After 15 years I build myself another PC for practicing Machine Learning/Deep Learning. It looks OK but pretty disappointed in terms of option when it comes to PC Case. Before this machine I was using a 9 year old HP z600 dual E5520, 24 GB Workstation and what a machine that was. I never had to touch any tool to replace any PC component. The scalebilty and build quality of HP Z600 is still unparalleled and my current build unfortunately cannot come close to the ruggedness that machine offered. Why did I buy this machine? Price. Any workstation from HP, Dell , Lenovo with an 7940x starts from 4000 $. See for yourself, an HP workstation with similar specs cost f.... 8700 $
I tried to find a good solid case below 150$ and could not find even one. NZXT H700 was something close to the quality I was looking for. I like its minimalist looks and the air flow. The case routing and fans are also good, but this machine is not built for workstation. I was very used to internal 3.5 hot swap tray of HP and I could not find even one internal 3.5 hot bay. Most of the internal hot bays were kind of an adapters for 2.5 hdd. Therefore had to come up with quickfix solution. I inserted an external 5.25 hotswap bay from Silverstone and it kind of worked but not as smoothly as I like it to be. The tempered glass window is joint by four thumb screws and its a big hassle if you continuously wants to swap things out of case. The other side back window is another story. It is the best feature of this case as you can dislocate that window by just pressing a button behind the case. This was another reason that I placed my hotswap in that window so that swapping of 3.5 HDD remains an easy task.
Also do not understand why there is such a shortage of core i9 7940x. Luckily I got it for below 1000 $ and that was the main intention to build this PC. Now coming to the actual use, I like how the machine works. I am using lubuntu and the PC just flies in most of the task. The main purpose of this machine was MAchine LEarning and I see a jump in performance. The only question is I bought my earlier PC, HP Z600 refurbished at peanuts and for its cost that machine was terrific. I learnt most of the MAchine Learning task on that machine. Its just that when my dataset and features started to become larger that machine was taking inexplicable amount of time. The gridsearch, K-Folds for larger datasets for proving to be a strong challenge and thats when I thought of building this machine.
Anyone on the similar boat who is trying to build a PC for Machine Learning/Deep Learning should buy this kind of CPU if he or she can afford. I also looked upon threadripper 1950x and 2950x but both were lacking when it comes to ML algorithms. 1950x cannot keep up with 7900x and 2950x cannot keep up with 7940x. It seems lot of ML algos. and Linux kernel is still optimized for Intel. Also, you would find lot comments on some blogs saying "CPU does not matter in Deep Learning" Avoid those. CPU does matter. You would certainly do things which cannot be migrated to GPU and for that you would need a strong CPU.