A New Computer


I wanted to write this step-by-step guide for myself in the event that I wanted to upgrade my current computer build. Since it takes an eternity to pick the best possible components and make sure they’ve even compatible, I thought I might as well go through the mental exercise, if for no other reason than to learn the nuts and bolts of what goes in a computer.

And hopefully others might find it useful and informative as well.

Note that things change over time, and the purpose of this post is to help you (or mainly myself) figure out how to pick the parts for a computer, and now what to put in your computer. Or, just one with an Intel CPU and Nvidia GPU in this case.


Cases exist in a variety of form factors defined by the sizes of motherboards they can contain

Motherboard Case
Mini-ITX SFF (small form factor)
Micro-ATX Mini tower
ATX Mid tower
E-ATX Full tower
Comparison of motherboards
Photo of motherboard form factors from VIA Gallery in Hsintien, Taiwan.

A full-tower case can contain both an E-ATX, ATX, microATX, and mini-ITX motherboard, while the SFF case can only contain the mini-ITX. In other words, the smaller the case, the more limited your motherboard options are.

You’re probably not going to run into an E-ATX motherboard, so the main advantage of the full-tower form factor is the clearance and room for additional components like HDDs and additional cooling. Optical drives aren’t a hot commodity nowadays.

When performance is important, you’ll want some space for airflow, referred to as “clearance”. This rules out smaller form factors like mini-ITX.

Airflow is intertwined with heat dissipation, which in turn is intertwined with noise as your computer works overtime to cool all components.

If airflow, heat and noise—or cable management and dust—aren’t major concerns for you, and you don’t have a lot of components that will go in your computer, I would check out Fractal Design’s Node 202, which is smaller than a videogame console:

The Node 202 case
Node 202 case by Fractal Design.

Another thing to consider is whether you graphics card can even fit inside your case. More on that in the GFX section.

As for me, I’ll be going with a silent performance computer, so I’m looking for a mid or full tower.

  • [ ] mini-ITX - SFF (small form factor)
  • [ ] microATX - Mini tower
  • [x] ATX - Mid tower
  • [x] E-ATX - Full tower

Some other things to consider:

  • 140mm fan support
  • Quality default fans—you’ll rarely end up replacing them
  • How many fans are supported: top, buttom, front, side?
    • Some cases have nominal support for fans which ends up not working under typical circumstances
      • beQuiet’s Dark Base 900’s one bottom fan is blocked by most PSUs
  • Filters for all fans and air intake
    • How easy are they to clean—how much time and effort to extract?
  • Are parts like screws easily replaceable if misplaced?
    • (Lian Li’s sure aren’t)
  • If the case has a front door panel, is it a piece of crap?
    • Do the front fans have any intake with the door panel closed?
  • How easy is it to carry?
    • You’ll have to lift and carry it at some points regardless of what you tell yourself
  • How easy is it to mount and install components—and maintain them?
  • How much of an eyesore are the front ports for headphones and USB?
  • How easy are the ports to reach—will cables have to run all the way from the back?
  • Good cable management
    • Rubber grommets for cable funnelling
      • Brackets for velcro ties to the side of the grommets might be better for cable management
    • Cable tie points on the back
    • PSU shroud or bottom funnel
      • At least offered on their website
      • Ideally with mounts for SSDs
      • Holes and gaps for airflow might be preferable
      • Grommet on top for PCI-e cables for the GPU(s) aren’t bad either unless you want to use the side’s grommets
    • Grommets at the very top of the motherboard for fan cables
  • Sound dampening foam on side panels
    • Good for noise reduction, might bring up the temperature somewhat
  • Are the HDD bays removeable to free up space and airflow?
  • 2+ SSD mount(s) in such places as
    • behind the motherboard plate
    • on top of the PSU shroud
    • elsewhere with SSD brackets

While a smaller tower would be nice, the best all-round case is the beQuiet! Dark Base 900:

The beQuiet! case
Dark Base 900 case by beQuiet!.

It’s also available in a “Pro” version with a transparent side and wireless Qi charging.

