Under the hood of a M2090

[atypicalguy]

No clue about that jack. Wild guessing, if Nvidia monitors temp like Intel, then the sensing is on chip but still requires a device nearby as the interface. I think that may be missing on this board. But, I have never plugged one of my M2090s in as I just have not had the time. I have had a i7-990X for several months to upgrade from a i7-920. I am hoping to be able to do the upgrades in 1 weekend.

I think the jack for the fan is in the opposite corner ... where there is no jack but a place for one.

If you are in CUDA mode for most of your analysis running 25% CPU load, why upgrade the processor? Just curious.

 

[Psi*]

Not all models fit in the 6 GB of the Tesla. Even then the software has substantial pre-/post-processing with some single threaded & some multi- and not CUDA. So a fast host CPU is needed.

 

[atypicalguy]

Long story short I went with the following:

koolance GPU-210

Swiftech H2O-220 Edge HD

The koolance block is smaller than the heat spreader on the GPU, but it seems to work anyway. Bolt on to existing heat sink mounts; I used the old screws. Runs about 45 deg on nbody simulation now fullscreen double precision 16,000 bodies or whatever. Be sure to order hose fittings with this block if you get one - I did not and had to drive a long way to get two extras. Kind of cheap of them to not include them I think.

I wasn't planning on cooling the cpu, but since the cpu block came with the radiator kit, and the block seemed to fit, I figured I would use it. The old CPU fan was interfering with the radiator installation also, so it is a win-win. I plumbed it in series, so it goes pump-> gpu-> cpu-> radiator-> pump. Should probably put the CPU first but oh well.

Apart from my motherboard sending me alarms on boot about the cpu fan not working, it is great. I will plug the pwm cable from the pump into that jack and see if I can get rid of the beep.

Sort of a pain to get the radiator to fit inside my case; Cooler Master Elite. You can do it, but I had to zip tie the radiator up there because the radiator mounting holes would not align with the mounting holes predrilled for the fans. The case was really not designed for an internal radiator. Next computer will definitely have a full tower case...it is tight on the memory cards but seems to be OK.

I plugged both radiator fans into the temp controller from CoolerGuys. It is binary off/on and it would certainly be nice to have that aquaero programmable one, but I kind of doubt I will get one as this one works well enough for now. I wish I knew how hot the cpu was getting so perhaps there is a role for a fancier unit after all. I can look in the motherboard setup screen and see a number, but that is obviously not under load. To be safe I just set the temp control to some low kick-in temp. The thermocouple is at the gpu heat spreader again.

Thanks again for your input everyone. I don't think I would have thought to go water except for this forum.

 

[Tharic-Nar]

Glad you got everything up and running. Let us know how things are holding out after a week or so. Guess you have to figure out what to do with all the air cooling kit now .

I wouldn't be too concerned with going from GPU > CPU in the chain. Water cooling is much more efficient and temperatures are unlikely to exceed what they were on air, even being pre-heated by the GPU. However, if you do become concerned, you could pre-cool by adding a small, single slot 80-120mm radiator after the GPU block, before feeding into the CPU. This could be back plate or side panel mounted, with the larger radiator in the top of the case. The pump should be able to handle the extra pressure (since it's rated for up to a triple rad). It all depends on space, location of the radiators and mounting slots available in the case.

 

[DarkStarr]

LOL, it doesn't matter where the parts are in the loop. The loop will equalize temp wise and the difference may be 1C. Even high end GPUs creating massive heat (overclocked and all) are typically only 1C apart if you have several and the water passes through in series.

Sounds like a win on the cooling however. What size rad are you running?

 

[Psi*]

haha ... so the "air guy" went water & the water guy (moi) is probably going to go air! I had a lot of work over the weekend & made no progress other than a dry fit.

I like the setup you describe. Nice job.

 

[atypicalguy]

haha ... so the "air guy" went water & the water guy (moi) is probably going to go air! I had a lot of work over the weekend & made no progress other than a dry fit.

I like the setup you describe. Nice job.

I should have put some valves in but whatever - it is all bolted together now.

The best part is being able to bolt the water block directly onto the GPU plate using the same four screws/springs/holes. The 3/4" hoses rub on the black heat plate a bit, and it is a bit tight so no compression fittings - I used hose clamps everywhere. One could use 45 degree fittings but it is nice having the lines come straight out parallel to the card - one out the top and another out the "back".

