Tesla K80's in Dell R720 Server?

Hi,

I want to share this link with you https://dell.to/3n3rZON as it is a similar topic.

The K80 works on R730 but the R720 may not be working. I could not see the K80 in their spare parts list for the R720. Apart from the any video card you will install, I would also like to draw your attention to power supplies. When you install a video card, please make sure that it meets the power. https://dell.to/3gzrZDO

Let us know if it helps!

Thanks for the reply Erman.

Yeah, the K80's aren't in the R720 parts list as they are unsupported.  They are on the R730 parts list since they are officially supported on that model, but I don't have any at this time.  I'm good on the power, as I stated the R720's are currently running with dual K20's, I just need faster, that's why I have the K40's.  The K80's are spares, but hey, if I can get them to work, then they're going in.

As an update, I drilled down deeper in the university's computer cluster specs.  Deeper in the documentation the reference to R720's magically turns into R730's.  My guess is that the R720 reference in the cluster spec overview was either a typo or someone just mistakenly referenced the wrong model number.  I was bummed.

Still waiting to hear about actual attempts.  Surely I'm not the first person to have all of this equipment sitting on a bench and wondered if it might actually work regardless of what the "official" docs indicated.

So, the answer to my original question is:  YES

You can run a nVidia Tesla K80 in a Dell R720 Server.

I have a R720 on the bench running a single Tesla K80 at this time.  Everything appears to be functioning correctly and operating as normal.

I will attempt to install a second K80 after some tests complete.

I will update as I get more info.

James

Hi, 

Which OS are you using? I use Win10 and the GPU reports Code 12. I use one Dell R720 server. 

Hi.  That server has Windows Server 2019 DataCenter on it.  I have no idea how Windows 10 would run on a R720 with GPU's.  Sorry for the late response, but I just saw your message.  Did you ever get your Code 12 error resolved?

Thank you very much for bringing up this topic.

I will go ahead to buy R720 and two Tesla K80! 

What did you have to do to get the 720 to boot?
mine sits at a single prompt after post... never boots

I verified I set within the BIOS above 4gb under the Integrated Devices > video area....

is there something else I need to do?
I plugged it into the x16 slot and powered from the slots GPU to the card... with a 8pin to 8pin

Hi.  What happens if you leave the card in the machine, but unplug the cable to the GPU and try to boot it up?  Does it halt still, or does it continue to boot up normally?

I successfully deployed the R720 system with two Tesla K80, however, the 2 Tesla cards cannot reach their theoretical performance (such as double-precision FP64 should be 1.37 TFLOPs for a single GPU, but only 0.9 TFLOPs or 3.6TFLOPs for 4 GPUs as shown in the AIDA64 GPGPU Benchmark). If you have any idea to solve this problem, please share it. Thank you.

I successfully deployed the R720 system with two Tesla K80, however, the 2 Tesla cards cannot reach their theoretical performance (such as double-precision FP64 should be 1.37 TFLOPs for a single GPU, but only 0.9 TFLOPs or 3.6TFLOPs for 4 GPUs as shown in the AIDA64 GPGPU Benchmark). If you have any idea to solve this problem, please share it. Thank you.

p/s: I am using Windows Server 2016.

Some things for your consideration:

1.   How EXACTLY are you testing this?  Let us know and we can check our own machines to get some benchmarks.

2.   What is your ambient air inlet temperature of the server?

3.   What are the GPU core temperatures at 100% with nvidia-smi?

4.   Keep in mind that the system and the K80s have shutdown temps.  If the K80s start getting too hot they will throttle down to try to prevent this by slowing down.

5.   While I have never tried this, I have seen where you can "overclock" and "underclock" the K80s from the commandline.  Might want to lookup how to check this and see if yours have been changed from the default settings.

If I think of anything else I will add it later.

After being blank squares all day your images finally started displaying for me.  Kinda clears up how you are testing.  Now to sort out the results.

Hi, thanks for choosing Dell. I don’t see installing a R720 system with two Tesla K80 in our guidelines. That is a lot of modification from the original I’d have to say. Sorry I can’t be of help this time but we could wait to see what other users on the community may have experienced. Wish you a good one. 

1.   How EXACTLY are you testing this?  Let us know and we can check our own machines to get some benchmarks.

>> It is clear now as you see.

2.   What is your ambient air inlet temperature of the server?

>> 20'C

3.   What are the GPU core temperatures at 100% with nvidia-smi?

>>

4.   Keep in mind that the system and the K80s have shutdown temps.  If the K80s start getting too hot they will throttle down to try to prevent this by slowing down.

>> The system worked under the shutdowm temp.

5.   While I have never tried this, I have seen where you can "overclock" and "underclock" the K80s from the commandline.  Might want to lookup how to check this and see if yours have been changed from the default settings.

>> I checked (nvidia-smi -q -i 0 -d CLOCK) and it shows "On" as a default setting. Based on what I found (for example here https://xstream.stanford.edu/specs/ ), it seems like the results from AIDA64 is the performance at Base Clocks, not GPU Boost Clocks. So, I wonder the best way to benchmark FP64 performance.