All of these comments... The cost of something is not a measurement of its efficiency. I find some of these comments downright disrespectful. Mr. Qasem, Great work here!
If you're worried about getting performance for the dollar value, the cost of a case for a server is a pittance compared to the card. Get used hardware because there's a ton of it from various sources (including server farms that have upgraded since) without much trouble. I'm 100% sure, no matter what your application is, that for 4000$ one could assemble at LEAST twice the amount of power, in a more convenient package like an actual server case with actual cooling you can literally remote into from wherever you are. So not only would you get more power for your money, exactly what you're attempting to do here, you'd have better accessibility as well as scaliblity because depending on the application you can split a task among-st different machines, further reducing cost per performance because old hardware = cheap hardware. So please, What's this research? Or is this video just an excuse to show everyone you managed to get a fancy new videocard to work in your desktop, that i'd understand.
Ok, in case people don't know - in most cases you have to buy a multi-gpu motherboard to use one of these, even if you're using an APU for the processor, as many motherboards will only allow one graphics adapter. Doesn't matter that it has no graphics output, it still counts as one to the computer. Secondly, you will need a much bigger power supply, and don't think you can get away with a high-power cheap brand. It will just go bang, literally. Don't make the mistake of thinking you can just get a cheap old server instead because the power supplies tend to be unique and you won't be able to upgrade. My advice also is to buy a whole bunch of micro heatsinks and use thermal glue to attach them to the chips on the reverse side of the motherboard, and set up a fan directly blowing across those too. They put out a fair bit of heat too. Watch out also for centrifugal fans that get too blocked with dust, can become unbalanced, and the vibration can also upset operation of your hard disk. I would also set up an automatic temperature-controlled fan so it adjusts itself rather than having to do it manually
Thank you for this hack, it helped me a lot with cooling my card as I was very, very worried at first when it started shutting down on me. This is definitely helping me in my research, so thanks!
The best system for this card which can be used as a barebone workstation is from Supermicro and the design already compensates for the heat issues. It is called the: Supermicro Superserver 7048GR-TR or you can speak to a company situated in the United States called Boxxtech that use the same motherboard but customised chassis. Their version of the system is the Boxxtech Apexx 5
hey Mohamed, aren't these cards made for server cases that move air through the case from front to back with their large (loud) fans? nice hack though! ^^
I would think that they were designed for special servers. Since I didn't have one, I designed my own fan duct. Then I made this video. I thought that this might help those who are in my situation.
+Harry Lawton it's is precisely because I couldn't afford to buy a server that I made this cheap duct. Being able to afford one thing doest mean I can afford the other.
Mohamed - We built a similar workstation for CST Microwave applications, only we used the ANTEC TriCool 120mm fan at 3000 RPM - 79 CFM. The fan you used is only 47 CFM. Our software works well and the Tesla K80 works well, however it still gets very hot (90C) during big computations. We run the nvidia-smi on a 3 second loop while using CST. One of our problems is that once CST grabs the K80, it keeps it - maintaining 65-75 watts per GPU until the software is closed. Matlab does the same thing too. We are considering one of the 120mm Delta fans at 150 to 200 CFM to increase the air flow.
The Thermaltake A1888 I installed is 3000 rpm as well. It's a bit smaller than the Antec TriCool. I didn't consider the CFM before. I will definitely upgrade to a fan that pushes more air next time. I don't know how much gain I would get out of it, but any bit of computational speed is welcome. True, Matlab keeps the GPU working until it is closed. I've read some complaints about that. I don't know why it keeps working the GPU even if there were no computations going on.
Hi Roger Cox and Mohamed Qasem, I am trying to get my k80 working on my pc but it doesn’t work with its motherboard, I think because it has no above 4G decoding option in its BIOS. What motherboard are you using to get the k80 to work? Can you recommend a motherboard please?
Very smart indeed. Your tube will work like a vacuum cleaner, and it makes really sense. Good job. I am gonna buy a card on ebay. They are very cheap at the moment ($550-$700 oct 2019)
Thank You for making the effort of creating and uploading this video, just to help others that might run into this or similar problems, or are just considering doing something similar. Best regards.
Hello Mohammed .. Thanks for the video .. i need to know more specifics about the installation and the drivers .. so i have this tesla card and i have the quadro k6000 .. so u said i have to install the tesla first and install its driver .. then install the quadro k6000 and install its driver ?
here's an idea: what about making a small duct, but stack multiple fans together to amplify their performance? I've seen someone do that with a bunch of 120mm fans for cooling a CPU, and that resulted in extremely high performance. Try stacking 4 or 5 of those little fans together, all pulling air through, and it should cool sufficiently well.
