NVIDIA K80 was the most powerful GPU when it appeared on the market in 2015. Since them a lot of new products have been made, but this one remains extermely fast despite being 6 years old. Essentiallly more than 4000 Cuda Cores and 24 Gigs of RAM. It is a great product for neural networlks and deep learning programming, but there are a few important facts for the ones who want to use it on a regular computer. 1. This is a GPU which has no display port. It has 24 gigs of video memory (5th gen) and runs on PCIe 3.0 but one cannot connect a screen to it. So the computer which runs it needs another GPU to connect the screen, meaning the motherboard has to support two GPUs, which is not always the case. 2. The motherboard has to support PCIe that can do above 4G Decoding, this is usually a configuration option on the PCIe or on the GPU part of the BIOS, typically in the advanced options. Without this switched on the computer will not even boot. This option is never set to on by default on a motherboard, so after putting this card on a machine, the machine will not boot and it is normal - find if the motherboard supports it before buying. 3. Cooling this GPU is a challenge. This GPU uses 300 Watt when used at max speed, but it will still use approx 70 Watt when not used, which means it will heat up to a level that will force a computer shut down if the cooling has not been set up properly. The cooling of a usual PC will not do anything because the card is enclosed so the cooling of the computer case is not enough. There are several Youtube videos on how to cool the K80, but the challenge is to make it practical: not too big to fit into the PC, but with enough airflow to evacuate 300 Watts of power. The air flow has to be directed to go through the card and exit the machine through the side where the display port is usally placed. 4. Cooling it down: my experience is that we need a fan which can move aprox 280 cubic meters per hour to fully cool it down - i.e. exploit its massive power at max speed for long periods of time. Otherwise the GPU willl self adjust to a lower speed or even turn off the PC if it gets too warm. It is not easy to create a device to push or pull the air thourhg it because of its narrow shape, but this is not impossible - be creative and get materials in a DIY shop. My own set up uses a Noctua 140 mm Industrial grade fan that can spin at 3000 rpm. 5. Cooling it down: there are alternative cooling devices that are supposed to be "universal" for any GPU. But this one is special, it is actually two GPUs in one, so the cooling requires a device that can actually cool down two GPUs in a very narrow space. I have not tried them but it seemed to me they would not work. 6. Cooling it down if you have money: I have not tried but the best way seems to be watercooling. There is one Chineese supplier of a waterbox designed for the K80 which still sell it on AliExpress. Other waterboxes may work as well, depending on their shape and cooling capacity. But this is expensive, it requires a waterbox, a pump, a reservoir, a radioator, etc. 7. It is actually two GPUs in one, each with 12 Gigs or RAM. Depending on application design it may not matter, but two devices with 12 Gigs of RAM is not exactly one with 24 Gigs. 8. Energy consumption: this draws 300 Watts, so be sure to have enough power 9. Space and weight: this is large (27 cm) and heavy (1.4 Kg) so only modern motherboards which are designed for the last generation NVIDIA or AMD GPUS will accept it - anyways only modern motherboards can do the aboce 4G decoding so check the motherboard specs before buying. 10. Noise can be an issue if not water cooled. In my set up the Noctua 140 mm industrial fan has keep running at 900 rpm to cool it down when not used - which is OK, sounds like a normal computer - but at full capacity the Noctua runs at 3000 rmp which is noisy. I have not tried the watercooling which should be better. Unplug the card if not used to avoid the energy consumption and the noise. 11. That's a very good piece of hardware, high quality from every angle, very stable computing power, 12. This was originally designed for purpose made hardware, not to be put on a private PC. It only works with windows 10 or windows server. Download the driver on the NVIDIA website and keep it in a safe place, this may go out of support one day. No Linux version as far as I know. But for me no problem on that side, I am running it on Windows 10 Pro - I guess it works on Windows 10 Home. Check on NVIDIA website to be sure. Read full review
Verified purchase: Yes | Condition: Pre-owned
It gets hot, so you'll need a lot of cooling, the drivers were rather complicated because I'm using a Quadro p400 for display, and so i needed to use cuda toolkit intead of normal drivers.
Verified purchase: Yes | Condition: Pre-owned
It starte to work inmediately. My system has a Nvidia 8g card. This is a great investment. A lot of fun comming my computational way!!
Verified purchase: Yes | Condition: Pre-owned
Great for simulations, intense math calculations and video rendering! Most successful image processing companies could use these cards.
Verified purchase: No
The pcie mounting brackets were all kinds of bent and the cards are definitely a little more beaten up than I anticipated. I am yet to power the cards on.
Verified purchase: Yes | Condition: Pre-owned
for video editing fast
Verified purchase: Yes | Condition: Pre-owned
Yippie
Verified purchase: Yes | Condition: Pre-owned
Current slide {CURRENT_SLIDE} of {TOTAL_SLIDES}- Best Selling in Graphics/Video Cards
Current slide {CURRENT_SLIDE} of {TOTAL_SLIDES}- Save on Graphics/Video Cards
Current slide {CURRENT_SLIDE} of {TOTAL_SLIDES}- You may also like
