• Log In
  • Register


Viserion's Deep Dive

by AfzalivE



Date Published

Oct. 16, 2017

Date Built

Oct. 9, 2017

CPU Clock Rate

3 GHz

CPU Temperature While Idle

32.0° C

CPU Temperature Under Load

54.0° C

GPU Core Clock Rate

1.594 GHz

GPU Effective Memory Clock Rate

8.008 GHz

GPU Temperature While Idle

37.0° C

GPU Temperature Under Load

67.0° C


Note: All of the prices are in Canadian Dollars, after tax and shipping.

I'm taking a Machine/Deep Learning course and decided that eventually, I'd need a more powerful machine than my 5 year old Macbook Pro for it.

Had the Corsair cooler and the NZXT S340 Elite Black/Blue from a previously planned build, but decided to sell the case and switch to Phanteks P400S Tempered Glass, white because it looks like snow!

Absolutely love the lighting in this case as well as the white non-glossy body. And because of the water cooler, it runs suuuuuuper quiet. Quieter than my laptop! Called it Viserion because it's white and blue, and it's a beast! (Well in my eyes anyway)

Had to make a sacrifice in the CPU budget to get a motherboard that supported SLI for future upgrades. Got my hands on the cheapest 16GB kit available at the moment, sadly it doesn't go that well with the lighting.

Got a B-Stock GTX 1070 from EVGA. It's been running without issues. It's a better bang for the buck than the GTX 1080 for DL and even better at this price! With the added benefit of being able to mine some Eth if I want to. The 1080 is notorious for being horrible with that.

Got a 750W PSU for the same reason as that motherboard, future SLI compatibility.

It's been working well for one week now. Running Ubuntu 16.04 LTS on it, absolutely no gaming. Haven't done any intense training yet. That is still to come in a few weeks. Ran a few benchmarks to see what the temps looked like.

Part Reviews


It's alright. Didn't need a very powerful processor for pre-processing data, and most of the libraries are optimized for single-threaded performance, so this works well in the price range.

CPU Cooler

Had it for a 6 core Xeon computer I was going to build but ended up not building that. Works pretty well. Keeps the CPU cool. The only thing I would say is that the pipes are too short and so are the fan wires. They barely reached the connections on the nearest side of the motherboard.


This motherboard gave me a headache. I spent a decent chunk of money trying to update the BIOS without a Skylake processor at hand. At this point, this should come updated from the factory! Cost me enough in the end that I could've bought a Gigabyte Z270 SLI board (with a better audio chip) with that money!


Snatched it on sale. Nothing to say other than the fact that all the sticks work.


Had this in my Macbook for the last year or two, upgraded that to 512GB so kept it for this system. It's been serving me well.


Got two of these. One for the gaming PC and one for this PC. Seems alright.

Video Card

Got this B-Stock at EVGA's website. While the other parts were arriving, I had it in the gaming PC and it breezed through everything Max at 1080p. Forza Horizon 3, Assassin's Creed: Black Flag, NBA 2k18. There was a stutter issue (even with Rocket League) but a driver reinstall fixed that on Windows. Temps are pretty low under stress (compared to my RX 470) and so is the noise. Idle temps are higher because the fans don't turn on unless there's load, and I won't connect the monitor to it so there's essentially no load until some CUDA computation is going on.

We'll see how the ML/DL performance is but from what I've read, this is great bang for the buck!


I had the S340 Elite Black/Blue from the Xeon build that never happened. Sold that and bought this as soon as I saw this. I love, love, LOVE this case! The magnetic RGB strip, the grommets, the velcro cable management ties at the back, the intuitive hard drive bays (with dampeners), the magnetic fan filters (AND covers if you want). This is how a case should be!

Power Supply

I've been using this guy's Bronze brother for the last 3+ years now in my gaming PC. The ML rig required a bit more dependability since it's gonna run for longer everyday. Got 750W for when I inevitably end up buying another GTX 1070 for SLI. Got this on MIR, which is a hassle but EVGA's MIR is quite good. I had an idea while doing cable management, maybe these guys should bundle in cable combs with PSUs. That would be really neat!

Comments Sorted by:

Deeplearningmoney 2 points 17 months ago


How do you like your rig so far? I am new to DL/ML and building a rig is my goal for the next 10 days.....

AfzalivE submitter 2 Builds 1 point 17 months ago

Hey, I like it so far. Gonna do CNN with images today. Did an ANN run yesterday with not so big of a dataset (10,000 rows, 6 nodes, 11 features, 100 epochs, 100 batch size), took like a minute or so. Example: Numer AI's dataset is around 500,000 rows and 50 features.