That would be a great spot for the 'Former' Thermaltake Smart fans with the thermal sensor to automatically increase the cooling when things get worm. Shame the fans were discontinued and there is nothing like them now that is of good quality (not everything that has a 120mm or 80mm fan has a x86 CPU to run fan control software on, lol). That's a great exhaust fan cooling duct, it looks good. If you do not have one, a second intake fan up in the CD ROM area may help keep temps low in the computer. A single intake fan just is not enough for modern computers. The CPU may be pulling air away from the K80, so a second intake fan for the CPU area may help, It's just a thought.
Hey Mohamed, awesome video. Can I ask you a question? How much CPU usage do you see while the GPU is processing? We have a supermicro server with two K80's installed and while the K80's are processing the CPU is close to 95% busy. I'm wondering if that is normal or if there might be something to change in the motherboard bios?
In my experience, the CPU usage doesn't increase that much. The calculations are all done on the GPU. I don't know that you need to change any settings beyond those I mentioned in the video.
Hi Paul Gomez and Mohamed Qasem, I have a k80 also but it doesn’t seem to work with my motherboard, I think because it has no above 4G decoding option in its BIOS. Did you find a motherboard that works with your k80? Can you recommend one please? Thanks.
Hi Mohamed, thanks posting this! Your tutorial inspired me to buy a Tesla K80 card from eBay earlier this year. It has been working just fine until today (using it for fluid dynamics simulations). However, since today it let me down, due to a corrupt infoROM. Do you still have your K80? If so, could you dump the infoROM from the two GPUs to some files? I know this is a lot to ask, but I am a little desperate, sorry :). Instead of a delightful number-crunching GPU, now I just have a very expensive paperweight...
2 years on, do you still use this card? I'm planning to get dual AMD Instincts and want to do a similar cooling solution, but pushing inside the case instead of pulling outside, to avoid negative pressure in the case (minimise dust).
did you thought about using Thermoelectric Peltier Water Cooling Device and get a bit more room on the thermal budget that you need to cover even if you push it ?? good approach and very pragmatic ... just another option to compare with
as a matter of fact I got one of these... and I'm sure I'll have the same problem.... the size of the duct you made... is is necessary? or whaever size as long as the fan keeps cooling is ok?
those bigger blower style fans, maybe? still pretty cheap but probably a cleaner solution because you can put it inboard pushing out, and possibly more effective considering they are usually much better with static pressure.
+Mohamed Qasem excuse my ignorance of workstation type cards and use cases but are they overclockable? if so would that even be advised because reduced stability/reliability?
Can I ask a question. I would like to with this card, how long it takes to train let say VOC12. or if you have any other popular benchmark dataset trained. can you let me know how long it takes?
What if you took the heat sink shroud off and installed two fans on it? Or remove the heat sink entirely and replace with a 3rd party GPU heat sink and fan?
4:30 it's not about "modern" motherboards or not , it's about the chip and the CPU if it's made with GPU inside of it or not . both are made to be together , like generations . for example Intel Z87 Z97 Z170 they made to fit with CPUs that have integrated GPUs , those are usually can take maximum ram of 32GB . and other chipsets are made to fir with CPUs that haven't GPUs inside of them , these chipset is for large scaled work over the the other one with GPU inside of the CPU , example like x79 x99 (x99 can work with 5th gen first real 8 cores intel's CPU) and this days can take 64GB of ram . anyways is this card have something to do with 3d rendering and video composing softwares , 3d photo-realistic simulation (cinema industry) , if it can go this way? , i'm learning/using these softwares , will be amazing if yes
it is possible to use this model of tesla k80 for realtime visualisation (activeshade, nvidia maximus) of 3dsmax and vray like the Tesla K20c that i have or this must only be used for computational and research programs only?
Hello there, I am just thinking of getting one of this card for my computations. Which motherboard are you using to run the Tesla K80?¿? thanks for all Manuel
Hello! I have an intel i9 9920X and this processor does not have a GPU included, so i supose i must have a different video card in my system already like a cheap ATI for example to not overdrive two Nvidia drivers if i have an nvidia video card like a GTX for example. So, could be this a solution to install K80 along an Nvidia GTX card? Need an ATI video card if i dont have a cpu with gpu intel HD included? SO, from your video, i understand you take out MSI video card, going with GPU from processor, install K80, modify the BIOS, and after that....? How you install the MSI video card? I assume it's an NVIDIA card or is an AMD VIDEO CARD? :) Thanks.
Did you consider installing a liquid cooler for the GPU, I have one for my CPU it used to heat to more then 100C and now it goes like 60 max. I know it's possible for GPUs but I didn't install one mine is good. It's like 60$ a single fan one a double would be like 120$. It take space in the case it is frustrating if you don't have space.(very effective)
I just compared the FLOPS of this card with the GTX1080 graphics card. Apparently the 1080 has around 8,4 TFLOPS and the K80 has 2,9 TFLOPS. So wouldn't you get a better performance for less money with a standard graphics card? Or am I completely on the wrong track and these cards can't be compared? If so, what makes them different?
