Truly amazing that you were able to get this working on a desktop PC build. Wish I had seen your video sooner, since I was trying the same thing. 3D printed a fan shroud to connect 2 Noctua A6x25 60mm fans, pulling air through the length of the card (video of it is on my channel). In hindsight, I should have spaced the two fans along the surface of the heat fins and blown air onto the card, as done by most actively cooled GPUs nowadays. I ended up giving the card to a colleague, since I wasn't able to get working in my system (Windows 10, X570 MOBO with resizable BAR and above 4G decoding support, 850W PSU, GTX 1660S for display). It was detected as a device under device manager, however I couldn't access the card in MATLAB or task manager. Probably an issue with GeForce/Tesla drivers not being able to exist simultaneously. Anyways, amazing setup! Although the video's over a year old, definitely has some utility nowadays for ML/image segmentation projects as most high end desktop GPUs are hard to come by. Out of curiosity, were you able to overclock the K80s and mitigate any thermal throttling that would ensue with your cooling setup? Wonder if changing the decade old thermal paste would also help.
Pretty sure you need to install quadro drivers and switch to graphics mode in order to find in task manager. I had to do that on mine. Also how was the cooling? Im thinking of adding a dual 60mm noctua fan system to blow air though the card, but I want to know if it works or not before I attempt.
@@TacticalPhoenixYT True, however I had a GTX series card at the time that had a lot of graphical issues when I installed the drivers needed for the K80. I ended up selling it and buying a RTX 3090 card instead. With regards to cooling, having two Noctua fans blow across the body of the card in a blower-style-GPU manner was a terrible design idea in retrospect, if you're referring to what I made on the video in my channel. Instead, I would have done what's being done in this video, but with two fans being controlled by a single PWM header (open air-style GPU).
@@pataporon but is there any real way to make a quiet exhaust fan for these cards? I'm using a workstation, so the airflow isn't great. I'm thinking of making a card that blows the air out.
@@TacticalPhoenixYT I'd recommend going with the same design shown in @Robin Grosset 's video, but instead of the delta fan, have two Noctua NF-A9 PWM fans side by side, each operating at 100% load (can be configured in your BIOS). Delta fans are extremely powerful, but extremely loud as well. The NF-A9 is the same size as the Delta fan, but operates much quieter (comparable to the two 60mm fans I had in my video operating at 100% load). The purpose of the Arduino controller was to give variable fan speeds proportional to the temperature of the card to help keep the card quiet when it's not under heavy load, but you can safely operate the two Noctua fans at 100% without it being any noisier than most CPU coolers. Also, I highly recommend changing the thermal paste on your m40 if you haven't already, since it likely hasn't been changed in the 10+ years the card was last manufactured. There's a video by @FUTC doing that on a tesla K80, but the procedure should be almost identical for your M40. Alternatively, you can also go about the watercooled approach. There are a few waterblocks available for the M40, but they cost around $100 USD which is comparable to the cost of the card itself.
I can only use this card on Linux... on Windows it appears configured with the drivers correctly, but no editing software recognizes it... it's not a defect, because on Linux it works
Any fan you can get to fit, will work. Passively cooled because it's designed to go into a server rack. Also explains why the power plug is in the back.
I come back to this video every couple of months. That delta fan is hilarious to me. I think about using my old delta fans from my old opteron servers and install them in my gpu workstation. The audio reminds me not to do that.
I'm looking to build a dektop specifically to use this GPU, and the question of how many PCIes it should have has arisen. I was thinking maybe buying 2-4 GPUs. Your DIY solution increases the width of the GPU, does that mean, regardless of the number of PCIe x16 ports, a desktop PC can only have 1 nvidia tesla k80 gpu? Unless the cooling is done on the side instead of on top, thereby not increasing the width, and allowing more space between GPUs. But these GPUs are already pretty thick, so even if I did this, do you think you could stack 2-4 of these GPUs side by side on a motherboard that has the PCIe ports? Thanks for sharing your solution.
First, bless you for making this video, it’s an inspiration to those of us who dream of having our own serves to render on at home. Could you have used a server style chassis like the rosewill 8gpu mining case for adequate cooling? There are fan mounts at the front and back of case.
Jackson Lukas as long as there is enough airflow it would work. Regular case fans don’t seem to push enough air but you could try. The cards will thermal throttle or shutdown I believe to avoid overheating.
hi Robin! Ok thanks for the reply. I’m doing a test with a server box and 6 5500 rpm 250 cfm fans to see if it moves enough air. It will be ready to test within the next 10 days, so I’ll tell you how it goes. If it doesn’t work I’m going to put one of the fans on the gpu itself like you’ve done here.
Hey Jackson , that airflow is more than enough I would say. The fan in the video is a Delta PFC0912DE with only 200 CFM and its good for cooling the card it never gets above 80 degrees with one of them.
Sir, could you please share all the names of the component that make up this dektop PC. The motherboard, the cpu, the case, etc? I have never built a PC, but am looking to build one for the Tesla K80 GPU, so this information would be very valuable for me, as you have had success with this setup.
Thanks for posting this video... I would like to run LLM (LLama 2 model) on my local machine and looking for a way where a cluster of GPU can be built... I understand that the model would require at least 32GB on a single card to load the entire model in the memory... Could you please confirm how did you create a Arduino circuit board or is it available in the Amazon marketplace and what's the name of it?
I have a question for you man :-) so I want to hook up a m40 Tesla to my HP computer but I want to set the Tesla on the top of the computer not mount it inside and use one of those brackets you have to take out of the back of the tower so you can insert the card or the Tesla and just have the wires from the power supply the motherboard going out of that hole in the back of the computer now up to the top of the tower where the Tesla GPU is sitting. Okay now on to the real point would it be possible to 3D print a new fan housing for the m40 and k80 etc where you take all of those things off of the casing and expose the chips and whatever you were talking about and reverse the fans so that they are sucking air out of the GPU and blowing it straight up so it would be kind of heating my bedroom not just blowing hot ass air around inside the tower.?? What do you think about this Am I crazy and dumb I'm new and kind of dumb and a little crazy but what do you think about this idea
hello nice video! this card is compatible with intel core i5 10400 2.90ghz processor (integrated intel UHD graphics 630 card)? you recommend it for the gaming?
I use a multi-gpu monitor in Zabbix on my headless linux machine. It uses data from the nvidia-smi command, which reports temperature, memory usage, and other data. Also there are alerts for high temperatures. I don’t regard the device he has created in this video as a monitoring tool beyond regulating fan speed based on relatively general temperature levels.
I am really interested on developing ML models for such small games as you have shown in the video. Could you please share more details on that ? I mean how to interface games with Tensorflow etc.?
Thank you for this Robin. I am trying to recreate the setup you demo here, but have no experience with the Arduino hardware. Can you also provide a link to the breakout board you used like you have for the fan and temperature sensors on the GitHub page? I appreciate your help.
