Tesla M2090

atypicalguy

Obliviot
I am running 295.41 or whatever the latest download is.

Nvidia-settings gives temp info on the quadro 600 but not the tesla.
 

atypicalguy

Obliviot
On a side note that I'm certain has been discussed ad nauseam somewhere by gamers:

What would the target max operating temp for an M2090 GPU be? Is there consensus? If so, based upon what? The glass transition temp of the circuit board material?

Thanks.
 

Tharic-Nar

Senior Editor
Staff member
Moderator
Supposedly, the max temp is around 100-120C, but that's more like the critical temp when things go funky. Ask any gamer though and they get very uncomfortable around 90C. Also, that's core temp, typically memory or VRMs go before that, but they don't have any temp monitors.
 

atypicalguy

Obliviot
Just wondering whether to monitor exhaust temp or stick the thermistor way down at the base of the heat pipes somehow. I will probably end up with a couple of those small fans blowing into one end on a fan controller.

Supposedly, the max temp is around 100-120C, but that's more like the critical temp when things go funky. Ask any gamer though and they get very uncomfortable around 90C. Also, that's core temp, typically memory or VRMs go before that, but they don't have any temp monitors.
 

Tharic-Nar

Senior Editor
Staff member
Moderator
Using a probe will result in much lower recorded temps, which means you'll need to compensate for that. I'm not certain, but I think 80-90C on the core would be about 60C on the heatsink.

Placing the probe on the heatsink is a better option than exposed to exhaust air, dust will give inaccurate readings after a while. Most probes in fan controller packs come with tape for the sensors anyway. Placing the probe on the underside of the heatsink would be ideal, away from direct airflow.
 

atypicalguy

Obliviot
Ideally I would find a thermostatic multiple fan speed controller with definable hysteresis parameters for 12v supplied power and 12v fans. I have yet to find one; most are just manual speed control with temp readouts. Better than nothing I suppose, but sort of basic considering the hardware...Anyway putting the temp sensor on the heat sink is imperative for an on/off controller; there is no exhaust if the fan is off.
 

Tharic-Nar

Senior Editor
Staff member
Moderator
Useful, but limited to 2 amps. That's enough for one of your fans.

Aquacomputer Aquaero 5 LT - that is fully programmable, complete with USB interface and can be used for water cooling too. Each fan header can support 1.65 amps for a total of 5 amps across 4 headers. You can buy additional header mounts for up to 10 fans I think. This is a new release, so can't say much. More for water cooling, but if you decide to go that route in the future, it might be worth it (or even for Psi).

Lamptron FC Touch - Which we've reviewed.
30 Watts per channel (so over 2 amps), up to 6 channels. (120 Watts total). Ask Optix for more details if you need them (he's the one that reviewed it).
 

atypicalguy

Obliviot
Thanks for that.

The mini squirrel cages only draw 0.12 amps, so if one or two of them can be made to do the job by applying them efficiently directly to the end of the heat sink, I am OK with the coolerguys.

The Aquacomputer LT seems like it would need a Windows machine to set it up, which I have, but I am crunching in Linux, so it would be a pain to have a separate laptop just to reprogram the thing, or reboot. The screen version is also $225 or so, whereas I am out $50 including shipping for the black box with the nice LED readout.

Other than that it looks like a very capable device. Thanks for the info; who knows where all this will end up.

Useful, but limited to 2 amps. That's enough for one of your fans.

Aquacomputer Aquaero 5 LT - that is fully programmable, complete with USB interface and can be used for water cooling too. Each fan header can support 1.65 amps for a total of 5 amps across 4 headers. You can buy additional header mounts for up to 10 fans I think. This is a new release, so can't say much. More for water cooling, but if you decide to go that route in the future, it might be worth it (or even for Psi).

Lamptron FC Touch - Which we've reviewed.
30 Watts per channel (so over 2 amps), up to 6 channels. (120 Watts total). Ask Optix for more details if you need them (he's the one that reviewed it).
 

atypicalguy

Obliviot
This is a closed equation system. I know the card draws 225W. Max temp is 90. Ambient air is 70-75. How much air flow do I need to dissipate 225 joules? Time cancels out and the answer should be in cubic meters (per sec) Need to know heat capacity of air I guess.
 

atypicalguy

Obliviot
It looks like I need about 30cfm to maintain a difference of 15 degrees btw inlet and outlet at 225W, sea level, 50% humidity. Micro fans are looking inadequate at 9cfm, unless I can run four of them somehow.

All the squirrel cages that blow that kind of volume are over 50db, as are the motors I have.

Not looking good for air cooling this thing at reasonable noise level.
 

