- Jan 8, 2019
- 2
- 0
- 510
Hi!
I am biding a ~7k$ workstation for data processing (large microscopy recordings - 3d arrays 10s of GB that require various sorts of filtering and other computations) and simple simulations (some PDEs), rarely machine learning. Fair amount of matlab, python and C/C++ code, large amounts of data going through
I am in between of committing to the amd threadripper 3970x or it's server counterpart epyc 7502. Those two seem almost identical (same core counts, same cashes), but threadripper has a much higher base clock (3.7-4.5 vs 2.5-3.35 Ghz), and epyc version allows for much more RAM (2TB vs 256 GB) with twice more channels (twice more bandwidth ) and also has more PCI lanes. If we are considering newegg factor-installed EPYC variants, they are even priced the almost the same (when the motherboard price is taken into account).
While it would be super expensive to have >=1TB of ram in 8 slots, it would often be convenient to have like 512 GB for some large processing. Larger bandwidth also seems quite a benefit in that case (I don't really know how much of a bottleneck it is right now), while extra PCI lines can be useful for adding extra GPUs. However, 30-50% more clock speed also looks huge.
There aren't many face-to-face comparisons of the two chips. I was able to find a namd benchmark (epyc version ~1.5 times slower). The linpac flops performance of 3970x is somehow almost 2x lower than the theoretical(?) 7502 performance, but not sure that means anything.
I would appreciate any advice!
I am biding a ~7k$ workstation for data processing (large microscopy recordings - 3d arrays 10s of GB that require various sorts of filtering and other computations) and simple simulations (some PDEs), rarely machine learning. Fair amount of matlab, python and C/C++ code, large amounts of data going through
I am in between of committing to the amd threadripper 3970x or it's server counterpart epyc 7502. Those two seem almost identical (same core counts, same cashes), but threadripper has a much higher base clock (3.7-4.5 vs 2.5-3.35 Ghz), and epyc version allows for much more RAM (2TB vs 256 GB) with twice more channels (twice more bandwidth ) and also has more PCI lanes. If we are considering newegg factor-installed EPYC variants, they are even priced the almost the same (when the motherboard price is taken into account).
While it would be super expensive to have >=1TB of ram in 8 slots, it would often be convenient to have like 512 GB for some large processing. Larger bandwidth also seems quite a benefit in that case (I don't really know how much of a bottleneck it is right now), while extra PCI lines can be useful for adding extra GPUs. However, 30-50% more clock speed also looks huge.
There aren't many face-to-face comparisons of the two chips. I was able to find a namd benchmark (epyc version ~1.5 times slower). The linpac flops performance of 3970x is somehow almost 2x lower than the theoretical(?) 7502 performance, but not sure that means anything.
I would appreciate any advice!
Facebook
Twitter
Instagram