Did you watch at least 3 minutes of video? This card is not consumer grade card that you will use for playing games or anyting similar. Its designed to be used for complex task such as physics simulations,scientific research etc..
+Steva Мађија what do you think gaming cards do? draw pictures out of their fantasies? dont make it out so simple as you do. a flop is a mathematical process, so addiction and multiplication. the big price difference isn't due to increased performance but due to enabled features like double precision, cherry picked cores of either a gtx 970 or 980, availability time/market targeting, customer support and for big orders you even get a nvidia worker to help programming what you want. basically speaking the dude in the video could have saved a few thousand bucks if he didn't plan on using any of the features except the double precision, and get himself a Titan z, because that card happens to have that feature too.
and I almost forgot to mention. that Tesla card are the ones nvidia uses to power their grid servers, doing exactly the same as their Geforce cards at a larger scale basically. probably the second biggest advantage to this card specifically is the super high vram of 24gb with error correction
The Tesla has a bunch of extra features that are important for research and computation that aren't present on GTX cards. stuff like way more VRAM, double precision compute, ECC VRAM, etc.
+chopcookies Thanks! That was the answer I was looking for. I mean I knew that there must have been a difference, because otherwise no one would buy it, but I didn't know what exactly sets them apart.
Fellow Commenters: the paraphrasing and generalizations I'm reading make it obvious you have very little understanding of the different purposes of accelerators. There are far more effective ways to stroke your ego
It's not a standard heatsink with lots of mass and many fins with lots of surface area and heatpipes. It is like a closed rectangular tube with a handful of straight channels inside. It needs massive airflow through the channels to get the heat out.
this card dont have display.me planning to build my beast with a threadripper 2970wx.it dont have igpu too..so,i need to use another nvidia gpu to use it?will i get 24gb vram if use a GTX750 for display and this as a vram?pls help if any1 knows these infos.
Hello. I have little question , i bought two K80 tesla for calculating. I have several problem. problem is my tesla are not working in windows. but you seems use in window? am i right?
I am looking at a cheaper way to render now the price has crashed on these in the used market!! Very cool very useful info! thinking about finding the cheapest board to mount 8x tesla cards.. or more.
if the card overheats just contact nvidia. if it would be a consumer grade card i would buy a aftermarket cooler (all in one loop or something like that) but with a 5k card i would not risk it (just like that)- first contact customer support.
It gets hot because he doesn't have the airflow nescessary to cool the card in his desktop. This card has no fan to move air with but is designed to be run in a server which has very powerful fans that can move the air to cool the car. His solution was good.
@Mohamed can you elaborate on that please. I wonder if you could've just gotten a gaming card that costs 10 times less to do the same job and with less headsache
+intra morph I don't think so, for any kind of research you'd need very high quality memory not found on consumer cards, also if he plans to do any kind of neural networking with cuda cores a lot of VRAM will be consumed and a lot of CUDA cores would also be needed.
Actually you could, especially pascal is already available with higher VRAM capacity. But when we are talking about computation for research, usually researchers need more sophisticated GPU. Gaming cards is good for computation only up to single precision which is a standard for reproducing real time rendering for games. While researchers usually need something called double precision computing which GPU for gaming cant handle it nicely. Moreover researchers do not want errors in their calculation due to bit error so they need VRAM with ECC equipped.
Adnan Hutomo Honestly those high end Cuda Core cards should come with some type of cooler already installed. I understand they usually go in very well cooled servers and so on but you should have the option to choose how and where to install it especially after paying so much for it
just bought two of those badboys on Amazon for 450$ each... used though... directly from Dell but the cards are brand new, how can you tell if the card is new?... well the wrapper and the original bios was still on the card... thanks Dell... it's a Titan XP with 24 gb vram for 450 shipped LOL... I fell a bit sorry you bought it for 5K though..
Hi Hernan Ventura, I have a k80 also but it doesn’t work with my motherboard, I think because it has no above 4G decoding option in its BIOS. Did you find a motherboard that works with your k80? Can you recommend one please??
Uhm, why exactly did you need a tesla, for most professional applications a quadro is fine, it has error correction double precision floating point accuracy and all that good stuff
Well not really. The two products differ to much in their intended usage. In absolute numbers (when you compaire this 2 gpu card with 2x 1080 (to also get 2cpu's)) you will see that the K80 has 4992 cuda cores @ 870Mhz and the 1080 combo will have 5120 cuda cores @ 1733Mhz (stock). So in absolute computing power the 1080's are faster. But both are optimized in another way.
