So i managed to reserve 4x Vega VII for 750$ each.
I need compute power for Tensorflow, Keras, Pytorch etc
The reason i got them is because of the 16GB ram each and because they are based on MI50 compute cards.
Im currently running a RTX 2080super but im running out of memory so i have to decrease the batch and IBM large memory support is not that fast for me. The idea would be to run the four Vega VII and hopefully combine the 4x16GB Ram to a total of 64GB ram.
I haven’t picked up the Vega VIIs yet so i cant verify it but from what i read the RTX 2080s would be faster at FP16 right but Vega does outperform the RTX 2080 in FP32 by x5?
Could you guys please test the RTX 3080 in Compute like Tensorflow, Keras, Pytorch etc and put it against a RTX 2080 or 2080s and Vega VII? I can find only gaming benchmarks and im really curious.
Because i have 3000$ to spend and im not sure if i should pick up the 4x Vega VII or perhaps go for 4x rtx 3080 since they cost the same or perhaps 8x 3070 since that would also fit my budget and add up to 64GB of ram. I was thinking to get a thread ripper board with enough PCIe 4.0 lanes because from what i understand PCIe lanes are basically double the number of PCIe 3.0, so would it be possible to use PCIe 3.0 Raiser hub cards to install a total of 8 RTX 2070 at full pcie 3 x16 speeds?
How would my 3000$ GPU budget be spend the best if what i need the most is compute power in tensorflow, keras and pytorch? (most coded in python).
Hope my questions make sense and im sorry for my english.
Many thanks!