Tesla K80 24GB x3 rendering

I have more problem to setup gpus with amd cpus than intel cpus.

My new build, 96 cores intel xeon platinum.
100% everything is modded/OC even the motherboard :innocent:
Installed win10 for workstations and all 8 cards was detected no need of uninstall/reinstall wtf :joy::rofl:

2 Likes

Nice one:).
About the OSā€¦ I use just Windows 10 Home x64, could that be a problem?
What would you suggest, try to just install drivers until they finally work?
I also found this: https://www.nvidia.com/en-us/geforce/forums/game-ready-drivers/13/252042/conflict-geforce-and-tesla/
suggesting to use this: https://www.guru3d.com/files-details/display-driver-uninstaller-download.html.
Looks a little bit scary but maybe that would work. Also inside the DDU folder I found the ā€œissues and solutions.txtā€ and it says:

To solve the infinite boot loop issue (stuck in Safe Mode) / password lost that some users are having, you must boot with the DVD/USB (Windows installation disk) in recovery with Command prompt OR restart the computer (not the reset button) and HOLD THE SHIFT KEY until you have the advanced option menu and the option to boot with a Command prompt and type :

Option 1 (recommended) : bcdedit /deletevalue {default} safeboot

  • Replace ā€œdefaultā€ with the identifier of your operation system. (can be seen when you run bcdedit)

If option 1 is not working,

Option 2 : bootrec /rebuildbcd

There may need to be some adjustement if you have a dualboot system. but if not , it should work correctly.

From my personal experience, the UXTHEME patcher was a possible cause of this.

Safe Mode may not work especially with the ā€œWindows 10 Anniversary Updateā€ when CSM in the bios is ā€œDisabledā€. (Microsoft issue)

The ā€œinfinite boot loop issueā€ seems just the problem I had.

Have never use a home edition.
It probably works, its only 2 gpus :+1:

I have CSM problem with one of my Amd Threadrippers.
I can install win10 on a 3x nvme raid0 CSM ON.
but after reboot
CSM is off in bios and it dont find any bootdrive.
AMD :sweat_smile:

Donā€™t know how but I did it!:smiley:
Well ok, I used the DDU first to clear all GPU drivers. After that I pretty did all you wrote before but in Windows safe mode and didnā€™t have any problems.
After trying to boot the system normally I did have a blue screen again couple of times but this time I decided to trust the Windows repair/recovery menu and let it do some stuff, it took him a while but finally got to the Windows logon. Both GPUs are detected in device manager, HWinfo and Blender:).
Now time for some testing.

Great :+1: happy rendering :ok_hand:

Tesla k40 is really powerful and lowcost on ebay.
Its 15% more powerful than GK110.
And its easy to biosmod up to 1200mhz with low voltage.
Overclocked tesla k40 GK180 is like 45% faster than a original tesla k20.

I think Iā€™ll stick to K20 as long as it works. I did it more as an experiment (to be honest I donā€™t need much rendering power but it is always good to have some). This particular Tesla cost me 237PLN (about 65$), which is equal to free in those strange times:D. Of course it would be way easier and more effective to just get another 1660 but right now it is hard to get ANYTHING in reasonable price.

One more thing about the temperature - do you know what is the maximum safe temperature for GK110?
Right now I have 30-35C in idle, when I start rendering or computing, it slowly goes up, so far it went up to 82C and I stopped the rendering as I didnā€™t want to check the limits :wink: .
Fortunately the temperature doesnā€™t go up rapidly so maybe I could just underclock it a little. Since I have medium tower, I donā€™t want to play with any sophisticated cooling, Iā€™m using a 120mm fan at 1100rpm with an air duct.

Oh, and a strange thing - it seems that my Tesla started to work when I changed the ā€œAbove 4G decodeā€ to DISABLED, opposite to what I have read over the internet. I changed it to ā€œenabledā€ for a while just to make sure and I started to have blue screen again.

Optimal is 80c , 95c maximum

Some teslas work with 4G disable.
K40 K80 need 4G enable.

Thank you, I see mine is throttling the clock frequency at 91C so at least it is not going to burn.

I also wonder if it is a good idea to get rid of the plastic casing from Tesla. The radiator could be additionally cooled by another fan on the side of computerā€™s case but I donā€™t know how the main airflow through the Tesla would be affected without the casing.

I had removed plastic and put a fan on the gpu cooler.
Save the plastic if u sell card later.

Mount the fan with plastic stripes.
It look like sh.t but works.

Amazing!! Incredible setup!!!

I am still assembling my system, it is not easy to put all the cards and make it look nice, my RTX 2080ti is triple slot, plus the four Tesla k80 cards ā€¦ I am going crazy: Yes but also enjoying a lot! ^^

I hope I will upload photos soon to show you.

A question Retrotrollet, which liquid cooling do you recommend for the four k80 cards?

Greetings!

BYKSKI waterblock for K80 is incredibe good quality.
I use 3 of them.

  • one EK 420x60 + 2x EK 120x60.
    The pump is a standard 10watt.
    This is for 2xcpu+ 8gpus.
    Gpus and cpus is around 45-50c when rendering and really silent :+1:
    With aircooling the pc sounds like a airplane.
    I upload cinebench test :yum:

Not yet finisch with this build, waiting for some stuff.
Going to be really cool with leds in all waterblocks even cpus and fans, like 40-50 diff color chemes

2 Likes

If u mount some gpus vertical u get more space around pcie, and its cool :sweat_smile:

I didnā€™t expect removing the casing would give such good results - the temperature now doesnā€™t exceed 78-88C. However, the ambient room temperature makes even bigger difference - when simply open the window (it is about +5-8C outside) the Teslaā€™s temperature goes down in seconds.

PRO-TIP (or already a well know fact;) ): Teslas have ECC capable VRAM, I doubt anyone would need this feature for rendering so it can be disabled to gain some extra performance (about 2% in my case).

Great, going to do some benchtest after work :ok_hand: