Tesla K80

[copey96]

ebay has some cheap deals on Tesla k80s. I knoiw there not exactly powerful but i want to use them with my dell poweredge r200.

 

so the poweredge r200 is pcie 8 lane. not sure what pcie version and the tsla k80 is pcie 3 16 lane. is it going to see a major performance decrease running something like comfy ui and deepseek r1 llm?

[YoungBlade]

I'd be more concerned about how the K80 lacks Tensor cores. Those weren't added until Volta and Turing. It has Cuda capabilities, but that's still not ideal for AI tasks, and Kepler's Cuda performance was weak compared to later generations.

 

Basically, while I haven't seen any testing with Kepler for modern AI stuff, my hunch is that the GPU core itself is going to be a bottleneck before the PCIe bus is.

[copey96]

Hmm I see what you mean. im guessing the option of upgrading my gaming rig with a new GPU and using my current rtx 5070 is probably a better option. No sure if this would start hitting bus issues though?

 

[Needfuldoer]

That card has a good amount of VRAM, but that's pretty much its only virtue. It's a decade old, supports a relatively early version of CUDA, and lacks tensor cores. Your 5070 is orders of magnitude faster.

 

The R200 itself is also a very, very old platform. It's Core 2 era stuff. It's low power (for its time) and only a 1U chassis, so a dual slot, power-hungry card may not even fit and get enough power.

 

You might run into trouble running software that requires CPU instructions it doesn't have. (I tried running Topaz Labs software on an Ivy Bridge era machine and it failed to run.) You may want a newer platform to play with, even if it's a Haswell/Broadwell era machine like a Dell Precision 5810 or HP Z440.