I used a prototype board. I have lots of different proto board lying around. It comes in a lot of sizes and is pretty cheap. Here is a link which has a bunch of sizes for $17. www.amazon.ca/dp/B072Z7Y19F/ref=cm_sw_em_r_mt_dp_hxVwFbPM66JBP
The Arduino board is something called a Arduino Pro Mini 5v version, you can get them on eBay or here is a link to Adafruit www.adafruit.com/product/2378?gclid=EAIaIQobChMIvbLqyrjo6wIVQAiICR2bogN6EAAYAiAAEgK1WfD_BwE With these you need a separate programmer like an ‘FTDI friend’ . The ones with built in programmers would also work fine like Arduino Nano. The link in the description will take you to the schematic and wiring diagram.
@@robingrosset6941 thanks! I already bought the set you use there in the video and built the whole (legacy) setup based on the arduino. works nicely so far!
Hi Robin, I am planning to use this for training Deepfakes. Is the K80 recognized in Linux/Windows as either: ONE GPU with 24GB of VRAM total, or two GPUs with 12GB of VRAM each? The program I use would benefit immensely if it were the first option. If not perhaps I should stick with my Titan X because it doesn't scale well with more than one GPU, and I already have 12GB of VRAM.
Tesla K20's are really cheap now, is it possible to to use them for gaming like how laptops switch between integrated graphics and discrete graphics. In windows 10 you can go to graphics settings and select a certain gpu to be used in certain programs. This would work for tesla cards?
The K80 by itself acts as 2 GPUs for the purpose accelerating certain workloads, not necessarily graphics related. It is not designed for gaming for example. This card can help with video rendering, 3D rendering, machine learning and scientific computations. Its a hardware accelerator called a GPU but it does not output graphics directly to a screen like a RTX 2070 Super. They are hard to setup as they are designed to be used in servers in data centres and not for consumer desktop use.
@Cedric Moore exactly you got it. It does not need the graphics card at all. In a server it returns this to the cpu or whatever process needed it. It uses a technology called CUDA from NVIDIA which can be used for all kinds of computing needs. A K80 is more like a CPU with 1000s or little cores. These cards can work 10 to 50 times faster than a CPU on the same computational problem. They are popular with high performance computing people and crypto miners for this reason.
Hello I'm not much of a linux person I'm using ubuntu the latest one. But i was wondeing Sense i have a server with two of the K10 cards on there i would like to share one card with my brother so he can use it. Problem is I'm not much of a linux person and i have no clue how to set these cards up on linux. What would be a good way to start and get the ball rolling as they say?
I noticed on a machine with intel integrated graphics that once you install one of these it will use one of the GPUs in the K80 a little for visual desktop display things. To it seems to use a GPU for video related stuff despite it having no direct output. I may try to capture this to show it. So I think the answer is maybe !
I had this problem too, I found ASUS ‘WS’ workstation class motherboards worked well. Like this one rover.ebay.com/rover/0/0/0?mpre=https%3A%2F%2Fwww.ebay.ca%2Fulk%2Fitm%2F303128322721
@@robingrosset6941 Thanks. Nice motherboard. I looked into it, and it will require I get a new CPU and RAM as well which is going to be too expensive at the moment, so I decided to swap the Tesla for a GeForce RTX 2070 instead (recommended here: timdettmers.com/2019/04/03/which-gpu-for-deep-learning/). I would have loved to get the Tesla up and running though. They are coming down in price fast. I got one for $450 at Amazon. It'll have to be next time. Do you know if your motherboard can support more than one Tesla k80?
so. We know how to cool it by taking it apart, but is there a desktop way of cooling it the inteneded way? (I know this does that but I don't fancy putting metal tape near my eletronics.)
For sure there is a way to attach a 3d printed fan shroud to the machine. You also need a specific kind of fan. Here is a link www.ebay.com/itm/NVIDIA-Tesla-K80-P100-V100-FHHL-M40-cooling-fan-shroud-duct-/153175788721 Its $22 USD for the shroud and about $30 for the fan as I recall. The problem you hit doing this is how long the card becomes with the shroud and fan. You need a BIG case to fit this. The other option is simply to buy a rack server chassis like this one you might find a used one cheaply ebay.us/m7ab5z
Nuclear Armament if you don’t put any load on it it would be okay for a few minutes. Under load it will heat up very quickly. If you watch other videos of people doing this they say without cooling they thermal throttle and shut down. I’ve never not cooled one deliberately.
A question that came to my mind: Is there thermal paste inside the GPU? Because, since those GPUs are quite old, we should change the paste, if there is any. Does anyone know?
I can take a look at this and see if it needs replacing. So far these cards have been pretty well behaved thermally and I actually reduced fan speed as it was not needed to be so high. Will keep in mind and see if replacing thermal paste at some point improves temps
it might be worth it. The K10 has pretty good/similar performance relative to a 1050ti based on the specs. If you are doing Machine Learning you may be able to do more with 2 cards. The K10 uses a whole lot of more power than the 1050ti however. I don't know if your motherboard will boot with the K10 in it but it should be easy to figure out. In Asus BIOSes you need to make sure Above 4GB option is enabled in the Boot menu. If it boots and gets to the OS and can see the card you are good. The K10 won't overheat during a normal boot process.
@@robingrosset6941 Ok, thanks... I was thinking about pairing it with my 1050 ti, but I don't think I could do that on this motherboard. Just wanted to know if it was safe and I guess I would need some extra fans to keep it cool, but thanks again... Much appreciated. 👍
So far I have been able to get this to work with an ASUS motherboard. You need to set the Above 4GB Decoding option in the boot menu. Not setting this means your motherboard won’t POST. I also understand Intel c612 chipset boards should also work but I don’t have one to try.
@@robingrosset6941 i have asus B85M-G R2.0 but i cant install the drivers (265.718126] nvidia-nvlink: Nvlink Core is being initialized, major device number 237 [ 265.718469] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid: NVRM: BAR0 is 0M @ 0x0 (PCI:0000:03:00.0) [ 265.718470] NVRM: The system BIOS may have misconfigured your GPU. [ 265.718475] nvidia: probe of 0000:03:00.0 failed with error -1 [ 265.718535] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid: NVRM: BAR0 is 0M @ 0x0 (PCI:0000:04:00.0) [ 265.718535] NVRM: The system BIOS may have misconfigured your GPU. [ 265.718538] nvidia: probe of 0000:04:00.0 failed with error -1 [ 265.718577] NVRM: The NVIDIA probe routine failed for 2 device(s). [ 265.718577] NVRM: None of the NVIDIA devices were initialized.) SORRY MATE
@@Dannyramos73 your board POSTs which is good, make sure you have the BIOS setting on your motherboard called 'Memory Remap' and set to [Enabled] this is the above 4GB setting I think. From what I have read there is something call PCIe BAR support and this memory mapping is set during the POST sequence. Some motherboards, like my ASRock X99 Taichi, seem to choke on the memory remapping required for the 24GB RAM in the GPU. There was nothing I could do to get it to work. So if the Memory Remap setting does not work, you could try updating your BIOS the latest for that board is from 2018 so it might work. Here is the link www.asus.com/ca-en/Motherboards/B85MG_R20/HelpDesk_BIOS/ . let me know if you get it running.
nice video, was hoping my HP DL380 server could handle this card not arrived yet, but has only 2 PCI x 16 gen 2 slots. it has great 128GB ram and 2 Xeon CPU. looks like I am going to external with an GPU caddy that uses USB-C thunderbolt cable to my laptop.