If one already had a water setup installed, this would be a no-brainer. It would also be extendable to multiple M2090s pretty easily. The CPU block even has multiple outlets so you could rig return lines to this radiator in parallel; main line ->CPU -> separate lines to two or three GPUs -> back separately to the radiator.

@Darkstar it is a two place radiator 120mm fans. It does not seem to have any problem with the E3-1280 or GPU going full tilt. It is hard to find an app that loads them both up at the same time; maybe BOINC or something along with a CFD run.

Where can I get a UV light for inside the case? Just want to be sure to see any leaks if they develop.

 

[Madmodman]

Hello again everyone,

I finally am home again and am ready to install my tesla card with the EK waterblock into the loop/system. However before I do, id like to toss the OE heatspreader back on the card and get a temp reading on it for comparison sake. Does anyone here know of a software application that supports temp monitoring of the Tesla m2090?

 

[Tharic-Nar]

This is part of the problem atypicalguy had. Although he was in Linux, he couldn't get any sensor information that correlated with the Tesla. This may be either a driver problem or it just doesn't have a sensor. Only thing that could be done was to attach a sensor to a fan controller or something, that's stuck to the underside of heatsink/block.

 

[DarkStarr]

I should have put some valves in but whatever - it is all bolted together now.

The best part is being able to bolt the water block directly onto the GPU plate using the same four screws/springs/holes. The 3/4" hoses rub on the black heat plate a bit, and it is a bit tight so no compression fittings - I used hose clamps everywhere. One could use 45 degree fittings but it is nice having the lines come straight out parallel to the card - one out the top and another out the "back".

If one already had a water setup installed, this would be a no-brainer. It would also be extendable to multiple M2090s pretty easily. The CPU block even has multiple outlets so you could rig return lines to this radiator in parallel; main line ->CPU -> separate lines to two or three GPUs -> back separately to the radiator.

@Darkstar it is a two place radiator 120mm fans. It does not seem to have any problem with the E3-1280 or GPU going full tilt. It is hard to find an app that loads them both up at the same time; maybe BOINC or something along with a CFD run.

Where can I get a UV light for inside the case? Just want to be sure to see any leaks if they develop.

Not bad, the 240 should be enough for the most part, just don't go adding things to the loop lol The rule is basically, 120mm of a good thick rad per component, none of that "slim rad" junk. After that its fit as much rad as possible lol That's why I have a 360 and a 280, both of which need new fans though.

 

[Madmodman]

Ive been getting the same results myself....

Ok, thank you- I will bring home the Infrared Thermometer today and see what I come up with. I will get back with everyone later on, hopefully tonight...

 

[atypicalguy]

Ya modguy read the thread it's all there. No temp sensor. Go fish.

Personally I think it doesn't matter though; once you put a waterblock on there all you need is a water temp sensor. I am going to either get one and screw it into a free port, then use that to power the fan, or use the PWM jack on the temp control and control both the fans and the pump speed from a single water temp thermocouple. I might also just tape the thermocouple to one of the block fittings and call it good; I didn't put any drain fittings into the system so I don't plan to take it apart any time soon!

I could use a fan controller to take the fans down a notch; so far the hottest I have seen on the gpu block is about 45 degrees in double precision fullscreen nbody simulation with the fans full tilt and the pump doing whatever the ASUS board tells it to from a PWM jack. Right now I have the CPU fully loaded on eight threads of BOINC and another BOINC thread running on the Tesla and I can't get it over 35C.

So I might just put a resistor in there and give the fans 7 volts instead of 12; the kit I bought has the resistors prewired for 7 and 5 volts on separate harnesses to plug and play. Pretty nice.

Thanks again everybody. Sorry I don't have a pic of the block on the card before it went in, but that part is really easy.

PS up is to the left on the photo

This is part of the problem atypicalguy had. Although he was in Linux, he couldn't get any sensor information that correlated with the Tesla. This may be either a driver problem or it just doesn't have a sensor. Only thing that could be done was to attach a sensor to a fan controller or something, that's stuck to the underside of heatsink/block.

 

[atypicalguy]

Not bad, the 240 should be enough for the most part, just don't go adding things to the loop lol The rule is basically, 120mm of a good thick rad per component, none of that "slim rad" junk. After that its fit as much rad as possible lol That's why I have a 360 and a 280, both of which need new fans though.

