The Official NVIDIA GRID M40 Thread

NotSoRandomJoe

New Member
Dec 1, 2017
13
1
3
40
its a waste for AI they are too slow for any purpose really unless you get them for $200. just buy a gtx 1070.
I can't imagine they would be too slow, maybe limited by the 4GB address space and the CUDA 5.0 sure, but not speed.

The 1070 offer up to 8GB and CUDA 6.1 (which means you'll be just behind the next Tensorflow when released since they've standardized their builds on the nvidia TX platform (aka CUDA 6.3) yea they've got me tongue-in-cheek about this too.

This reminds me of another question I forgot to ask here.

Can you virtually SLI these cards and run all 4xGPUs as 1 single config?
 

NotSoRandomJoe

New Member
Dec 1, 2017
13
1
3
40
I doubt there is OpenCL access to these cards so you'll have to install CUDA then compile ethminer with CUDA support.

Note that I haven't purchased one yet, but I'm fairly certain this will be the case.

What is the ethminer command line you're attempting to use on this card?
I might be wrong about the lack of OpenCL support.

The HP driver listed in this thread clearly states OpenCL.

https://support.hpe.com/hpsc/swd/pu...740c61e3f9724e678408d0060a&swEnvOid=4231#tab1
 

LukeP

Member
Feb 12, 2017
175
18
18
41
I can't imagine they would be too slow, maybe limited by the 4GB address space and the CUDA 5.0 sure, but not speed.

The 1070 offer up to 8GB and CUDA 6.1 (which means you'll be just behind the next Tensorflow when released since they've standardized their builds on the nvidia TX platform (aka CUDA 6.3) yea they've got me tongue-in-cheek about this too.

This reminds me of another question I forgot to ask here.

Can you virtually SLI these cards and run all 4xGPUs as 1 single config?
ok well if you are just doing it for fun no probs. just buyer beware.
 

Kirill Lebedev

New Member
Jun 12, 2017
12
5
3
San Francisco
OpenCL libraries are installed together with Cuda on Ubuntu. The card looks like 4 GM107 cards. ML is not so crazy slow as you can think about it. And actually, it is great in terms of TF code debug if you actually need to do some multi-gpu development. TF sees it as 4 gpus as expected.

In terms of vGPU it is hard to say will it work properly on Windows combinations, but Xen sees it as Tesla M10 and allows to pass it to VM by separate cores. Ubuntu guest has no problems to pick it up with Cuda driver
 
  • Like
Reactions: NotSoRandomJoe

NotSoRandomJoe

New Member
Dec 1, 2017
13
1
3
40
It would be great if anyone could do an ETH miner test on a single GPU and post Mh/s and Wattage results.

I already know what I'm getting on the AI side in regards to CUDA 5.0
 

chilipepperz

Active Member
Mar 17, 2016
212
63
28
50
It would be great if anyone could do an ETH miner test on a single GPU and post Mh/s and Wattage results.

I already know what I'm getting on the AI side in regards to CUDA 5.0
You don't do eth on these cards. Monero and cryptonight = best on Maxwell. Then neoscrypt. Then Equihash. Then some randos then Ethash.

I have one in a server I don't have power monitoring on, but it's still earning $3/ day using nicehash and it was like 39W per core x4 last I looked.
 

NotSoRandomJoe

New Member
Dec 1, 2017
13
1
3
40
You don't do eth on these cards. Monero and cryptonight = best on Maxwell. Then neoscrypt. Then Equihash. Then some randos then Ethash.

I have one in a server I don't have power monitoring on, but it's still earning $3/ day using nicehash and it was like 39W per core x4 last I looked.
Thank you for the reply and additional information, that looks like great news.

However, the insinuation that ethminer wont' work on a card that supports OpenCL/CUDA 5.0 is equivalent to spreading false information.

Real data vs the interpretation of data is preferred.

I'm still interested in seeing this test if anyone is willing to spend the 10 minutes on it.
 

PharmEcis

New Member
Jan 25, 2018
1
0
1
43
Hey all, first post. So I picked one of these up to try out and I seem to either be stupid or missing something simple. EWBF cuda miner is only showing about 240H/s doing zcash. It shows all 4 gpus but shows each doing ~60. I'm running Server 2016 using driver linked here specifically for 2016. From what I gather, this card should do a bit better for XMR / Zcash, correct? I pulled 4x 1060's out of a DL580G7 to toss the M40 in to test, so I'm going to test the card on a clean install now on a second G7 to make sure it wasn't a driver issue. When I put the 1060s back in, they kept stopping after a few minutes until I wiped all video drivers and installed clean with proper drivers.

Any advice would be appreciated.
 

jtylis

New Member
Jul 22, 2017
1
0
1
36
Hey guys,

Now that the Grid M40 can (apparently) run on Win 10 with the help the Win Server 2016 drivers from HP, and given its' awesome price on ebay, I've been thinking of snatching one up. Thing is, I'm wondering if my concept of use, as below, is valid:

I'm looking to use the card on CAD applications, mainly Rhino 3D. Does it make sense to have a Grid M40 handle the number crunching and renders + have a basic Quadro to drive the monitor? Is this even possible technically? Also, since it's four GPUs in one card (4x640 CUDAs) will I get the total 2560 CUDAs to work, or will the active program (eg. Rhino) use only 1 GPU? (in which case I'll be better off with a single newer and cheaper 640 CUDA / 4 GB - P1000)?

Thanks for the help!
 

DASHIP

New Member
May 4, 2016
10
0
1
51
Has anyone gotten the VDI drivers working for these? Would love to be able to use it as vGPU's for vDGA or vSGA in ESXi or a with a VMWare View virtual desktop cluster.