@@robingrosset6941 nope managed to get it going on a bf-450f rog board bio flashed to 2021 to enable above 4g. I've got 2 fans on it but gets really hot quickly.
@Scott Patnode I am currently using a BF450F board with latest flashed Bio to unlock the above 4g option, using windows 10 64bit, I am having trouble using the K80 for bitcoin gold mining, but works good for my rainbow tables, but didn't have experience in particle simulations etc even broken 1080ti etc all selling for £200+ and working ones £400 etc the prices are so high because of people using them for mining Ethereum etc at the moment. from reports k80 isn't a great mining card compared to 1050ti, 1080ti etc is makes it cheaper to buy compare to the others. another reason why 3080 etc are hard to get or prices are inflated to sky the limit as people buy them up to thrash them for mining.
If you can map the GPU into the virtual machine and use it like a GPU to run machine learning stuff then for sure. I know VMWare ESXi can use GPUs this way but I have not tried with freeNAS. Linus from LinusTechTips has some videos where he virtualizes GPUs like this so it might work and would be beneficial because machine learning jobs can take hours or days so running on a server makes complete sense.
Might be possible but you would still have to control the fan speed somehow and link the sensor to that. Discovered a little board on Amazon which has temp sensor and fan control in one DC 12V PWM PC CPU Fan Temperature... www.amazon.ca/dp/B019P0FLHW?ref=ppx_pop_mob_ap_share
@@robingrosset6941There is a windows app called speedfan where you can link a fan to a temp signal, that would work for sure. Maybe you can also just do it in the BIOS. I will try to get a card and test it. I have seen working cards for 190€ on eBay...
Thank you for the great tutorial, Robin! I also have a Tesla K80 from eBay, and I would like to try your mod (great application for an Arduino). However, recently my K80 has some problem with a corrupt infoRom. This is why I am reaching to you, with a small request: if your card is working, can you provide some dumps of the two infoROM sfor the two GPUs? I know this is asking a lot (especially considering that I am a complete stranger), but I would be eternally grateful! It is embarrassing to ask such a thing, and probably you don't have time for such endeavors :(.
Hello Dragos, I think the easiest way to fix this is just to use this utility which includes the bios. support.hpe.com/hpsc/swd/public/detail?swItemId=MTX_f7bea664267845a683084432dd . Best
Thank you for the idea, @@robingrosset6941! I just tried this, and it did not help. It seems that the two infoROM are independent of the vbios chips. And (for some reason) the infoROMs on my card got corrupted. So, here I am, trying to convince people to run nvflash on their working GPUs, to get these 2x 32kb dump-files. It is crazy that such a small amount of data can make or break a chip, right? Anyway...I understand that my chances are slim (I am not sure that I would touch nvflash because someone else asked me to; but it was worth a try). Thank you again for your time!
Thank you very much, @@robingrosset6941 ... I would really appreciate that! To minimize the time you lose with this, below are the steps. Of course, this process should not affect your card in any way. - unload nvidia kernel-modules: sudo rmmod nvidia_drm nvidia_modeset nvidia # (CentOS/RHEL 7 ... might be different on Ubuntu) - get the nvflash_linux utility from here: www.techpowerup.com/download/nvidia-nvflash/ - dump the infoROM from the card, into a file (ifr file-extension is important): + sudo ./nvflash_linux --save inforom_dump_card0.ifr # (and select the index of the 1st K80 GPU) + sudo ./nvflash_linux --save inforom_dump_card1.ifr # (and select the index of the 2nd K80 GPU) - re-load nvidia module (or restart the machine), to continue with your work: sudo modprobe nvidia If this works, can I at least buy you an Arduino Mega to play with? It's a symbolic gesture, but I want to thank you somehow ;).
@@robingrosset6941 I can imagine that you are quite busy with other things, so I apologize if I am annoying. Do you think you will have time (at some point...no pressure) to save those infoROM files? I guess I am a little desperate, so I am "rising the stakes" :)). I would be more than happy to get you one of the newly-announced Raspberry Pi 4's (which are apparently good-enough for ML inference): arstechnica.com/gadgets/2019/06/faster-raspberry-pi-4-promises-desktop-class-performance/
maintherboard ASUS Z10PA-D8C For Dual Xeon E5-2600v3/v4 + CPU Intel Xeon E5-2678v3 12C/24T 2.5GHz Upto 3.3GHz 30M Cache. Do you think this set up is good for k80 Tesla ?
Yes that Asus board is Workstation or Server class so it should work. If there is an ‘Above 4GB’ setting in the bios be sure to enable it before installing.
Jay Thai the ‘Above 4GB Decoding’ setting is on on page 78 of the manual in bIos section under PCI Subsystem Settings. It *should* work but as nobody publishes compatibility with K80s it’s hard to be 100% but I am pretty sure it will work.
In the video description there is a link to a schematic and code for the Arduino. I used a Pro Mini just to keep it small but you can use any Arduino. Here is the link github.com/rgrosset/CoolingTeslaK80
The card uses standard CUDA drivers form NVIDIA so I believe this would just work with DaVinci Resolve. I don’t use DaVinci Resolve myself but from what I have read I think it will work.
@@robingrosset6941 I bought a Tesla k80, to try it out in mining but i get an error message saying "This device cannot find enough free resources that it can use" It is in my understanding that this GPU already comes in compute mode. So i don't understand what could be going on. As you might be able to tell, I am not an expert.
It can still do machine learning and run big models bigger than most GPUs on sale today. Its performance is not as good as most recent cards. For $200-$300 it likely can't be beat particularly on model size.. But if you have $600 and don't need the RAM size then go for an NVIDIA RTX card .. it will be faster and more energy efficient. Today I would go for a 10 series or RTX series card as they are so much faster.
