Description

Being abroad as a freelancer I needed a somewhat powerful PC to help me prototype and test things and working for some time out of a 7700hq/1050 ti laptop (still decent as a temporary solution, but becomes tiresome after awhile), I ended up with this setup. I started looking for components at the end of July 2018. The build took almost 6 months to complete and there are a couple of changes/upgrades pending, thus some component choises might seem a bit odd today. The main purpose of the build is software development, machine learning and of course some gaiming on the side.

CPU At the time I started looking, I was between the 2700x (€310) and the 8700k(€360). Pros for 2700x: Two more cores (four threads), lower MSRP, comes with a decent cooler, better performance at full load and overall more future proof as software continues to adapt to more core/threads. Pros for 8700k Faster single clock speeds, AVX 256, lower/better latency (cache/ram), noticeably faster for my workloads (but not significantly), comes with integrated graphics (GPU was ordered seperately and arrived a month later). From a financial stand point, the 2700x is the clear winner when it comes to new, but before buying one, I decided to see how the 8700k did in the second hand market. So I placed a couple bids on ebay. Lo and behold I managed to win one for €245. I was surprised by the price and at the time I thought that it would be a bad overclocker (no delid), which I did not care that much. Currently I am able to do 4.9Ghz (0 AVΧ) with reasonable voltage/temperatures (adaptive -80mv, level 3 LLC asrock , cinebench r15 70c (16c ambient))

Cooler Went for the cheapest available 240 AIO. Tricky to install in ITX case but does a decent job keeping the CPU cool. I chose to install it vertically in the case as an intake, to help improve air pulling. Seems to work.

Motherboard I wanted an itx system that it would be easier to move and take with me if necessery. Originally I went with the Gigabyte z370n which was the cheapest one, but I managed to brick it. On the first installation at stock settings the temperatures were through the roof. Not only the CPU (+40c idle, cinebench 85c, 25 ambient), but also the motheboard itself was boiling. HWMonitor would report +80c on TMPIN5 at idle and on load it would go above 110c and continue climbing until I stopped what I was doing and let the thing cool down a bit. I tried to reapply the thermal paste and remount the cooler when I did a big mistake. I reinstalled the CPU on a vertical position and the system would not boot if RAM slot 1 was occupied. Yes I bent a few pins. Then I tried to straigthen them with twizzers, which completely destroyed the mohtherboard. Buying the same MOBO again was out of the question thus I ended up with the Asrock z370 itx (the second cheapest option €20). The difference between the two MOBOs is night and day. Absolutely no thermal issues and a decent overclocker (a lot higher than I expected). I managed to do 4.9 fully stable within a day. I gave it a shot at 5.0, but I could not get it fully stable at -20mv and the temperatures where starting to go beyond my comfort zone, so I gave up. 100Mhz do not worth it.

RAM I knew I needed 32GB (16GB is OKish, with a lot of paging going on). I checked the PCPartpicker and found a G.Skill set with decent speed, but somewhat loose timings at a reasonable price. Samsung B dies were at least €100 more expensive, which is way too much for any gains they might offer. I managed to overclock it at 3600Mhz(17-19-39) at 1.35v, which is good enough.

SSD It was the time that SSD prices started to go down and managed to find the Samsung 970 EVO 500gb at €120. Excellent product and outstanding performance. PC boot and program loading is almost instantaneous. I paired it with a Samsung T5 and moving files around is blazing fast.

GPU Finding a GPU was tough. For sure I knew I wanted a RTX card due to the tensor cores. No I do not care about raytracing. I am mostly interested in the AI capabilities they offer. At the time of my purchase only the 2080 and 2080 Ti were available but expensive and out of my budget. I end up getting an EVGA GTX 1070 to get me started. EVGA is the only manufucturer that offers the option to stepup. I have chozen to upgrade to a 2070. The 1070 did the job very well. Looking forward though to see what improvements the 2070 brings. According to some initials tests it is 60% faster at 16bit training. I will update this section, when I play more with it.

Case Went with the classic Phanteks Enthoo ITX Mini case. Good build quality at relatively affordable price. A bit dated by todays standards. No front USB type C. Restrictive front panel air flow. Limited side panel cable management.

PSU SeaSonic is considered an industry leader and for a good reason. Excellent quality at very good prices. It works silently and flawlessly.

Fans The case comes with a 200mm front fan and I have added 3 be quiet!. One on the back as an exhaust and two on top as intake. Currently on the way are two InWin Mars fans

Monitor I have multiple programs open and more screen real estate is very useful. I went with two Samsung 4k monitors that were on sale. Decent quality for the price.

Benchmark https://www.userbenchmark.com/UserRun/13891129

Part Reviews

CPU

Probably I would not pay MSRP to buy it right now. Currently there are better options, but if you can find one in the second hand market at lower price, it worths considering. Good overclocker, very snappy and can handle anything you throw at it.

CPU Cooler

For its price, it is ok. My biggest complain is that it is too complicated to assemble and difficult to install it in the socket. You have to adjust as you go and hold it as you screw it in place.

Motherboard

Good CPU and RAM overclocker with multi phase power delivery. VRMs stay cool under load. No RGB fluff, but who cares

Memory

Works at advertised XMP setting and event goes higher. 5 stars for the price,

Storage

Absolutely amazing. Super fast and reliable

Video Card

I have just received the RTX 2070 from EVGA as a stepup from the GTX 1060. The performance difference in machine learning training is very significant (60%)

Case

Good case, but dated. No type C USB, restrictive front airlow and limited side panel cable management.

Power Supply

Excellent. Good for overclocking. No stability issues under full load.

Monitor

For the price a really good choise for anyone who wants 4k

External Storage

Nice drive and very useful. You can literally work out this drive, without noticing any delays or that it is an external drive.

Comments

  • 10 months ago
  • 2 points

I hate it when it is 4.9 or 3.9 GHz. It's awful.

  • 10 months ago
  • 1 point

yes. So true, so true

  • 10 months ago
  • 2 points

Well-reasoned, thoughtful build. I get a kick out of hearing how people use their computers for things other than gaming. May I ask how you are planning on making use of the tensor cores? Is it that certain programs interact with that part of the video card driver to process things?

  • 10 months ago
  • 2 points

I am using tensorflow and keras for machine learning. You can configure those apis to explicitly execute parts of your program with tensor cores for some significant speed up.