M2050 cooling Passive cards in a non-server case

Apologies if this is not the right forum.

I have just picked up (nice and cheap) a couple of tesla M2050s for one of my cuda machines. These are passively cooled (unlike the C2050 I have elsewhere) and will sit in adjacent slots in a standard Corsair D800 case. I guess I am concerned about cooling. The case is well designed and already has ample fans but is there anything I should do to make sure these things do not go into meltdown?

Any advice would be welcome as long as it does not involve turning the machine off every 30 minutes :-)

Darrel

Hi Darrel.

I think you can get standardised heatsink/fan based coolers after market, this should help aleviate over heating. The passivity of the heat sink is designed to work with the airflow in blades, other 1U stuff, etc. It might be possible to emulate that somehow. I have read that overheating can damage these cards quite easily.

I have a question you might be able to answer. I have just bought a m2050 designed for HP, an sh885b. Any ideas if this will work in a non-HP pc? I was hoping to build something around it.

Thanks.

Adam

Hi Adam,

Sounds like we are in the boat!

The M2050s I just purchased are HP parts and I am running one (too nervous to try both at present!) currently in a custom built machine comprising an Asus Rampage IV extreme motherboard with a water-cooled intel 3960X and 64Gb of RAM. I have a 3Gb GTX580 driving the display.

The M2050s is getting hot but my CUDA kernals are intense short burst activities at present - this is for code development (we willl deploy over a large cluster via EC2 when the code is ready). I have just ordered one of these:

http://www.scan.co.uk/products/antec-vcool-graphics-card-cooler-keeps-graphics-cards-upto-15-degrees-cooler-extendable-3-speed

and one of these:

http://www.scan.co.uk/products/antec-super-cyclone-blower-exhaust-fan-for-hot-pcs-fits-in-2x-expansion-slots-3-speed

I will use the former adjacent to the M2050 heatsink and hope that it cools it a little, the latter will be used to help exhaust a little more air out of my Corsair D800 case (which is pretty well cooled already).

If this works sufficiently well I will deploy the second card in my server box (custom made on an asus W6 Supercomputer motherboard, intel 980X) alongside an existing C2050 and a GTX470 driving the display.

My office is getting rather toasty though,nice in winter but…

This solved my MPI under windows problem but gave me a few more headaches.

Darrel

Sorry should have added:

An additional problem with the OEM M2050s is that the drivers do not report card temperatures to software monitors i.e. there is no way to check the actual core temperatures. Apparently the cards talk directly to server hardware monitors and OEM supply software solutions to manage the fans etc. For me this will be guesswork unless someone else has any good ideas :)

http://www.tomshardware.co.uk/accelero-xtreme-plus-v6000-vf3000,review-32041.html - do you know any of these would fit the M2050? I thought the form factor of the recent GTX cards was similar to the tesla cards?

I will get on and build something around my card and report back to you how I get on. I understand you’ve kindly given me the specs of the systems you’re using the HP Tesla cards in, but what’s the main requirement? PCIe x16 Gen. 2? Is that the only stipulaion, or is there something else, chip sets, etc?

Thanks for your advice.

Adam

Hi Adam,

I don’t know whether the GTX cooler solutions would fit. If you did fit one you would need to remove the existing heatsink - I did not want to do this, hence I am looking to just try and improve the airflow through the existing heat sink. My GTX580 has a special Gainwood Phantom cooler which would probably not fit on the HP tesla - certainly you would need to be brave to try!

As you suggest I think PCIex16 gen 2 is the only requirement, there is nothing ‘extra’ special about the card itself just the usual tesla stuff i.e. ECC, availability of TCC mode etc. The lack of temeperature monitoring worries me but I am hoping it has some kind of internal overheat protection - i.e. it will just shut down before melting! My two rigs have been built to accommodate some heavy use of sequential code and large datasets as we try to migrate our code to CUDA, hence the high-spec CPUs. I have run the tesla C2050 on my rather more modest hardward at home without a hitch.

You will need a fairly beefy PSU ( mine are 1050W Corsair and a 1250W Enermax ) and a big case with good cooling ( again I have the Corsair D800 and a Lian Li P80B Armoursuit). These can take a lot of additional fans which I believe I am probably going to need :( . If you want to fit more than one card you will need a motherboard which has the PCI slots far apart - I suspect putting M2050s next to one another in a normal case would give some real temperature headaches.