This video has nothing to do with Tesla the car company. These are high performance computing cards from NVidia they just happen to have a brand name Tesla too. There is a connection in that cards like these are used to build artificial intelligence systems to make self driving cars. blogs.nvidia.com/blog/2019/04/23/tesla-self-driving/
Thanks for the comment. In a server, which is where this card normally belongs the card has no fans its cooled by airflow through the server chassis. The main thing is to ensure the card and each GPU is cooled, with this delta fan it is really overkill, these fans at full speed will lift off a desk by themselves. Its a bit too much cooling. I have tried 2 fans together and the issue comes back to where will the hot air exit the card. By design of the cooling fins the hot air flows over one GPU and then over the other there is usually a temperature increase over the 2nd GPU. You see in the video one GPU is 48 degrees C the other 54 degrees C. This is normal for these cards in server environments. There are some examples of overheating cards and normal cards at the NVIDIA forum link google "tesla-k80-overheating 38690" The temperatures in this video are well below the thermal throttling levels. If you look at NVIDIA SMI screen shot in the video you'll see the GPUs are loaded up 50% and 30% But notice GPU RAM usage is 100% on both GPUs.. so they are working hard and at the RAM limit. RAM limits are often the big factor in why folks use these GPUs 10:52 nvidia-smi 100% RAM used. Its normal for CPUs to work hard pushing data to the GPUs, but it does depend on the workload. This test is CPU and GPU intensive because I am running 8 parallel machine learning workloads and an element of this workload is CPU bound because its displaying the OpenAI Gym Space Invaders user interface at high speed. In terms of longevity for this setup I just took a K80 out of a machine after 3 years and its still running fine.
@@robingrosset6941 got both , M40 single eCPU and k80 dual , full load on image generate they stay under 70 temp with 2 fans like you did on this video
Can you sell me this as a prebuild ... Since they are $175 on amazon... How much for this? I really would love to pair this with a p2200 in my threadripper desktop for video editing.
Sorry, I don't make these to sell. They are for my own use. The first one I made is certainly overkill you don't need all that air moving over the card. I found a better combination which is to build some prebuilt parts still a little overkill but cheaper to build and from pre-exisitng parts on amazon. Here is what I would recommend. Fan controller if you want temperature control. DC 12V PWM PC CPU Fan Temperature Control Speed Controller CPU Alarm Module www.amazon.ca/dp/B019P0FLHW/ref=cm_sw_em_r_mt_dp_TAE6FbSR9HZ7F?_encoding=UTF8&psc=1 And then you need a 80mm PWM fan 4pin version of this, I used 2 of these. You can also plug these fans into any motherboard header and control them from your motherboards fan controller. GDSTIME 8025 80mm x 80mm x 25mm 12v DC Brushless Cooling Fan 4PIN www.amazon.ca/dp/B086HCBH9S/ref=cm_sw_em_r_mt_dp_pPE6FbV3GJDCZ
@@robingrosset6941 do you think the k80 amd a quadro combo is good for extra compute power for video editing or would i be wasting a graphics card slot
These K80 cards are great. The real advantage is the massive amount of memory over others however they are a few generations back so depending on what you are doing a RTX 20xx might be better. An RTX 2080ti for context is about 4 times faster than a single GPU in a K80 but if you can use both GPUs and the memory the K80 is still a contender. To get more memory in a GPU still today you are taking $ thousands.
Your motherboard was released in 2012, 2 years before K80 was launched. It also does not have the 'Above 4GB Decoding' option in the BIOS which in my experience is required to get the system to power on with a K80 installed. Sorry to say this means that it is unlikely to work. I could be wrong, this is just my opinion based on what I have tried.
@@purewbattogtokh6557 A K80 will cost around $400 to $500, you could simply buy a high end graphics card from NVIDIA. It depending on the application. Based on this web site below it looks like 3dStudio works just as well with older NVIDIA GTX cards like GTX 980 or GTX 1080. These would be a cheaper route and you could install more than one. www.pugetsystems.com/labs/articles/AutoDesk-3ds-Max-2017-GeForce-GPU-Performance-816/
hi is me again, i am thinking to buy the PNY NVIDIA Quadro P1000 Scheda Video 4 GB GDDR5 PCIe for the video and the Tesla K80 do you thing that will do it fine ?
Hi, i am going to buy k80 for my workstation with 2x xeon 2683v4 in huananzhi x99-f8d, but xeon dont have integrated graphics, what should i do in this case? I think about buy cheap and old another graphic card. Sry for my bad english , hope that u understand me
Truly amazing that you were able to get this working on a desktop PC build. Wish I had seen your video sooner, since I was trying the same thing. 3D printed a fan shroud to connect 2 Noctua A6x25 60mm fans, pulling air through the length of the card (video of it is on my channel). In hindsight, I should have spaced the two fans along the surface of the heat fins and blown air onto the card, as done by most actively cooled GPUs nowadays.
I ended up giving the card to a colleague, since I wasn't able to get working in my system (Windows 10, X570 MOBO with resizable BAR and above 4G decoding support, 850W PSU, GTX 1660S for display). It was detected as a device under device manager, however I couldn't access the card in MATLAB or task manager. Probably an issue with GeForce/Tesla drivers not being able to exist simultaneously.
Anyways, amazing setup! Although the video's over a year old, definitely has some utility nowadays for ML/image segmentation projects as most high end desktop GPUs are hard to come by.
Out of curiosity, were you able to overclock the K80s and mitigate any thermal throttling that would ensue with your cooling setup? Wonder if changing the decade old thermal paste would also help.
Pretty sure you need to install quadro drivers and switch to graphics mode in order to find in task manager. I had to do that on mine. Also how was the cooling? Im thinking of adding a dual 60mm noctua fan system to blow air though the card, but I want to know if it works or not before I attempt.
@@TacticalPhoenixYT True, however I had a GTX series card at the time that had a lot of graphical issues when I installed the drivers needed for the K80. I ended up selling it and buying a RTX 3090 card instead.
With regards to cooling, having two Noctua fans blow across the body of the card in a blower-style-GPU manner was a terrible design idea in retrospect, if you're referring to what I made on the video in my channel. Instead, I would have done what's being done in this video, but with two fans being controlled by a single PWM header (open air-style GPU).
@@pataporon but is there any real way to make a quiet exhaust fan for these cards? I'm using a workstation, so the airflow isn't great. I'm thinking of making a card that blows the air out.
@@TacticalPhoenixYT I'd recommend going with the same design shown in @Robin Grosset 's video, but instead of the delta fan, have two Noctua NF-A9 PWM fans side by side, each operating at 100% load (can be configured in your BIOS). Delta fans are extremely powerful, but extremely loud as well. The NF-A9 is the same size as the Delta fan, but operates much quieter (comparable to the two 60mm fans I had in my video operating at 100% load).
The purpose of the Arduino controller was to give variable fan speeds proportional to the temperature of the card to help keep the card quiet when it's not under heavy load, but you can safely operate the two Noctua fans at 100% without it being any noisier than most CPU coolers.