It depends. A single 1080 would be stronger in single-precision (stronger than P100 too when overclocked), but if you need double-precision a GTX card won't do. You won't be doing any gaming on a tesla card though.
talk more about your machine learning research? neural netwroks / depe learning / SVM , what are you doing? what algorithms are you using ? are you working for a big company or doing personal computations(you don't need a card this powerful to do that)
I've developed my own neural network/method. It's my own algorithm, and yes, it's another deep learning method. I'm not working for a company. I work at the College of Technological Studies in Kuwait. However, right now, I'm on sabbatical at the University of Southampton, UK, and I'm doing this work for my own research. This card performs very fast computations on large matrices. It can outperform the best CPUs (up to 20x with large matrices).
All of these comments...
The cost of something is not a measurement of its efficiency.
I find some of these comments downright disrespectful.
Mr. Qasem, Great work here!
Thanks. I was hoping to help those who want to do the research without paying extra for the server casing.
If you're worried about getting performance for the dollar value, the cost of a case for a server is a pittance compared to the card. Get used hardware because there's a ton of it from various sources (including server farms that have upgraded since) without much trouble. I'm 100% sure, no matter what your application is, that for 4000$ one could assemble at LEAST twice the amount of power, in a more convenient package like an actual server case with actual cooling you can literally remote into from wherever you are. So not only would you get more power for your money, exactly what you're attempting to do here, you'd have better accessibility as well as scaliblity because depending on the application you can split a task among-st different machines, further reducing cost per performance because old hardware = cheap hardware. So please, What's this research? Or is this video just an excuse to show everyone you managed to get a fancy new videocard to work in your desktop, that i'd understand.
Ok, in case people don't know - in most cases you have to buy a multi-gpu motherboard to use one of these, even if you're using an APU for the processor, as many motherboards will only allow one graphics adapter. Doesn't matter that it has no graphics output, it still counts as one to the computer. Secondly, you will need a much bigger power supply, and don't think you can get away with a high-power cheap brand. It will just go bang, literally. Don't make the mistake of thinking you can just get a cheap old server instead because the power supplies tend to be unique and you won't be able to upgrade. My advice also is to buy a whole bunch of micro heatsinks and use thermal glue to attach them to the chips on the reverse side of the motherboard, and set up a fan directly blowing across those too. They put out a fair bit of heat too. Watch out also for centrifugal fans that get too blocked with dust, can become unbalanced, and the vibration can also upset operation of your hard disk. I would also set up an automatic temperature-controlled fan so it adjusts itself rather than having to do it manually
4,000$ card
Doesn't come with a cooler.
K
it's probably meant for server cases that have huge loud ass fans that move air through the whole case from front to back
+Thunder Draws yeeeep.
+Schiken cards today shut down themselfs to prevent that from happening
+Schiken why did you commented then?
+Schiken ahhhhhhh okay well then...
Thank you for this hack, it helped me a lot with cooling my card as I was very, very worried at first when it started shutting down on me. This is definitely helping me in my research, so thanks!
I'm glad to be of help. :)
The best system for this card which can be used as a barebone workstation is from Supermicro and the design already compensates for the heat issues. It is called the: Supermicro Superserver 7048GR-TR or you can speak to a company situated in the United States called Boxxtech that use the same motherboard but customised chassis. Their version of the system is the Boxxtech Apexx 5
hey Mohamed, aren't these cards made for server cases that move air through the case from front to back with their large (loud) fans?
nice hack though! ^^
I would think that they were designed for special servers. Since I didn't have one, I designed my own fan duct. Then I made this video. I thought that this might help those who are in my situation.
they are designed for servers like these:
www.supermicro.com/products/system/4U/4028/SYS-4028GR-TR2.cfm
+Mohamed Qasem if you can afford a Tesla gpu why not just buy a server case?
+Harry Lawton it's is precisely because I couldn't afford to buy a server that I made this cheap duct. Being able to afford one thing doest mean I can afford the other.
Idiot.
Mohamed thanks for the video. I know it's not like a gaming GPU, but it would be cool to see some gaming test/FPS on the top tier latest games. lol
Thank you Mohamed for a great idea, that has so many applications. You can't push air from front (no space) so suction is the way>
Mohamed - We built a similar workstation for CST Microwave applications, only we used the ANTEC TriCool 120mm fan at 3000 RPM - 79 CFM. The fan you used is only 47 CFM. Our software works well and the Tesla K80 works well, however it still gets very hot (90C) during big computations. We run the nvidia-smi on a 3 second loop while using CST. One of our problems is that once CST grabs the K80, it keeps it - maintaining 65-75 watts per GPU until the software is closed. Matlab does the same thing too. We are considering one of the 120mm Delta fans at 150 to 200 CFM to increase the air flow.
