HHCJ6 Dell NVIDIA Tesla K80 24GB GDDR5 PCI-E 3.0 Server GPU Accelerator (Renewed)

It works great in ANSYS, but it runs SUPER hot like I'm not even kidding.I have 2 stacked 140mm fans blowing at it through a fan shroud and it's idling at 54 and 87 degrees on the 2 cores.Loaded it up to 100% and it shot up to 95C within 30 seconds, though it finished the analysis within that time so no complaints.If you are running this in a desktop, you better water cool it.A couple of little "features"I'm running an 8700k on Asus Maximus Hero X with 2*Strix GTX1080 already installed.This card goes through PCH and does not take up any PCIe lanes, did not have any impact on SLI performance.There is also no performance hit on my 970Evo M.2 SSD which is also going through the PCH.When it overheats and crashes, it doesn't bring the whole system down. It just freezes for a second but your computer can keep running just fine.

This GPU did not work.

Picked up for $59 and committed to not spending a dollar more to get it running on its own server. I wound up having to buy a couple flash drives for $11, so $70 all-in.I used an old bargain-bin motherboard with a dual-core FM2 CPU with integrated graphics which was living in a dusty closet box and which had a stick of RAM left on it, some leftover 25mm fans (and yes, these ARE awful to listen to) I affixed to the card directly with spare acrylic glue, and a free PSU which came with one more CPU power connector than I required for the motherboard -- and, of course, drivers from Liz II's reign. I've spent somewhere around 10 hours getting everything to work -- I tried a free flash drive which failed before caving and buying a couple decent ones; this wasted a lot of time -- I also started off working on two other motherboards, but found them to be unsuitable for different reasons. The current working setup is sitting on the floor lifted by a clothespin and a tube of chapstick, and I think the stick of RAM is bad, but the K80 is now remotely addressable when the rest of the system feels like working, by God.Just know what you're getting into: a project, and Fun with a capital F. Note, too, that setups like this are not supported by Nvidia and is not how the cards were designed to operate. Don't expect it to work with any random PC setup; in theory, you should use a certified-compatible motherboard and power supply in an enclosure made to handle cooling cards like this.

Had to redo the thermal paste on the GPU (suggest you do the same if you get one) also suggest getting fans for it as it’s built with passive cooling which is no bueno… the active cooling systems are easily found and run about $30 bucks… a bit loud but well worth it. The gpu is good for LLM local builds and some other large tasks. It functions fine and for the price I will pick three of these any day over a new 5080. Not as fast as the Nvidia Instinct but more applicable use…

Let me clarify a few things here:1. First and foremost - this is a CPU based GPU and it not a graphics accelerator with no output. This is for data mining. Does that mean you cannot reset it to be a graphics accelerator, NO not at all, but you will need a compatable GPU card to run along with it to get the benifits and you will need to manually change this on the card via command line. It is not hard and if you search the web you will find out how to do this.2. There is no HDMI interface even though it is in the specs for this device. That is just a flat out lie. In fact there are no outputs at all - VGA? NO - DVI? NO - HDMI? NO - Video port? NO there are none, Zero, Nada, Zilch output ports on this card - Hopefully that will clear that up for anyone thinking you can just plug a monitor into the card as there is a ZERO % chance of that. Did I say there are no ouputs? There are none.3. To power this card on you will need a power cable which was not included which is probably okay as it will depend on the system you are placing this into. I am putting this into a rack mounted Dell R740XD which is a 2u rack mounted server with 256GB of RAM loaded into it. (server supports up to 1.5TB of RAM with its 24 memory slots inside using 64GB chips) This card is huge so make sure you have plenty of space for it. takes up two slots and it long.4. This GPU will run extremly hot and the server I have has 6 high speed fans blowing over the 2 processors and yes that beast has 2 processors in it, each with 24 cores. Hence the Fans. and if you are putting this into a desktop you had better come up with a cooling soltution as there are no fans on this GPU and you will have to make your own. There are sites that refers to this and offers solutions so I will not recrete the wheel if you have searching skills you can find them. Some people have even made 3D-printer models for this purpose. Bottom line is if you cannot hear you fans they are probably not going to have the airflow you need. Servers have high speed fans that dwarf those desktop fans. They are not as large but they move a ton of air.5. This card has two GPUs and most likely in windows you are only going to get one of them to work.I gave this 2 stars - might deserve only 1 due to the false description of the card being HDMI. If you enjoy technology and want a challenge and have another GPU that would benifit from the added memory then this could be a fun project. If you want something plug and play then run away as that is not what this beast is.

I selected one GPU out of the two in my K80 card in MATLAB, and Bench marked the Renewed K80 using the gpuBench function.It is just as expected, the performance approaches the score of K40. (I suppose the K80 is just, roughly speaking, two K40s integrated together).One inconvenience for PC users is that you need to put some really-really powerful fans to force the air flow.Another small hinder is that the GPU memory chip one the back side has no metal back-plate or heat sinks attached to it, when running the bench, they become really hot. Unless you actually put it in a server or a case with server-level air flow design, I might try to put some heat sinks on them for PC usage.But I am glad that this card work out eventually.

Unfortunately wasn't compatible for my Precision workstation

I got 2 of these at the $80 mark heck yea for an AI server its status que but still 48gb vram. cant beet the price