Also, I highly recommend changing the thermal paste on your m40 if you haven't already, since it likely hasn't been changed in the 10+ years the card was last manufactured. There's a video by @FUTC doing that on a tesla K80, but the procedure should be almost identical for your M40.
Alternatively, you can also go about the watercooled approach. There are a few waterblocks available for the M40, but they cost around $100 USD which is comparable to the cost of the card itself.
@@pataporon thanks!
I can only use this card on Linux... on Windows it appears configured with the drivers correctly, but no editing software recognizes it... it's not a defect, because on Linux it works
Amazing video, you just killed any second thoughts I had on doing the same on a remote desktop for myself
Any fan you can get to fit, will work. Passively cooled because it's designed to go into a server rack. Also explains why the power plug is in the back.
i like how you make the card run without inserting it into PCIE slot
Its the fan running (testing) not the card. 🙂
I come back to this video every couple of months. That delta fan is hilarious to me. I think about using my old delta fans from my old opteron servers and install them in my gpu workstation. The audio reminds me not to do that.
If PFC0912DE has max rpm 9000, then 40% is around 3600 rpm, which seems enough. Good project !
One of these with this mod or Two K40c (the c denotes active cooling) for gpu acceleration? What's your take on this?
I'm looking to build a dektop specifically to use this GPU, and the question of how many PCIes it should have has arisen. I was thinking maybe buying 2-4 GPUs. Your DIY solution increases the width of the GPU, does that mean, regardless of the number of PCIe x16 ports, a desktop PC can only have 1 nvidia tesla k80 gpu?
Unless the cooling is done on the side instead of on top, thereby not increasing the width, and allowing more space between GPUs. But these GPUs are already pretty thick, so even if I did this, do you think you could stack 2-4 of these GPUs side by side on a motherboard that has the PCIe ports?
Thanks for sharing your solution.
First, bless you for making this video, it’s an inspiration to those of us who dream of having our own serves to render on at home. Could you have used a server style chassis like the rosewill 8gpu mining case for adequate cooling? There are fan mounts at the front and back of case.
Jackson Lukas as long as there is enough airflow it would work. Regular case fans don’t seem to push enough air but you could try. The cards will thermal throttle or shutdown I believe to avoid overheating.
hi Robin! Ok thanks for the reply. I’m doing a test with a server box and 6 5500 rpm 250 cfm fans to see if it moves enough air. It will be ready to test within the next 10 days, so I’ll tell you how it goes. If it doesn’t work I’m going to put one of the fans on the gpu itself like you’ve done here.
Hey Jackson , that airflow is more than enough I would say. The fan in the video is a Delta PFC0912DE with only 200 CFM and its good for cooling the card it never gets above 80 degrees with one of them.
Sir, could you please share all the names of the component that make up this dektop PC. The motherboard, the cpu, the case, etc? I have never built a PC, but am looking to build one for the Tesla K80 GPU, so this information would be very valuable for me, as you have had success with this setup.
Thanks for posting this video... I would like to run LLM (LLama 2 model) on my local machine and looking for a way where a cluster of GPU can be built... I understand that the model would require at least 32GB on a single card to load the entire model in the memory... Could you please confirm how did you create a Arduino circuit board or is it available in the Amazon marketplace and what's the name of it?
I have a question for you man :-) so I want to hook up a m40 Tesla to my HP computer but I want to set the Tesla on the top of the computer not mount it inside and use one of those brackets you have to take out of the back of the tower so you can insert the card or the Tesla and just have the wires from the power supply the motherboard going out of that hole in the back of the computer now up to the top of the tower where the Tesla GPU is sitting. Okay now on to the real point would it be possible to 3D print a new fan housing for the m40 and k80 etc where you take all of those things off of the casing and expose the chips and whatever you were talking about and reverse the fans so that they are sucking air out of the GPU and blowing it straight up so it would be kind of heating my bedroom not just blowing hot ass air around inside the tower.?? What do you think about this Am I crazy and dumb I'm new and kind of dumb and a little crazy but what do you think about this idea
Grate Job! Waiting for my P100 delivery , will do same trick except temp sensor
hello nice video! this card is compatible with intel core i5 10400 2.90ghz processor (integrated intel UHD graphics 630 card)? you recommend it for the gaming?
I would have placed the temp sensor in direct card output airflow, so it gives you a more reliable reading. great work
Precision of the temperature reading would not be worth any obstruction of airflow.
@@davidmchugh6908 but in that way you can also see if the first gpu is running while the last is shut off. also you can use a "smaller" probe
I use a multi-gpu monitor in Zabbix on my headless linux machine. It uses data from the nvidia-smi command, which reports temperature, memory usage, and other data. Also there are alerts for high temperatures.
I don’t regard the device he has created in this video as a monitoring tool beyond regulating fan speed based on relatively general temperature levels.
Hello i got a tesla m40 when i remove the shroud the fins of the cooler are completely closed unlike yours. Is there any way to open them?
I am really interested on developing ML models for such small games as you have shown in the video. Could you please share more details on that ? I mean how to interface games with Tensorflow etc.?
Thank you for this Robin. I am trying to recreate the setup you demo here, but have no experience with the Arduino hardware. Can you also provide a link to the breakout board you used like you have for the fan and temperature sensors on the GitHub page? I appreciate your help.
I used a prototype board. I have lots of different proto board lying around. It comes in a lot of sizes and is pretty cheap. Here is a link which has a bunch of sizes for $17.
www.amazon.ca/dp/B072Z7Y19F/ref=cm_sw_em_r_mt_dp_hxVwFbPM66JBP
The Arduino board is something called a Arduino Pro Mini 5v version, you can get them on eBay or here is a link to Adafruit
www.adafruit.com/product/2378?gclid=EAIaIQobChMIvbLqyrjo6wIVQAiICR2bogN6EAAYAiAAEgK1WfD_BwE
With these you need a separate programmer like an ‘FTDI friend’ . The ones with built in programmers would also work fine like Arduino Nano.
The link in the description will take you to the schematic and wiring diagram.
Thank you, i’ll see if i can work it out For now i’ve just got a fan running at max velocity.
Hi Robin,
thanks for the nice tutorial. Which screwdriver is needed to open the card?
I figured this out its a Torx, I cover this in this video ua-cam.com/video/nLnICvg8ibo/v-deo.html
@@robingrosset6941 thanks! I already bought the set you use there in the video and built the whole (legacy) setup based on the arduino. works nicely so far!
amazing demo
Nice cooling! What motherboard do you use for this setup?
I am using an Asus Z170 WS. Seems Asus works but I found an ASRock board would not.
@@robingrosset6941 why did the asrock board not work?
Hi Robin, I am planning to use this for training Deepfakes.
Is the K80 recognized in Linux/Windows as either:
ONE GPU with 24GB of VRAM total, or
two GPUs with 12GB of VRAM each?