Psi*

Tech Monkey
I am pretty sure that "90" is 90 deg C which is 194 deg F. The 60 deg C that my C2070 idles at is 140 deg F ... for convenient reference. These cards run very warm.

Your ambient is undoubtedly 70 deg F.

Squirrel cage fans are hardly worth it. Maybe consider larger fans with reducers like these. There are all sorts of bendy duct work in that link also.

Also, go wander around the local Home Depot or Lowes. Maybe you can figure out or get the inspiration of how to make some kind of plastic shell to fit over the heat pipe heat sink that will duct the air flow across & thru the fins ... with a larger fan into a reducer (above).
 
Last edited:

atypicalguy

Obliviot
Ha! Thank you for that. It seemed like a lot of airflow (!)

At least the error was in a conservative direction...so ambient is 24degC, which means a maximum usable temp differential of 65degC - gulp! The good news is, the hotter it runs, the less cfm are required to cool it.

I take your point on big fans and ductwork, but the space is pretty limited at the end of the card. I only have one 16x slot so I can't move the Tesla to a better spot in the case.

Bottom line is the olc-fan 9cfm micro squirrel cage might actually do the job. That would be great, because that fan is not terribly loud at 22dB (not to mention it is in the mail already) But the fins will have to get pretty hot before it begins removing heat at the proper rate. My plan is to plug it into the Coolerguys cheapo binary temp controller; hopefully it will only have to run when the card is active, but at 22dB it can run all the time without headache if required.

If that doesn't do the trick, I will probably order one of these and use a speed controller to throttle it back: http://www.sofasco.com/products/dc_cross/df40.html

At that point, I will have spent enough on various options to pay for a water rig. Certainly with multiple GPUs requiring many fans, water would be the way to go. I am still hoping to get by with the low profile rodent fan option though.
 

Psi*

Tech Monkey
I am considering this vga cooler. I think it will fit in place of the OEM heat pipe heatsink. I am looking for competitive products. I don't mind water cooling but not sure that the M2090 merits the effort given my experience with the C2070.
 

DarkStarr

Tech Monkey
I added my 280 to my 360 and now only run my 280 with 140s on it (a pair in pull) and my 360 has 2 200mm fans above it in pull. I may add some CM R4s on the insides since they are dang quiet and then my loop would be really chilly (vs air) and still very quiet.

I had been running 2 deltas on the 360 rad at 2k rpm moving a large amount of air with ridiculous static pressure. (2k is a low as they could go and they were really loud even then) I call them fingerslicers :p 4k rpm at full speed, it is not fun to touch one at that speed......

I am considering this vga cooler. I think it will fit in place of the OEM heat pipe heatsink. I am looking for competitive products. I don't mind water cooling but not sure that the M2090 merits the effort given my experience with the C2070.

Grab this, Arctic Cooling is AWESOME: Newegg Link My 5750 used a much smaller AC cooler (made for HIS) and it was awesome. Stayed cool enough to clock it to the wall no problem.
http://www.newegg.com/Product/Product.aspx?Item=N82E16835186048
 
Last edited:

atypicalguy

Obliviot
Is this basically a bolt on deal or what? Anybody know the M2090 gpu bolt pattern? I haven't pried my case off yet to check. I think I saw it posted somewhere so I'll do another search.

I added my 280 to my 360 and now only run my 280 with 140s on it (a pair in pull) and my 360 has 2 200mm fans above it in pull. I may add some CM R4s on the insides since they are dang quiet and then my loop would be really chilly (vs air) and still very quiet.

I had been running 2 deltas on the 360 rad at 2k rpm moving a large amount of air with ridiculous static pressure. (2k is a low as they could go and they were really loud even then) I call them fingerslicers :p 4k rpm at full speed, it is not fun to touch one at that speed......



Grab this, Arctic Cooling is AWESOME: Newegg Link My 5750 used a much smaller AC cooler (made for HIS) and it was awesome. Stayed cool enough to clock it to the wall no problem.
http://www.newegg.com/Product/Product.aspx?Item=N82E16835186048
 

atypicalguy

Obliviot
The arctic unit is definitely high capacity, but it is PWM. I think most graphics cards have a PWM out jack regulated by GPU temp. So far, I have not seen a jack like that on an M2090 and have reason to believe there is no temp sensor on the card.

so...

you would need a separate fan control for that unit, assuming it would bolt onto a 2090, which it probably would, though whether it would do so with the existing heat spreader plate in place is an open question.

The size of the AC extreme 3-fan seems to be an issue also; length being the primary concern in reviews. But an M2090 is already pretty long, so I'm guessing that it is no longer than the card, i.e. fine.

I'm also guessing the dual plugs will clear the heat AC sink, as the 8-pin on the end is still only single plug height, and the other 6-pin goes in on top of the unit.