The Thermaltake A1888 I installed is 3000 rpm as well. It's a bit smaller than the Antec TriCool. I didn't consider the CFM before. I will definitely upgrade to a fan that pushes more air next time. I don't know how much gain I would get out of it, but any bit of computational speed is welcome.
True, Matlab keeps the GPU working until it is closed. I've read some complaints about that. I don't know why it keeps working the GPU even if there were no computations going on.
Hi Roger Cox and Mohamed Qasem, I am trying to get my k80 working on my pc but it doesn’t work with its motherboard, I think because it has no above 4G decoding option in its BIOS. What motherboard are you using to get the k80 to work? Can you recommend a motherboard please?
Hello Mohamed. Could you please tell me which motherboard and other specs are you using in order to make your tesla K80 work? Thanks a lot
Thanks for the video. What make and model of motherboard did you use? Did you run the K80 in Windows or Linux OS?
Robin, did you figure out what motherboards would work with this card? I saw you have an Asus z170, but any other budget friendly one?
Hey Mohamed! Happy New Year. Would you still recommend this GPU and cooling solution today?
just picked up 2 of them for $180
Very smart indeed. Your tube will work like a vacuum cleaner, and it makes really sense. Good job. I am gonna buy a card on ebay. They are very cheap at the moment ($550-$700 oct 2019)
Buys a card intended to be in a server rack with high flow fans. Surprised when the card fails to cool it's self.
Thank You for making the effort of creating and uploading this video, just to help others that might run into this or similar problems, or are just considering doing something similar.
Best regards.
Great video. (1) How did you punch the holes in the aluminum? (2) How thick is the aluminum sheet? Thanks!
I have the same problems with my K10, will fix a ventilator to it like yours now. Thanks for this video
Hello Mohammed .. Thanks for the video .. i need to know more specifics about the installation and the drivers .. so i have this tesla card and i have the quadro k6000 .. so u said i have to install the tesla first and install its driver .. then install the quadro k6000 and install its driver ?
Great video!! What was your motherboard sir? My system looks like is not compatible.
Dear brother, up to your knowledge....
Can I use this GPU with hp z440 workstation... I am planning to use this GPU in quantum espresso calculations
here's an idea: what about making a small duct, but stack multiple fans together to amplify their performance? I've seen someone do that with a bunch of 120mm fans for cooling a CPU, and that resulted in extremely high performance. Try stacking 4 or 5 of those little fans together, all pulling air through, and it should cool sufficiently well.
A readily available solution costs $25 on Ebay. But there are still pretty cheap Teslas with active cooling
That was a given, card is made for server racks that push air through the cards
That would be a great spot for the 'Former' Thermaltake Smart fans with the thermal sensor to automatically increase the cooling when things get worm. Shame the fans were discontinued and there is nothing like them now that is of good quality (not everything that has a 120mm or 80mm fan has a x86 CPU to run fan control software on, lol).
That's a great exhaust fan cooling duct, it looks good. If you do not have one, a second intake fan up in the CD ROM area may help keep temps low in the computer. A single intake fan just is not enough for modern computers. The CPU may be pulling air away from the K80, so a second intake fan for the CPU area may help, It's just a thought.
Hi Mohamed. Can you share the model of the motherboard you used for your build? Thanks!
do you have temp readings in comparison to passive and fanned cooling? Wanted to see how much it cools the card down. Thanks!
Hey Mohamed, awesome video. Can I ask you a question? How much CPU usage do you see while the GPU is processing? We have a supermicro server with two K80's installed and while the K80's are processing the CPU is close to 95% busy. I'm wondering if that is normal or if there might be something to change in the motherboard bios?
In my experience, the CPU usage doesn't increase that much. The calculations are all done on the GPU. I don't know that you need to change any settings beyond those I mentioned in the video.
Thanks!!
Hi Paul Gomez and Mohamed Qasem, I have a k80 also but it doesn’t seem to work with my motherboard, I think because it has no above 4G decoding option in its BIOS. Did you find a motherboard that works with your k80? Can you recommend one please? Thanks.
Do you plan to get the nvidia pascal GP100 GPU?
Out curiosity what's your system specs look like if you don't mind? Processor, motherboard ram etc.
Hi Mohamed, thanks posting this! Your tutorial inspired me to buy a Tesla K80 card from eBay earlier this year. It has been working just fine until today (using it for fluid dynamics simulations). However, since today it let me down, due to a corrupt infoROM. Do you still have your K80? If so, could you dump the infoROM from the two GPUs to some files? I know this is a lot to ask, but I am a little desperate, sorry :). Instead of a delightful number-crunching GPU, now I just have a very expensive paperweight...
2 years on, do you still use this card?