The program I use would benefit immensely if it were the first option. If not perhaps I should stick with my Titan X because it doesn't scale well with more than one GPU, and I already have 12GB of VRAM.
In either operating system it’s seen as two GPUs with 12GB each.
what type of tape did you use? also would this setup work with two of those fans?
It’s aluminum foil duct tape. You can get it in hardware stores. It should work work with 2 fans.
Tesla K20's are really cheap now, is it possible to to use them for gaming like how laptops switch between integrated graphics and discrete graphics. In windows 10 you can go to graphics settings and select a certain gpu to be used in certain programs. This would work for tesla cards?
Yep, but it's tricky. Loads of videos on UA-cam showing you how, and which cards it works on.
nice! right on dude.
Thanks!!
im so confused by this hardware, if you had this and a 2070 super in your pc what would it do? how does this work
The K80 by itself acts as 2 GPUs for the purpose accelerating certain workloads, not necessarily graphics related. It is not designed for gaming for example. This card can help with video rendering, 3D rendering, machine learning and scientific computations. Its a hardware accelerator called a GPU but it does not output graphics directly to a screen like a RTX 2070 Super. They are hard to setup as they are designed to be used in servers in data centres and not for consumer desktop use.
@@robingrosset6941 so basically hit handles things like physics mechanics and algorithms than sends that mathematical data to graphics cards.
@Cedric Moore exactly you got it. It does not need the graphics card at all. In a server it returns this to the cpu or whatever process needed it. It uses a technology called CUDA from NVIDIA which can be used for all kinds of computing needs. A K80 is more like a CPU with 1000s or little cores. These cards can work 10 to 50 times faster than a CPU on the same computational problem. They are popular with high performance computing people and crypto miners for this reason.
Hello I'm not much of a linux person I'm using ubuntu the latest one. But i was wondeing Sense i have a server with two of the K10 cards on there i would like to share one card with my brother so he can use it. Problem is I'm not much of a linux person and i have no clue how to set these cards up on linux. What would be a good way to start and get the ball rolling as they say?
What fan for cooling do you added? 80 cm or above? Thanks.
Lazarescu Catalin it’s 92 mm but 80mm also work. Go for a high air flow fan 33 CFM worked well for me.
could you use this in a desktop as an assistance for a cheaper gpu that has an output?
I noticed on a machine with intel integrated graphics that once you install one of these it will use one of the GPUs in the K80 a little for visual desktop display things. To it seems to use a GPU for video related stuff despite it having no direct output. I may try to capture this to show it. So I think the answer is maybe !
Great stuff, especially what you are using it for!! Just got one myself, but i realize my mobo don't have the above 4 GB decoding option. Bummer.
I had this problem too, I found ASUS ‘WS’ workstation class motherboards worked well. Like this one rover.ebay.com/rover/0/0/0?mpre=https%3A%2F%2Fwww.ebay.ca%2Fulk%2Fitm%2F303128322721
@@robingrosset6941 Thanks. Nice motherboard. I looked into it, and it will require I get a new CPU and RAM as well which is going to be too expensive at the moment, so I decided to swap the Tesla for a GeForce RTX 2070 instead (recommended here: timdettmers.com/2019/04/03/which-gpu-for-deep-learning/). I would have loved to get the Tesla up and running though. They are coming down in price fast. I got one for $450 at Amazon. It'll have to be next time. Do you know if your motherboard can support more than one Tesla k80?
What would happen if my mobo didn’t have that option?
Will this work fir m 40
so. We know how to cool it by taking it apart, but is there a desktop way of cooling it the inteneded way? (I know this does that but I don't fancy putting metal tape near my eletronics.)
For sure there is a way to attach a 3d printed fan shroud to the machine. You also need a specific kind of fan. Here is a link www.ebay.com/itm/NVIDIA-Tesla-K80-P100-V100-FHHL-M40-cooling-fan-shroud-duct-/153175788721
Its $22 USD for the shroud and about $30 for the fan as I recall.
The problem you hit doing this is how long the card becomes with the shroud and fan. You need a BIG case to fit this.
The other option is simply to buy a rack server chassis like this one you might find a used one cheaply ebay.us/m7ab5z
Can I just plug this card into my HP Z820 without a fan? Would it overheat or would it be able to work fine even under full load without it?
Nuclear Armament if you don’t put any load on it it would be okay for a few minutes. Under load it will heat up very quickly. If you watch other videos of people doing this they say without cooling they thermal throttle and shut down. I’ve never not cooled one deliberately.
A question that came to my mind: Is there thermal paste inside the GPU? Because, since those GPUs are quite old, we should change the paste, if there is any. Does anyone know?
I can take a look at this and see if it needs replacing. So far these cards have been pretty well behaved thermally and I actually reduced fan speed as it was not needed to be so high. Will keep in mind and see if replacing thermal paste at some point improves temps
I have an old K10. Is it worth pairing it with a 1050 Ti? Also which motherboard would I need for it because I have the Asus Prime B350 M-A.
it might be worth it. The K10 has pretty good/similar performance relative to a 1050ti based on the specs. If you are doing Machine Learning you may be able to do more with 2 cards. The K10 uses a whole lot of more power than the 1050ti however. I don't know if your motherboard will boot with the K10 in it but it should be easy to figure out. In Asus BIOSes you need to make sure Above 4GB option is enabled in the Boot menu. If it boots and gets to the OS and can see the card you are good. The K10 won't overheat during a normal boot process.
@@robingrosset6941 Ok, thanks... I was thinking about pairing it with my 1050 ti, but I don't think I could do that on this motherboard. Just wanted to know if it was safe and I guess I would need some extra fans to keep it cool, but thanks again... Much appreciated. 👍
excellent !! one question do you recommended any other mother board for tesla k 80? thanks
So far I have been able to get this to work with an ASUS motherboard. You need to set the Above 4GB Decoding option in the boot menu. Not setting this means your motherboard won’t POST. I also understand Intel c612 chipset boards should also work but I don’t have one to try.
@@robingrosset6941 i have asus B85M-G R2.0 but i cant install the drivers
(265.718126] nvidia-nvlink: Nvlink Core is being initialized, major device number 237
[ 265.718469] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid:
NVRM: BAR0 is 0M @ 0x0 (PCI:0000:03:00.0)
[ 265.718470] NVRM: The system BIOS may have misconfigured your GPU.
[ 265.718475] nvidia: probe of 0000:03:00.0 failed with error -1
[ 265.718535] NVRM: This PCI I/O region assigned to your NVIDIA device is invalid:
NVRM: BAR0 is 0M @ 0x0 (PCI:0000:04:00.0)
[ 265.718535] NVRM: The system BIOS may have misconfigured your GPU.
[ 265.718538] nvidia: probe of 0000:04:00.0 failed with error -1
[ 265.718577] NVRM: The NVIDIA probe routine failed for 2 device(s).