3 fans is probably overkill. A single 92mm fan on a good set of fins is probably fine for this card. I'm guessing there isn't a lot of data to suggest it will crunch numbers better at 55 degC than at 70 deg C. The fans can certainly run slower rpms on the 3-fan unit, though, at any given temperature.

If I knew any of these vga coolers would bolt straight on to my unit with existing heat spreader plate in place, I would go for it.
 

DarkStarr

Tech Monkey
I was going to reply to the 3 fans overkill statement but then I realized this isn't OCN or it would probably already have been watercooled.

OCN, where overkill is nothing and nothing is overkill.
 

Madmodman

Obliviot
Hello everyone, first time poster here!

Id like to chime in here and throw out my solution for the ongoing watercooling of the tesla m2090....

I came across this site/thread/posts via google while tryin to research for a better/different cooling solution on this card....

First and foremost, a GTX 580 waterblock will NOT work with this card. The way nvidia has arrayed the mosfets and memory is different on the tesla. Remember, a Geforce card like that has/needs more mosfets than the standard tesla for the reasonings of it has output display, 2D mode, low and high power 3D mode to handle. More are just required when compared to something that is meant to just crunch numbers.

I found the pictures provided by PSI of the broken down m2090 (thank you!!) and the universal block solution from Swifttech via Typical guy- This is what got my gears turning....and yes this is still a fresh project, even im not 100% complete yet.

THIS IS AT YOUR OWN RISK!!

Two Solutions:

Going to swifttechs site and taking a close view at the block, it looks as though its a great possibility assuming you tend on leaving the OE vram/mosfet heat spreader intact. Going over everything that is required, swifttech wants you to purchase a mounting plate for the block(kinda dumb but ok- should come with it). This COULD cause a fit issue, but with the two different ways to mount the screws. That should allow you to play with the height and such if needed. The OE backplate that is attached to the card will provide a great mounting area.

I am unsure exactly how well the MCW block will perform on this card due to the fact it DOES have a small riser on block, thus creating a small gap. (I ordered a NON-AMD one, will confirm gap) I suppose you could fill it with thermal paste, but this is a $3K card, I feel we can afford to do it right....Time will tell once the block arrives in my hands.

Solution 2:

I have a older/spare EK Supreme HF waterblock for a 1366 socket cpu laying around that I recently have taken apart, wouldnt you know it, the Spreader side of the block is a PERFECT fit over the teslas GPU. The part that is in the way now is the initial mounting bracket that the block came with. Line up the bracket using the spreader half of the block as a point of reference while its sitting on the gpu. Mark the bracket, and Dremel it down to size so that it fits in between the gap of the vram/mosfet heat spreader. Be careful NOT to remove so much material from the bracket that you dont have room to drill your new 4 mounting holes.

Once you have it dremeled down, Mark your new holes in compliance with the 4 main holes surrounding the GPU on the PCB(outter ones!). Once you feel you have it all lined up, drill your new mounting holes. Reassemble your block and line it up when finished. If done right, this block should fit perfect between the gap on the OE vram/mosfet heat spreader!


Modifying(this is needed for either solution): THIS IS ALL AT YOUR OWN RISK!

Ok, so to get either of the blocks mounted, we are going to have to open up the 4 main holes that surround the GPU. On the back side of the card, nvidia supplies a smaller heat spreader for the vram containing a total of 9 torx head screws. You will have to remove this plate, putting the focus on the 4 torx head screw holes that surround the GPU. Open the 4 holes up more via a drill bit. Start small, and work your way up to allow your mounting screws to fit through(wont take much). Confirm this on the PCB side of the card as well(DO NOT drill into the PCB!!). Once you have this completed and you are happy with the results. ENSURE you have wiped it all down and that there are no tiny metal shavings on the plate. From here, you can reuse, replace, or upgrade the paste/pads for cooling the vram. Re-install when ready..


______________________________________________

now thats over with! its time to dry mount your block. If you have to, remove the Mosfet/vram heat spreader to verify you have a solid contact on the gpu. When you feel you got it all locked down and looking good. Apply your thermal paste (I still favor silver 5), and mount that sucker down!

Hook it up and let it crunch baby!!!

I will be happy to provide pics and such when I figure out how to add them to here. Always open to new ideas, improvements, etc...
 
Last edited:

Rob Williams

Editor-in-Chief
Staff member
Moderator
Thanks a ton for the post, Madmodman! Pics would be awesome if you could provide them.

Kind of bizarre how the LGA1366 block fits over the Tesla GPU...

I shudder thinking of all that has to be done to a $2,000+ GPU just to get it to work without overheating. I don't really see how NVIDIA could have overlooked this problem.
 
Top