charlesGodman

joined 1 year ago
[–] charlesGodman@alien.top 1 points 11 months ago

Thanks. From what I read, on a system with such a powerful cpu, it’s advisable to use software raid over hardware raid cards

[–] charlesGodman@alien.top 1 points 11 months ago (2 children)

I didn’t realise there were Epyc cpus that are actually cheaper than threadripper. small epyc 7003 are the clear choice. Do you know any epyc boards that are made for storage systems?

[–] charlesGodman@alien.top 1 points 11 months ago

I dont have input on your question. As someone owning a Samsung 870 QVO I can strongly recommend not buying it, though. Get EVO or something else. The sustained write on my QVO is abysmal.

 

I was thinking the Ryzen Threadripper 7945WX (or equivalent 5000WX) Pro (smallest, 12 core) might make a great platform for a data server. The motherboards don’t have many u2 connections though (only one slimSAS). With 128 PCIe lanes I should be able to drive around 20 x4 NVMe drives. But how do I connect all the drives to the board? Are there dedicated u.2 expansion cards? Do I use 4x m2 x16 cards and m2 to u2 cables? I am interested in connecting it eg with some 100Gb/s connection to another server for deep learning.

[–] charlesGodman@alien.top 1 points 11 months ago

To be precise: linear regression with L2 loss and L1 regularization. L1 regression with L1 penalty can be solved using a linear program.

[–] charlesGodman@alien.top 1 points 1 year ago

There has been LOADS of research on deep learning optimisation in recent years. However, TLDR nothing beats ADAM.