[ 265.718577] NVRM: None of the NVIDIA devices were initialized.) SORRY MATE
@@Dannyramos73 your board POSTs which is good, make sure you have the BIOS setting on your motherboard called 'Memory Remap' and set to [Enabled] this is the above 4GB setting I think. From what I have read there is something call PCIe BAR support and this memory mapping is set during the POST sequence. Some motherboards, like my ASRock X99 Taichi, seem to choke on the memory remapping required for the 24GB RAM in the GPU. There was nothing I could do to get it to work. So if the Memory Remap setting does not work, you could try updating your BIOS the latest for that board is from 2018 so it might work. Here is the link www.asus.com/ca-en/Motherboards/B85MG_R20/HelpDesk_BIOS/ . let me know if you get it running.
@@robingrosset6941 thanks mate li will try again ..
nice video, was hoping my HP DL380 server could handle this card not arrived yet, but has only 2 PCI x 16 gen 2 slots. it has great 128GB ram and 2 Xeon CPU. looks like I am going to external with an GPU caddy that uses USB-C thunderbolt cable to my laptop.
Cool! Nice setup. Did it work?
@@robingrosset6941 nope managed to get it going on a bf-450f rog board bio flashed to 2021 to enable above 4g. I've got 2 fans on it but gets really hot quickly.
@Scott Patnode I am currently using a BF450F board with latest flashed Bio to unlock the above 4g option, using windows 10 64bit, I am having trouble using the K80 for bitcoin gold mining, but works good for my rainbow tables, but didn't have experience in particle simulations etc even broken 1080ti etc all selling for £200+ and working ones £400 etc the prices are so high because of people using them for mining Ethereum etc at the moment. from reports k80 isn't a great mining card compared to 1050ti, 1080ti etc is makes it cheaper to buy compare to the others. another reason why 3080 etc are hard to get or prices are inflated to sky the limit as people buy them up to thrash them for mining.
do you think this would be beneficial in a home server like freeNAS running virtual machines etc.
If you can map the GPU into the virtual machine and use it like a GPU to run machine learning stuff then for sure. I know VMWare ESXi can use GPUs this way but I have not tried with freeNAS. Linus from LinusTechTips has some videos where he virtualizes GPUs like this so it might work and would be beneficial because machine learning jobs can take hours or days so running on a server makes complete sense.
I think UnRaid is the software Linus used and not freeNAS.
Okay thank you
Thanks, that's good enough. Still quite loud. Some people still buy them, lol.
Couldnt you just use the onboard temp sensor to control the fan?
Might be possible but you would still have to control the fan speed somehow and link the sensor to that. Discovered a little board on Amazon which has temp sensor and fan control in one DC 12V PWM PC CPU Fan Temperature... www.amazon.ca/dp/B019P0FLHW?ref=ppx_pop_mob_ap_share
@@robingrosset6941There is a windows app called speedfan where you can link a fan to a temp signal, that would work for sure. Maybe you can also just do it in the BIOS. I will try to get a card and test it. I have seen working cards for 190€ on eBay...
Thank you for the great tutorial, Robin! I also have a Tesla K80 from eBay, and I would like to try your mod (great application for an Arduino). However, recently my K80 has some problem with a corrupt infoRom. This is why I am reaching to you, with a small request: if your card is working, can you provide some dumps of the two infoROM sfor the two GPUs? I know this is asking a lot (especially considering that I am a complete stranger), but I would be eternally grateful! It is embarrassing to ask such a thing, and probably you don't have time for such endeavors :(.
Hello Dragos, I think the easiest way to fix this is just to use this utility which includes the bios. support.hpe.com/hpsc/swd/public/detail?swItemId=MTX_f7bea664267845a683084432dd . Best
Thank you for the idea, @@robingrosset6941! I just tried this, and it did not help. It seems that the two infoROM are independent of the vbios chips. And (for some reason) the infoROMs on my card got corrupted. So, here I am, trying to convince people to run nvflash on their working GPUs, to get these 2x 32kb dump-files. It is crazy that such a small amount of data can make or break a chip, right? Anyway...I understand that my chances are slim (I am not sure that I would touch nvflash because someone else asked me to; but it was worth a try). Thank you again for your time!
@@dragoschirila Never done this but I can try tomorrow.
Thank you very much, @@robingrosset6941 ... I would really appreciate that! To minimize the time you lose with this, below are the steps. Of course, this process should not affect your card in any way.
- unload nvidia kernel-modules: sudo rmmod nvidia_drm nvidia_modeset nvidia # (CentOS/RHEL 7 ... might be different on Ubuntu)
- get the nvflash_linux utility from here: www.techpowerup.com/download/nvidia-nvflash/
- dump the infoROM from the card, into a file (ifr file-extension is important):
+ sudo ./nvflash_linux --save inforom_dump_card0.ifr # (and select the index of the 1st K80 GPU)
+ sudo ./nvflash_linux --save inforom_dump_card1.ifr # (and select the index of the 2nd K80 GPU)
- re-load nvidia module (or restart the machine), to continue with your work: sudo modprobe nvidia
If this works, can I at least buy you an Arduino Mega to play with? It's a symbolic gesture, but I want to thank you somehow ;).
@@robingrosset6941 I can imagine that you are quite busy with other things, so I apologize if I am annoying. Do you think you will have time (at some point...no pressure) to save those infoROM files? I guess I am a little desperate, so I am "rising the stakes" :)). I would be more than happy to get you one of the newly-announced Raspberry Pi 4's (which are apparently good-enough for ML inference): arstechnica.com/gadgets/2019/06/faster-raspberry-pi-4-promises-desktop-class-performance/
maintherboard ASUS Z10PA-D8C For Dual Xeon E5-2600v3/v4 + CPU Intel Xeon E5-2678v3 12C/24T 2.5GHz Upto 3.3GHz 30M Cache. Do you think this set up is good for k80 Tesla ?
Yes that Asus board is Workstation or Server class so it should work. If there is an ‘Above 4GB’ setting in the bios be sure to enable it before installing.
Jay Thai the ‘Above 4GB Decoding’ setting is on on page 78 of the manual in bIos section under PCI Subsystem Settings. It *should* work but as nobody publishes compatibility with K80s it’s hard to be 100% but I am pretty sure it will work.
Great video, can you please provide more info on the Arduino type? I wished the hardware was readily available
In the video description there is a link to a schematic and code for the Arduino. I used a Pro Mini just to keep it small but you can use any Arduino. Here is the link github.com/rgrosset/CoolingTeslaK80
what's the hash rate for Ethereum
can they be used for video editing in Davinci resolve ?
The card uses standard CUDA drivers form NVIDIA so I believe this would just work with DaVinci Resolve. I don’t use DaVinci Resolve myself but from what I have read I think it will work.
this is really over-engineered but I'll take the duck tape idea
What nvidia software are you using to control the gpu?