I'm planning to get dual AMD Instincts and want to do a similar cooling solution, but pushing inside the case instead of pulling outside, to avoid negative pressure in the case (minimise dust).
did you thought about using Thermoelectric Peltier Water Cooling Device and get a bit more room on the thermal budget that you need to cover even if you push it ??
good approach and very pragmatic ... just another option to compare with
as a matter of fact I got one of these... and I'm sure I'll have the same problem.... the size of the duct you made... is is necessary? or whaever size as long as the fan keeps cooling is ok?
Cant believe how much those cards costed when you made this video. They cost less than $100 on the second hand market today.
مبدع كعادتك يا دكتور 🌺
those bigger blower style fans, maybe? still pretty cheap but probably a cleaner solution because you can put it inboard pushing out, and possibly more effective considering they are usually much better with static pressure.
I purchased two differently sized blowers. Neither would fit in the case. This was the reason I designed an external duct using this type of fan.
+Mohamed Qasem ahh, still working out well?
+Miles11we yes. Perfectly.
+Mohamed Qasem excuse my ignorance of workstation type cards and use cases but are they overclockable? if so would that even be advised because reduced stability/reliability?
I don't know the answer to this question. Sorry.
Can I ask a question. I would like to with this card, how long it takes to train let say VOC12. or if you have any other popular benchmark dataset trained. can you let me know how long it takes?
Thank you so much for this video it's actually helps a lot. Can you make a video about its working in your research. for example in MD simulations?
What is motherboard which you are using for K80. Had a problem to install 2090 at Asus Sabertooth r2.0. Bios have no option "assign Irq to vga".
What if you took the heat sink shroud off and installed two fans on it? Or remove the heat sink entirely and replace with a 3rd party GPU heat sink and fan?
4:30 it's not about "modern" motherboards or not , it's about the chip and the CPU if it's made with GPU inside of it or not . both are made to be together , like generations . for example Intel Z87 Z97 Z170 they made to fit with CPUs that have integrated GPUs , those are usually can take maximum ram of 32GB .
and other chipsets are made to fir with CPUs that haven't GPUs inside of them , these chipset is for large scaled work over the the other one with GPU inside of the CPU , example like x79 x99 (x99 can work with 5th gen first real 8 cores intel's CPU) and this days can take 64GB of ram .
anyways is this card have something to do with 3d rendering and video composing softwares , 3d photo-realistic simulation (cinema industry) , if it can go this way? , i'm learning/using these softwares , will be amazing if yes
it is possible to use this model of tesla k80 for realtime visualisation (activeshade, nvidia maximus) of 3dsmax and vray like the Tesla K20c that i have or this must only be used for computational and research programs only?
You should consider the NZXT KRAKEN G10 Liquid Cooler for GPUs. as far as I know it should work with yours.
Would it work for Nvidia K40?
Just putting watercooling on this card, would be easier.
You can't use water cooling. The heat sink is encased in plastic.
You can always remove the heat sink
+Mohamed Qasem when you watercool, ypu always remove the stock heatsink and plastic case
Sure, why not?
i doubt there are waterblocks for it
Hello there,
I am just thinking of getting one of this card for my computations. Which motherboard are you using to run the Tesla K80?¿?
thanks for all
Manuel
Manuel Luis Aznar I think he mentions it.
Hellowdo you use adobe premiere pro cc, did you find this card its fast and smouve during playing multicame 4k? thank you. salutations. M
انا عندي k10 وركبت مروحه على الكرت بالنص وقسمت التوزيعه الهواء وصار المرت شغال مايفصل والامور تمام. بس طريقتك وايد أحسن اشوفها والله يعطيك الصحه والعافيه
Hello!
I have an intel i9 9920X and this processor does not have a GPU included, so i supose i must have a different video card in my system already like a cheap ATI for example to not overdrive two Nvidia drivers if i have an nvidia video card like a GTX for example. So, could be this a solution to install K80 along an Nvidia GTX card? Need an ATI video card if i dont have a cpu with gpu intel HD included?
SO, from your video, i understand you take out MSI video card, going with GPU from processor, install K80, modify the BIOS, and after that....? How you install the MSI video card? I assume it's an NVIDIA card or is an AMD VIDEO CARD? :)
Thanks.
Just put more fans on your case, so you have more positive or negative pressure and It should be fine.
Putting more fans in the case doesn't help. I've already tried that.
Can it Run In Windows For Photoshop , Autocad Or Some Cuda Software For Deep Learning ?
love how he puts the card down, VERY SLOWLY.
Heck may as well water cool it. But yea tesla cards are meant to be in server racks that have really powerful fans in them anyway.
Did you consider installing a liquid cooler for the GPU, I have one for my CPU it used to heat to more then 100C and now it goes like 60 max. I know it's possible for GPUs but I didn't install one mine is good. It's like 60$ a single fan one a double would be like 120$. It take space in the case it is frustrating if you don't have space.(very effective)
+rashed al rashed You can't use liquid cooling with the Tesla K80. You can't access the heat sink. Liquid cooling works with some GPUs.
