Raevyn

Hello, So I have been viewing various youtubes, including Linus obviously lol, on making my own NAS at home. It seems fairly simple but the one thing that is a bit new to me is ZFS. I keep reading it is resource intensive and needs decent hardware, but I am having a hard time determining what that actually means in terms of appropriate hardware. I am definitely wanting to go AMD. I was thinking of a Ryzen 3900x but I am worried that is going to be a complete overkill. Would Ryzen 7 3800X or even 3700X perform the same? All of this would be going on an ASUS microATX board with the max memory it can support, 64gb of ECC.

This was the board I was thinking of getting.. it holds 4 cards sorry I dont know why I was saying 6 lol https://www.supermicro.com/products/motherboard/Xeon/C620/X11DPG-QT.cfm I cant currently find the model that had that server rack riser that holds so many cards but that does make me ask another question: I see supermicro makes another board that would hold a lot more 8x cards if single slot. Since most GPUs are double slot would it make sense getting risers for those slots to be able to fill those up be a good idea for deep learning? Here it is: https://www.supermicro.com/products/motherboard/Xeon/C620/X11DPX-T.cfm

Okay that makes sense.. AI needs the high speed communication. Well then why do companies like Supermicro sell boards that can allow up to 10 cards when the processors cannot reach the full bandwidth that would provide? Wouldnt it be better, for learning and AI, to stay within what the CPU lane limit is?

Hello all! So I saw a video Linus and Co made about crypto-mining on an Asus board that had I think 19 PCIx1 slots and I am trying to understand something about that. In the video he hooked up some number over the traditional 4 or so a normal board will hold by using riser cards which allowed him to plug in 16x cards into the 1x slots. I know this sounds dumb but wouldnt that result is a huge performance bottleneck since obviously it is cutting down the bandwidth available to the GPUs for processing? I am wanting to build a system not for mining, but for learning and I was originally thinking of a Supermicro board that can hold 6 GPUs, although this question I had a while back when I saw a Supermicro board that had a riser board to hold something like 10ish cards in the same rig. The processors just dont support that many PCI lanes so the same issue seems to come up to me, bandwidth limits. Am I missing something in this? Because honestly assuming I can get that Asus board, it is priced several hundred cheaper than the Supermicro I was looking at and if I will get the same or even better performance for my application, it seems the better way to go though I would have to build a custom rig which is my plan anyway.