It has the NVIDIA Linux drivers, with additionally CUDA libraries and TensorFlow GPU for the machine learning aspects.
@@robingrosset6941 I bought a Tesla k80, to try it out in mining but i get an error message saying "This device cannot find enough free resources that it can use" It is in my understanding that this GPU already comes in compute mode. So i don't understand what could be going on. As you might be able to tell, I am not an expert.
Could this card beat the 3090 founders edition?
No, it’s a server class GPU from a few years ago now so it will not compare to a 3090 which will beat it easily I would expect.
Is this GPU still good for machine learning next year?
It can still do machine learning and run big models bigger than most GPUs on sale today. Its performance is not as good as most recent cards. For $200-$300 it likely can't be beat particularly on model size.. But if you have $600 and don't need the RAM size then go for an NVIDIA RTX card .. it will be faster and more energy efficient. Today I would go for a 10 series or RTX series card as they are so much faster.
@@robingrosset6941 thanks for the info!
Hi thank you for the video i just wanted to know will that work in n250 board and it has no video output port so where connect the monitor cable ?
How is tesla connected to this? Is this some battery powered car?
This video has nothing to do with Tesla the car company. These are high performance computing cards from NVidia they just happen to have a brand name Tesla too. There is a connection in that cards like these are used to build artificial intelligence systems to make self driving cars. blogs.nvidia.com/blog/2019/04/23/tesla-self-driving/
might rig something similar for something similar
K80 have 2 eCPU on them so need 2 fans or 3 on them ... Also on you video that demo it's used CPU and not GPU ... not normal to get 70-80% cpu usage .
Thanks for the comment.
In a server, which is where this card normally belongs the card has no fans its cooled by airflow through the server chassis. The main thing is to ensure the card and each GPU is cooled, with this delta fan it is really overkill, these fans at full speed will lift off a desk by themselves. Its a bit too much cooling. I have tried 2 fans together and the issue comes back to where will the hot air exit the card. By design of the cooling fins the hot air flows over one GPU and then over the other there is usually a temperature increase over the 2nd GPU. You see in the video one GPU is 48 degrees C the other 54 degrees C. This is normal for these cards in server environments. There are some examples of overheating cards and normal cards at the NVIDIA forum link google "tesla-k80-overheating 38690"
The temperatures in this video are well below the thermal throttling levels.
If you look at NVIDIA SMI screen shot in the video you'll see the GPUs are loaded up 50% and 30% But notice GPU RAM usage is 100% on both GPUs.. so they are working hard and at the RAM limit. RAM limits are often the big factor in why folks use these GPUs
10:52 nvidia-smi 100% RAM used.
Its normal for CPUs to work hard pushing data to the GPUs, but it does depend on the workload. This test is CPU and GPU intensive because I am running 8 parallel machine learning workloads and an element of this workload is CPU bound because its displaying the OpenAI Gym Space Invaders user interface at high speed.
In terms of longevity for this setup I just took a K80 out of a machine after 3 years and its still running fine.
@@robingrosset6941 got both , M40 single eCPU and k80 dual , full load on image generate they stay under 70 temp with 2 fans like you did on this video
Can you sell me this as a prebuild ... Since they are $175 on amazon... How much for this? I really would love to pair this with a p2200 in my threadripper desktop for video editing.
Sorry, I don't make these to sell. They are for my own use. The first one I made is certainly overkill you don't need all that air moving over the card. I found a better combination which is to build some prebuilt parts still a little overkill but cheaper to build and from pre-exisitng parts on amazon.
Here is what I would recommend. Fan controller if you want temperature control.
DC 12V PWM PC CPU Fan Temperature Control Speed Controller CPU Alarm Module
www.amazon.ca/dp/B019P0FLHW/ref=cm_sw_em_r_mt_dp_TAE6FbSR9HZ7F?_encoding=UTF8&psc=1
And then you need a 80mm PWM fan 4pin version of this, I used 2 of these. You can also plug these fans into any motherboard header and control them from your motherboards fan controller.
GDSTIME 8025 80mm x 80mm x 25mm 12v DC Brushless Cooling Fan 4PIN
www.amazon.ca/dp/B086HCBH9S/ref=cm_sw_em_r_mt_dp_pPE6FbV3GJDCZ
@@robingrosset6941 do you think the k80 amd a quadro combo is good for extra compute power for video editing or would i be wasting a graphics card slot
These K80 cards are great. The real advantage is the massive amount of memory over others however they are a few generations back so depending on what you are doing a RTX 20xx might be better. An RTX 2080ti for context is about 4 times faster than a single GPU in a K80 but if you can use both GPUs and the memory the K80 is still a contender. To get more memory in a GPU still today you are taking $ thousands.
Hey, I just posted this video which might
be a good solution for you. ua-cam.com/video/nLnICvg8ibo/v-deo.html
Can I connect Tesla k80 into ASUS ESC2000 G2?
Your motherboard was released in 2012, 2 years before K80 was launched. It also does not have the 'Above 4GB Decoding' option in the BIOS which in my experience is required to get the system to power on with a K80 installed. Sorry to say this means that it is unlikely to work. I could be wrong, this is just my opinion based on what I have tried.
@@robingrosset6941 thanks. please, suggest me which graphic accelarator best for my workstation. And is there any way to use k80?
@@purewbattogtokh6557 A K80 will cost around $400 to $500, you could simply buy a high end graphics card from NVIDIA. It depending on the application. Based on this web site below it looks like 3dStudio works just as well with older NVIDIA GTX cards like GTX 980 or GTX 1080. These would be a cheaper route and you could install more than one. www.pugetsystems.com/labs/articles/AutoDesk-3ds-Max-2017-GeForce-GPU-Performance-816/
hi is me again, i am thinking to buy the PNY NVIDIA Quadro P1000 Scheda Video 4 GB GDDR5 PCIe for the video and the Tesla K80 do you thing that will do it fine ?
How can we get you a cup of coffee my man?
Hi, i am going to buy k80 for my workstation with 2x xeon 2683v4 in huananzhi x99-f8d, but xeon dont have integrated graphics, what should i do in this case? I think about buy cheap and old another graphic card. Sry for my bad english , hope that u understand me
Just pop in a second, lower spec GPU or standard video card for the video. Sorted. 🙂
but will it run crysis
Dragzilla 66 the better answer is it cant
Neat!
Ubuntu?
Yep, NVidia do Linux drivers for their cards too. Look it up.
this card trains ai to be a jet engine,
and you hear it
7:44 - preparing to take off
Yes that fan is overkill... but fun to play with.
I have Tesla k20xm
How do I cool this?
With your nose. Loads of videos out there on UA-cam showing you how to sneeze properly and how often.
Thanks
at my shop it costs like 8300buc
Everyone and their DIY cooling solutions for these things.
What's your contribution?
7:40