I just put "Core Temp" on my windows desktop. This morning with about 22deg ambient the highest core is at 77 deg C with all eight cores at 3.6Mhz 100% all night and the Tesla grinding away at 40 deg C on my poor man's temp sensor.

So there is a big difference btw what software reports from sensors embedded in the CPU chip and what a thermocouple on the GPU heat spreader just outside the waterblock report for component temps.

From what I understand the CPU chip is a lot more dense, so the heat is more concentrated and it responds better to increased flow rate, which is why you always want it in the main flow line. This is from Swiftech's site technical papers.

But I also suspect the actual GPU chip is pretty hot and I am just measuring the heat sink outside of the center, so lots of heat has been taken out by the water block before it ever gets to the thermocouple.

For now there is just no known way to get a good heat reading on the GPU chip itself. But if the cooling were inadequate the temp would equalize across the heat spreader and I would see higher readings. I think a thermocouple on the CPU spreader would probably show a lot less than 77 also.

If safety is the main concern, then the CPU temp would appear to dominate the system limits (I hope). I have intentionally loaded it all up and put a temp limit on the CPU monitor to shut the computer down if it hits 90 deg on the CPU core during the day today. There is just no good way to compare the temp values from the two components directly the way I am measuring them.

 

[Psi*]

Nice atyp. My experience with the C2070 is that it just locks up when it gets too warm. I managed to do that a few times when I was setting the OC settings.

I have never managed to control the fan speeds on the 2 systems I have with CPU watercooled. My systems are push/pull fans on the radiators but they are not noisy ... there is some noticeable fan noise but it is more of a whisper and everything is in dedicated office space. I had considered fan speed control just to reduce dirt accumulation in the radiators, but the radiators are much easier to blow out than the air cooled heatsinks for CPUs. And, I put mesh filters on the fans so dirt accumulation is pretty minimal anyway. This all caused me to lose interest in any sort of fan speed control.

I used a pyrometer all over my systems during setup. With high water flow the temp is quite uniform. Some argue that more heat is removed if the water flow is slow as it is moved thru the radiator ... good argument. But the water is also slow thru the water block & absorbs more heat.

I decided not to mess with any more gadgets & controllers & just let it pump. My CPU OCs are about as good as they get & they can run for days 100% maxxed. Running from 4.2 GHz to 4.5 GHz so another 100 to 200 MHz is not interesting.

And, the OC for the C2070 ... the GPU is set to the maximum setting of the software I am using. I think it is the memory clock that had to be backed down a bit for stability. I also do not think the issue is temperature related although if some sort of peltier device was used higher clocks might be had .... ehhh! I made a thread on that someplace in this forum.

 

[atypicalguy]

OK thanks for the reassurance. I am definitely on the "let it roll" plan today also.

Maybe I need some quieter fans; the ones with the kit looked pretty basic.

Just checking my home machine now- - so far the CPU is hovering around 80 fully cranked up; for some reason it posted an 84 max on core 3 earlier today but it is down now. I had it backing the hard drive up so maybe that was the issue.

I can probably do better with airflow. Fans are pulling down into the case to be sure the rad sees the coolest air possible, but inside the case is now warmer. I had to take off the rear case exhaust fan to fit the radiator in; maybe I will put it on the outside with a grate to keep little fingers out = Should help get the warm air out and potentially increase the flow across the radiator as well.

So DarkStarr was pretty spot-on: the two place 120mm radiator is just enough for this job. The card draws up to 250W per NVIDIA documentation and the CPU draws ~70 full tilt so that is a fair amount of energy to dissipate, assuming most of it goes into heat (? Surely some of it goes to increasing the order of the electrons in the CPU, but I have no idea how much. Hmmm).

I can see why people go with big external radiators and large fans. I just want it all inside the case to keep my kids away from it.

Nice atyp. My experience with the C2070 is that it just locks up when it gets too warm. I managed to do that a few times when I was setting the OC settings.

I have never managed to control the fan speeds on the 2 systems I have with CPU watercooled. My systems are push/pull fans on the radiators but they are not noisy ... there is some noticeable fan noise but it is more of a whisper and everything is in dedicated office space. I had considered fan speed control just to reduce dirt accumulation in the radiators, but the radiators are much easier to blow out than the air cooled heatsinks for CPUs. And, I put mesh filters on the fans so dirt accumulation is pretty minimal anyway. This all caused me to lose interest in any sort of fan speed control.