I'm using Hamachi to log into it from anywhere and access R Studio or the Spyder IPython kernel and I really like that all that power is available from everywhere. Gonna write a blog post about how to set this up soon (especially for securing it and for Spyder).

The only thing I don't like is that there's no way of automating a fan curve on linux unless you run a script all the time, and by default, this EVGA card's fans don't kick in until 50+ C.

Your GTX 1070 Mini might generate a quite a bit of heat too, and your Xeon E3-1220 only supports PCIe 2.0. I would say give Ryzen a try if you can, you definitely have a nice enough RAM kit to try it.

Bodom0511 5 Builds 1 point 17 months ago

Nice looking build, but may I ask why you chose to water cool a locked processor? For the price you paid you should have certainly gotten an unlocked cpu and an aircooler.

AfzalivE submitter 2 Builds 1 point 17 months ago


Haha, fair question. Tbh, the cooler was for a Xeon X5650 CEH build I was trying to do in April. Sadly, I gave up on finding a motherboard in my budget for it, and then realized that the processor was kinda out of date for my needs (PCIe 3.0+). And then I promptly lost the processor somehow :/

So basically, I had the cooler way before I had the processor, lol.

mrkurtberry 1 point 16 months ago

Did you add that backplate to the GPU? Where did you get it? I been looking for something like that for my EVGA 1070 SC Black Editions

AfzalivE submitter 2 Builds 1 point 16 months ago

I didn't. That's the EVGA 1070 SC Gaming


tuninglegend 2 Builds 1 point 15 months ago

Is it normal if my top covers are a bit yellow-ish, instead of the same lucid white that the entire case has?

AfzalivE submitter 2 Builds 1 point 8 months ago

I don't think mine are. Depends on how okay you're with it I guess

[comment deleted]
[comment deleted]
AfzalivE submitter 2 Builds 1 point 17 months ago

So...All these are canadian prices (after tax and shipping). Lol.

I had a Ryzen part list going side-by-side. In the end, the prices were just not comparable. Closest thing in the C$250 range was the R5 1500X and then I'd end up spending about $50 more for the different motherboard. It probably would've been better and cheaper to get that in the long run, especially when I decide to upgrade the CPU.

Also read that Ryzen, right now, is very memory-hungry. Also, with a Ryzen, I'd have to power the display through the GPU. Not something I wanted to do when I want to use the GPU purely for computation. Right now, I can use the GPU for computing while powering the display with the integrated gpu.


That's not $500 even before tax, let alone AFTER tax and shipping. That's a $100 difference.

Gotta disagree with you on the EVGA PSU though. I think both of those are great in their own right and you can't pick one over the other. Again, big price difference, it's $198 on Newegg Canada right now before tax. That's a $90 difference after tax.

I will concede the CPU part though, but it was too late. I ordered the motherboard through Vuugo and it would've been a major hassle to return it sadly. Accounting for the stupid BIOS upgrade that I had to get done, yes, spending another $100 and getting the AMD combo would've been WAY better.

Lessons learned.

[comment deleted]
AfzalivE submitter 2 Builds 1 point 17 months ago

Haha, yeah they do. I wish they weren't so different. I'll keep an eye out, though this setup will still last for a while until I feel that it's not enough to crunch datasets fast enough. :)

By then, many Python pre-processing libraries will be way better at multi-threading than they are right now.

Lobsterareawesome 8 Builds 0 points 17 months ago

and no.. the 1600X does not surpass the 7700k in gaming

[comment deleted]
[comment deleted by staff]
manirelli staff 8 Builds 1 point 17 months ago

Please keep feedback polite and constructive.

AfzalivE submitter 2 Builds 1 point 17 months ago

Wow. Did you delete all the good conversation in the other thread too?

[comment deleted by staff]
[comment deleted by staff]
[comment deleted by staff]
manirelli staff 8 Builds 1 point 17 months ago

Please keep feedback polite and constructive.

[comment deleted by staff]
[comment deleted by staff]
[comment deleted by staff]
[comment deleted by staff]
[comment deleted by staff]
[comment deleted by staff]
AfzalivE submitter 2 Builds 1 point 17 months ago

I'm a programmer, make Android apps. But this course is pretty comprehensive. Starts from basic regression and classification lectures, teaching the "intuition" and pros and cons for each strategy, then going on to implementing them in Python and R. And then also guiding the student on making a template for later real-world use.

Only just starting the reinforcement learning section, then NLP, dimensionality reduction, and then neural nets. My issue was that I needed a good resource to understand what these layers and activation functions etc did with the data and how to experiment with them.

[comment deleted by staff]
[comment deleted]
[comment deleted by staff]
[comment deleted]
[comment deleted by staff]
AfzalivE submitter 2 Builds 1 point 17 months ago

Thanks! It was a Canon T2i, and then some post in Lightroom :)