a single 1060 and just realized the temperature @ stock is 76C (without any need to mod the settings with afterburner). It is not a card with a high end cooling solution, just two fans and radiator size gets two slots. Fan at around 56% I do not feel this is loud (and obviously that is a very subjective thing). Most important thing is that I do not feel any temperature discomfort when I touch the case on the back.

Even when I lowered the power of 2x1070ti setup and kept them running @ about 70C (I think that the Afterburner shows temp on a chip die), there was a huuuge difference in my PC case temperatures. With that setup I could barely touch the metal back of my PC case due to temp. IMHO cooling solution used on that 1070s was a top on the market with three fans and very big radiators. It seems the thermals on such powerful GPUs are very nonlinear processes comparing single card setup to combo.

You can see in the photo how tight that setup was.
https://imgur.com/a/YYJWpnE

[Paragon]

I've been running various configs of dual, triple, and quad air cooled GPUs in mid tower cases. The key for me was using blower style cards that exhaust 100 percent of the air outside the case, and adding cool air intake fans to the side panel to feed the cards directly.

I applied this to an old fermi-based GPU "space heater" here.
https://greenfoldingathome.com/2020/05/ ... u-want-to/

Cooling diagram is in the gallery at the bottom of the article

[MeeLee]

Zotac used to have a great open cooler design.
Their design had the fins in the length of the GPU, forcing the air out the back and front of the card, while some of the air went through it and exhausted on the sides.
It's like half a blower type, half an open type.
Unlike most modern GPUs that are fully open, and force air on the top/bottom of the GPU.

[HaloJones]

A GTX1070 and a TitanX powered by a G4400.

Liquid really is the only way with dual gpu

With single, you can make any card into a hybrid liquid cooled card with an NZXT G10/G12 and a used AIO.

Just blew up the motherboard

Oh well, chance to upgrade again. This was a B150, just won an auction for a Z170 so hopefully a little bit better and easier to debug as it has a debug LED.

Tried again this morning after an hour of using an iPhone11 as a microscope to better see the bent pins in the cpu socket and some gentle manipulation, the motherboard now works again! Shame I won that auction. Now I'll have to build another folding rig

[Skajaquada]

I had the same issues and choose the other way
My RX5700xt was too loud and hot for my wife, so i decide to change on custom watercooling.
After that it was so cold that i decided to buy another RX5700xt and a bigger Radiator.
Now it became a Corsair 780T with 2x 280mm and one 360mm Radiator which cool the 2 overclocked rx5700xt and the Ryzen 7 3700x down to 48°C max (Water temps at 38° max)
And yes, it´s so silent, my wife now wants her own watercooled System

[jrweiss]

Having GPUs that exhaust the hot air out the back helps, too. Those that just circulate the hot air in the case should be avoided.

Unfortunately, most GPUs will NOT exhaust [most of] the cooling air through their own backplanes. Even those that have slotted backplanes often have the cooling fins oriented parallel to the backplane, so airflow is mainly directed back into the case.

I recommend you use slotted backplane covers adjacent to the GPUs, and have positive pressure airflow (more intake than exhaust CFM) so that hot air will tend to go out those slotted backplanes.

[agent71]

I don't have 2 x RTX 2060 but am running 1 x RTX2060 and 1 x P106.

Temps are high but "fine" and in a case that isn't exactly great with airflow - Corsair Carbide 275R. It is pretty quiet though as replaced all fans with Noctua and placed CPU radiator at top of case to allow maxmium, unrestricted airflow in via front panel. It's still not really enough but ok.

You can see that the RTX peaks at 81C running at 150W and the P106 at 59C which is bearly an increase from idle temp...

I was power limiting both via nvidia-smi but it made very little difference to temps on the P106, +/- 3 degrees, so I let that run at whatever power it needs.
RTX at 125W sees temps peak around 75C.

As the 2060 runs hot and the PC sits in the home office I pause it via cron during working hours and fire up at 9PM although still limit to 150W. So not exactly a solution to improving cooling but one that means I don't care about the noise the fans make as I'm not in the room.

Code: Select all

0 7 * * 1-5 /usr/bin/FAHClient --send-pause 1
0 21 * * 1-5 /usr/bin/FAHClient --send-unpause 1

[MeeLee]

I don't have 2 x RTX 2060 but am running 1 x RTX2060 and 1 x P106.

Temps are high but "fine" and in a case that isn't exactly great with airflow - Corsair Carbide 275R. It is pretty quiet though as replaced all fans with Noctua and placed CPU radiator at top of case to allow maxmium, unrestricted airflow in via front panel. It's still not really enough but ok.

You can see that the RTX peaks at 81C running at 150W and the P106 at 59C which is bearly an increase from idle temp...

I was power limiting both via nvidia-smi but it made very little difference to temps on the P106, +/- 3 degrees, so I let that run at whatever power it needs.
RTX at 125W sees temps peak around 75C.

As the 2060 runs hot and the PC sits in the home office I pause it via cron during working hours and fire up at 9PM although still limit to 150W. So not exactly a solution to improving cooling but one that means I don't care about the noise the fans make as I'm not in the room.
...

the P106 uses a lower TDP than the RTX GPU.
Limiting the P106 to 125W, is setting it 5 Watt higher than it's stock rated '120W' TDP.
If I'm not mistaken, the P106 can be lowered all the way to ~75W; but it'll suffer in performance at those power levels.
I believe somewhere around 80-85W is the optimal setting for that GPU, but you'll have to play around with it, looking at GPU frequencies while power capping to different settings, to see if you can sacrifice 10% of GPU frequency (usually for ~25% of lower power settings).
10% of GPU frequency drop, doesn't always relate to 10% lower PPD. It more or less equals like 5% of PPD loss.
So instead of a 333k PPD, you'll get like 315-320k PPD.
Not a biggie, if you look at the power saved.