I’ll let you know if mine melt!

Darrel

Hey guys,

I’m in a very similar situation at the moment with one m2050 in a desktop box. I’d be really interested to know if the antec graphics card cooler works and how well. At this stage the card I’ve got is going to be used for long-ish duration work (probably a few hours flat-out at a time) and I’m considering whether it would be worthwhile mounting a 60mm fan or two to blow along the length of the card (i.e. longitudinally) to try and replicate a 1U server environment. Do you think this would be worth the effort or is there a better way to do this?

Ben

check out hwinfo64. I first thought that there is no chance for M-series cards to report the temps because they directly send the sensor data to a IPMI capable motherboard. I read that somewhere in the internet. however hwinfo64 is able to read my GPU core temp. search under sensors. and they are plausible. 47° C water cooled at 100% computing. they rise while computing and go down when stopping the compute process.

about your cooling question. I first wanted to air cool my teslas m2070, however I ended up water cooling just the GPU unit. see my post:

it seems like water cooling the GPU only works quite well, my second card must have another issue so i decided to also cool the ram with water, but that unfortunately did not help ME, but the card is really well cooled now.

Sorry a little late in picking up these responses.

Thanks for the hwinfo64 tip - I’ll try that later in the week. The Antec PCI cooler seems to be doing a pretty good job at present but I don’t know how well the M2050 would survive continuous use - I guess I will find out shortly (it will certainly help if I can monitor the GPU temp). Snikpo: Mounting a couple of fans on the heatsink would probably be a good idea but I have no idea how they would mount - I would certainly be interested to hear how you get on if you go that route. As appealing as water cooling is I am unable to do that at present.

Darrel

Just realised I can continuous poll the GPU temperature using nvidia-smi -l (the nvidia-smi.exe is in my C:\Program Files\NVIDIA Corporation\NVSMI directory. This is a pretty useful utility, not sure how I missed it earlier.

Hi guys,
Firsty, thank you for pointing me toward hwinfo64, it works perfectly!

Otherwise, I’ve had a go at putting some fans on the heatsink.

I’ve used two 60x13mm fans (delta AFB0612HHC’s) blowing longitudinally along the graphics card and out through the back of the case (see pictures). I would desirably have had the fans mounted such that they blow directly off the heatsink (i.e. blowing perpendicular to the length of the card), but in my case I have about a 30mm gap between the Tesla’s heatsink and the graphics card next to it, so that isn’t really an option.

The mounting isn’t particularly professional for the time being, I’ve mounted one of the fans directly onto the metal base-plate that the card came with (the one that is down the end of the card near the power sockets) and the other on the outside of the case - to make a push-pull setup along the card.

In regard to performance; it is idling at 50C and under load seems to stabilise at around 70C. This isn’t a particularly thorough test, but it’s been running at load for the last hour and has only hit 70 once (seems to hold 68-69C). When I’ve made some simulations that can be run for many hours I will let you know how it handles it.

Ben



Brilliant idea - I’ve just purchased the coolers you mentioned, as well as the “Akasa Vortexx Neo AK-VC03-BLUV VGA cooler” which attaches to nVidia cards.

… And does it fit? My main blocker at the moment is finding a water block that fits the screws round the base of the main GPU.

Thanks.

I just received the card in the post today. I’ll see how it fits tonight. I don’t check into this forum regularly, so email me on support at designrem dot com.

In the end, after a month of fruitless attempts to get the M2050 working without overheating, I gave up an swapped them for C2050’s. They worked beautifully.

Greetings,

Could you please help me put together a cooling system for a tesla M2075. I have a dell t7400, and I
don’t know what to do.

Best regards,

Sorry to breathe life into an old thread, but I picked up an M2050 and have 3d printed it: a brace, a shroud and a duct to funnel a 120 mm fan into it. It now maxes out at 74’C when running at max. Picture of it all setup: Imgur: The magic of the Internet

I probably would have been better off going with a centrifugal blower fan, which would have been the only way to do it if I didn’t have the space under the card for the duct and 120 mm 2000 rpm fan.