I used a pyrometer all over my systems during setup. With high water flow the temp is quite uniform. Some argue that more heat is removed if the water flow is slow as it is moved thru the radiator ... good argument. But the water is also slow thru the water block & absorbs more heat.

I decided not to mess with any more gadgets & controllers & just let it pump. My CPU OCs are about as good as they get & they can run for days 100% maxxed. Running from 4.2 GHz to 4.5 GHz so another 100 to 200 MHz is not interesting.

And, the OC for the C2070 ... the GPU is set to the maximum setting of the software I am using. I think it is the memory clock that had to be backed down a bit for stability. I also do not think the issue is temperature related although if some sort of peltier device was used higher clocks might be had .... ehhh! I made a thread on that someplace in this forum.

 

[Psi*]

Finding fans can be more more challenging than it should be. I check newegg (which has a great return policy) and have filtered all other WC sites down to mostly Performance PCs. There is always jab-tech & frozencpu which come up. Performance PCs has always shipped the same day.

Also, the Zalmon VF1000 I just received was missing 1 compression spring ... received 3 of 4. I contacted Performance, they contacted their supplier, who in turn emailed back copying me also & indicating that a spring was going out priority UPS. All of that transpired in the same day ... hard to ask for much more than that.

For radiators, thick fans are best. I have have the low speed version of these Fesers. According to Performance (I asked them) those low speed fans are not available any longer as "the company is difficult to work with". The hi speed fans in that link would be much too loud for any normal environment. 65 dBa!!!!! They also offer Pabst fans and are much quieter. I will admit that I have forgotten how loud might be too loud. Those Pabst at 39 dB still seem a little too loud tho and I do not believe that you need that kind of air flow ... I would stay below that noise level & is why I used push/pull.

Because I have the push/pull which is overkill I added these filters just to keep the dirt out.

 

[Tharic-Nar]

For fans, there are three choices that I'm aware of if you want to push air through the rad, rather than pull. These fans are focused with a high static pressure. Their CFM will appear lower, but are more effective at forcing air through tight spaces.

Corsair SP120 High preassure - (has a quiet version too)
Noctua NF-F12 PWM 120mm Focused Flow
Silverstone SST-AP121 Air Penetrator

It may just be a marketing vid, but Silverstone posted a very informative test on youtube that shows the difference quite clearly. http://youtu.be/8m8fC809TK0

If you are pulling air through, then high static pressure is less of an issue.

Oh the irony... looks like someone put an article together just for you. http://www.anandtech.com/show/6177/...ator-fan-testing-eight-fans-with-corsairs-h80

 

[DarkStarr]

I am getting Bitfenix Spectre Pros I need some 140s and 120s. If you want to go cheap then grab several CM R4 fans, they work great just don't expect them to last forever. I had 6 of them on my rad push/pull and got better temps with 2 deltas at 7v but the deltas were sooooo loud. That's why I added in the extra rad, so I can have slower fans and bring in less dust.

 

[Psi*]

The Zalman VF1000 keeps the M2090 ... uh ... working ok. As discussed there is no way to know the GPU temp like there is with the C2070.

I have compared the speed of the M2090 to the C2070 (overclocked I think) & the M2090 is faster. Almost twice as fast. But, the M2090 is in a faster system; the M2090 is in a i7-990X versus the C2070 in a i7-920. Both CPU are OC-ed to well over 4 GHz, but a utility I have that came with a commercial number-crunch software shows the PCIe bus in the i7-990X is almost 10X faster than the i7-920. Mother boards are identical.

When there is a lull in the current crunching I will upload a few screen shots. Still, the C2070 is substantially faster than the i7-990x crunching by itself which had been the comparison. And, I must def find the time to drop in the extra i7-990x.

 

[atypicalguy]

I ended up going push pull on the two fan radiator with a couple silent tornados (cant remember the name but the ones that get good reviews 1450rpm push and one of the swiftech fans in pull. The pull is thermostat switched along with a bunch of case exhaust fans with the temp sensor on the gpu. The push fans are plugged into the motherboard so they run continuously. Makes a pretty quiet machine unless it is crunching.

I would put the temp sensor on the cpu next time and put the cpu first in the loop. Maybe go with two pumps and a bigger case with triple radiator. But what I have will keep it running even on hot days, so it is just enough. Cpu temp is limiting factor and it it motherboard protected so it just powers down if too hot. Cpu runs 81-82 deg with everythink cranking in hot weather.