Mohamed Qasem I see, thanks for informing me.
What kind of research are you doing?
Hi Mohamed Qasem, what is the motherboard you use? That has the bios that way?
You could have watercooled it with an NZXT G10, would perform amazingly under water. Would only cost you about $150 or so.
bitspower released fullcovered waterblock for tesla k80! your problem will go easy now!
I just compared the FLOPS of this card with the GTX1080 graphics card. Apparently the 1080 has around 8,4 TFLOPS and the K80 has 2,9 TFLOPS. So wouldn't you get a better performance for less money with a standard graphics card?
Or am I completely on the wrong track and these cards can't be compared? If so, what makes them different?
Did you watch at least 3 minutes of video? This card is not consumer grade card that you will use for playing games or anyting similar.
Its designed to be used for complex task such as physics simulations,scientific research etc..
+Steva Мађија what do you think gaming cards do? draw pictures out of their fantasies? dont make it out so simple as you do. a flop is a mathematical process, so addiction and multiplication. the big price difference isn't due to increased performance but due to enabled features like double precision, cherry picked cores of either a gtx 970 or 980, availability time/market targeting, customer support and for big orders you even get a nvidia worker to help programming what you want. basically speaking the dude in the video could have saved a few thousand bucks if he didn't plan on using any of the features except the double precision, and get himself a Titan z, because that card happens to have that feature too.
and I almost forgot to mention. that Tesla card are the ones nvidia uses to power their grid servers, doing exactly the same as their Geforce cards at a larger scale basically. probably the second biggest advantage to this card specifically is the super high vram of 24gb with error correction
The Tesla has a bunch of extra features that are important for research and computation that aren't present on GTX cards. stuff like way more VRAM, double precision compute, ECC VRAM, etc.
+chopcookies Thanks! That was the answer I was looking for. I mean I knew that there must have been a difference, because otherwise no one would buy it, but I didn't know what exactly sets them apart.
any ideas for internal ducts .it's just looks ugly on our workstation.??
Fellow Commenters: the paraphrasing and generalizations I'm reading make it obvious you have very little understanding of the different purposes of accelerators. There are far more effective ways to stroke your ego
Well put.
Damet garm dada, pooldariha!!! Vali khodavakili honar nazde iranian asto bast!
What is the mainboard what you use ?
Are you running that from Molex connectors?
Wouldn't it have been easier to just take the card's shroud off and put 3 90mm fans on it's heatsink?
Nice video! Expensive card! Isn't that second GPU a GTX 660?
couldnt you just remove the plastic on the outside and mount some fans to it with a fan controller?
It's not a standard heatsink with lots of mass and many fins with lots of surface area and heatpipes. It is like a closed rectangular tube with a handful of straight channels inside. It needs massive airflow through the channels to get the heat out.
this card dont have display.me planning to build my beast with a threadripper 2970wx.it dont have igpu too..so,i need to use another nvidia gpu to use it?will i get 24gb vram if use a GTX750 for display and this as a vram?pls help if any1 knows these infos.
Hello. I have little question , i bought two K80 tesla for calculating. I have several problem.
problem is my tesla are not working in windows.
but you seems use in window? am i right?
I bought one for 140 bucks like wtf is this looks cool. I wanna sell it for more.
for 3d max it work? its fast? any example?
I am looking at a cheaper way to render now the price has crashed on these in the used market!! Very cool very useful info! thinking about finding the cheapest board to mount 8x tesla cards.. or more.
May the 4th be with you!!!
1:50... I don't see a girl :/ was disappointed
if the card overheats just contact nvidia. if it would be a consumer grade card i would buy a aftermarket cooler (all in one loop or something like that) but with a 5k card i would not risk it (just like that)- first contact customer support.
It gets hot because he doesn't have the airflow nescessary to cool the card in his desktop. This card has no fan to move air with but is designed to be run in a server which has very powerful fans that can move the air to cool the car.
His solution was good.
Shamie Ramsamuj
I guess you are right.
with this video card cpu mining ? good or not
new model "cpu mining" same no screen exits DVI or HDMI ..i think it is a miner card
Robert Dulany and let's say you can get this k80 for a fraction of market cost? Would you recommend it for mining? If you took ROI out.
how do install the drivers for the tesla without video output?
why would they make such a hot card without any effective ventilation and so much crashing?
+transcendtient but what about for cases like his? like situations not chassies
The card costs more then my PC, then double it lol
jesus $15 for a sleeve bearing 80mm Thermalfake fan
what kind of research u do? and what aplications?
I'm working on deep learning.
@Mohamed can you elaborate on that please. I wonder if you could've just gotten a gaming card that costs 10 times less to do the same job and with less headsache
+intra morph I don't think so, for any kind of research you'd need very high quality memory not found on consumer cards, also if he plans to do any kind of neural networking with cuda cores a lot of VRAM will be consumed and a lot of CUDA cores would also be needed.
Actually you could, especially pascal is already available with higher VRAM capacity. But when we are talking about computation for research, usually researchers need more sophisticated GPU. Gaming cards is good for computation only up to single precision which is a standard for reproducing real time rendering for games. While researchers usually need something called double precision computing which GPU for gaming cant handle it nicely. Moreover researchers do not want errors in their calculation due to bit error so they need VRAM with ECC equipped.
Adnan Hutomo Honestly those high end Cuda Core cards should come with some type of cooler already installed. I understand they usually go in very well cooled servers and so on but you should have the option to choose how and where to install it especially after paying so much for it
just bought two of those badboys on Amazon for 450$ each... used though... directly from Dell but the cards are brand new, how can you tell if the card is new?... well the wrapper and the original bios was still on the card... thanks Dell... it's a Titan XP with 24 gb vram for 450 shipped LOL... I fell a bit sorry you bought it for 5K though..
I need your help, I have a gpu tesla k80 , please recommend me a motherboard and a processor compatible with this gpu.
Hi Hernan Ventura, I have a k80 also but it doesn’t work with my motherboard, I think because it has no above 4G decoding option in its BIOS. Did you find a motherboard that works with your k80? Can you recommend one please??
What if my motherboard doesn't have video output? My CPU doesn't have iGPU.
Brilliant solution.
Nvidia did not put a fan on it because the final product would be very expensive.
NO its an accelerator. Its supposed to be in a server case where it would cool passively with multiple cars stacked on top of each other.
Just calling it a GPU is going to confuse the more simple people, I'm afraid...
Because it's still a GPU, technically.
you better be ! now over clock
eggs Not in the traditional sense. It doesn't output video. It's purely an acceleration/computation card.
+Paul J There is a GPU inside, it only lacks outputs. This makes it an accelerator and not a video card, but it's still technically a GPU.
+Paul J There is a GPU inside, it only lacks outputs. This makes it an accelerator and not a video card, but it's still technically a GPU.
Uhm, why exactly did you need a tesla, for most professional applications a quadro is fine, it has error correction double precision floating point accuracy and all that good stuff
why isn't it in the geforce family?
Which one is faster?
the gtx 1080 or NVIDIA Tesla K80?
tesla k80
Loki_Nyx_319 __ then this should be the fastest GPU Right?
Jim Romines the amd radeon r9 295xs is
Well not really. The two products differ to much in their intended usage. In absolute numbers (when you compaire this 2 gpu card with 2x 1080 (to also get 2cpu's)) you will see that the K80 has 4992 cuda cores @ 870Mhz and the 1080 combo will have 5120 cuda cores @ 1733Mhz (stock). So in absolute computing power the 1080's are faster. But both are optimized in another way.
It depends. A single 1080 would be stronger in single-precision (stronger than P100 too when overclocked), but if you need double-precision a GTX card won't do. You won't be doing any gaming on a tesla card though.
Whats the case in the background at the begining
The Titan V is $3K and outperforms this card I think, I hope you've upgraded to it.
why not take off the shell and install a water cooling device?
I would (figuratively) kill to get one of those!
what temps do you get at load?
Don't cheap out on the cooling of that card! Poor card.
put a Aftermarket Cooler like Morpheus on it, it will run 50-60 C....instead of playing engineer.
عربي ؟ الله يوفقك في شغلك بس اتمنى انك تسوي فيديو عربي تتكلم عن مشروعك و عن الكرت
كل من يعمل بهذا المجال يعرف اللغة الإنجليزية بإجادة، اللغة العربية تخدم شريحة لا تعمل بهذا المجال. أقول هذا الشيء وأنا آسف على هذه الحالة.
talk more about your machine learning research? neural netwroks / depe learning / SVM , what are you doing? what algorithms are you using ? are you working for a big company or doing personal computations(you don't need a card this powerful to do that)
I've developed my own neural network/method. It's my own algorithm, and yes, it's another deep learning method.
I'm not working for a company. I work at the College of Technological Studies in Kuwait. However, right now, I'm on sabbatical at the University of Southampton, UK, and I'm doing this work for my own research.
This card performs very fast computations on large matrices. It can outperform the best CPUs (up to 20x with large matrices).
can you send to me the PC or the server model which you connect the Card on it?
Can it work with DaVinci Resolve ?
@@WillCarterTech oh nooo....
@@WillCarterTech on Linux maybe?
@@WillCarterTech Well Resolve is developed on Linux in the first place
@@WillCarterTech
It should work better than other operating systems
It might be a good time for me to try Linux 🙂
very nice diy job... very interesting...