<Update 03/26 ‘17> Apparently, another difference is the lack of automatic fan control. What the fuck, beQuiet. If you want it built in, get another case. From what I can tell by the manual, the Pro version also has an additional 4- and 3-pin connector (4 of each vs 3 of each in the non-Pro).

The circuit board for the Dark Base 900 fan controller.
You can see the outline of all the stripped components for the fan controllers—like chalk lines for a murder victim.

My former case was a Lian Li that wasn’t perfect, but it did have advantageous feature of thumb screws that could be turned with your fingers without the use of screwdrivers. I miss that when I deal with my Dark Base 900.

<Update 05/23 ‘17> After it turned out the case’s door wasn’t working, I fucking finally received a replacement, about two months after I initially reported the issue. I’m pretty sure my seller just sent me a replacement, because beQuiet were taking so long to address the issue, since I got an order confirmation soon after I shot them an e-mail asking them what the hold-up was.

A few other things:

  • The manual is also extremely cheaply done, and it turns out there’s a separate section in the back with assembly instructions past all the different translations.
  • The instructions themselves are also pretty terrible.
  • The first case I received was also very banged up with scratches on screws and drive bays, and the replacement had a loose PSU mount foot.
  • On that note, the screws appear to be of the same cost-cutting quality as the manual; they scratch if you as much as look at them, and turn silver after their first use.

    I come from a Lian Li so maybe my expectations are too high, but this is embarrassing for a high-end case.

  • Mounting the PSU is a weird and abstruse requirement, and mounting it only reduces movement on one side and axis. Pretty sure I also scratched my PSU a fair bit installing it

    As the Germans in this PSU installation video I had to hit up described it: “eine kleine Besonderheit”.

    You can also hear the Hardware Canucks poking fun at the unsecured PSU in a video.

  • The bottom fan mounts are practically unusable because the PSU covers the first one, and PSU cables will obstruct the second one.
  • If you install a third front fan at the top, it won’t have a fan filter.
  • The case supports two top fans, but there’s only three infinitesimal slits above the one at the farthest back for exhaust
    • Besides, without the two bottom fans, you’ll end up with negative or neutral pressure by adding two top fans (2or3&timesF-(R+2×T))
  • No apparent way to buy another SSD bracket in place of their garbage fan controller either
    • At least you can install your SSD on a PSU sh—
    • Update: I was able to buy an SSD bracket after finding a way to contact them
  • No PSU shroud—and their Dark Base 900 Pro shroud doesn’t support the non-Pro, of course
  • No top grommets for EPS and fan cables
  • Contacting beQuiet about anything is a miserable experience

Put another way, you’ll have to find and buy these yourself:

  • Fan controller (or use your motherboard’s)
  • SSD bracket
  • PSU shroud
  • Front fan filter

I don’t see myself ever getting anything from beQuiet again. </End of update>

One thing to note is that the front door panel of the Dark Base 900, and the Pro version, doesn’t close well (3:13), which means you manually have to hold it in when carrying the case and such. Hopefully this is addressed in a future revision.

<Update 06/06 ‘17> For some reason, you the middle screw for the HDD bays are in the way of SATA cables … What?</End of update>

<Update 05/21 ‘18> I think the front door issue is solved by triple checking that the front panel is secured. I managed to fix an issue that arose after I’d taken off the panel to install a new fan in the first place.

Noise versus airflow

In their Dark Base 900 Pro review Tom’s Hardware note the distinction between noise and temperature management:

The Dark Base Pro 900 was quieter than the other cases, but was hotter at the same time. Therefore, [its] efficiency is lower due to the reduced airflow not effectively cooling the components.

Tom’s Hardware’s Dark Base 900 Pro review

In other words, you may have to choose between the two—or go with watercooling. I prefer the noise reduction and don’t plan on overclocking, so this suits me fine.

Here are bit-tech’s take on the matter:

The exterior does raise concerns over cooling and airflow, though, as there’s a lack of vents, especially in the roof, and any air being drawn in at the front or expelled at the rear will be hampered by having to bend at 90 degrees through the side vents. That said, this is done deliberately by Be Quiet! to deflect noise caused by airflow, so we’ll have to see how well the case cools before judging too harshly. It’s perhaps a little more porous than some of Phanteks’ efforts too, many of which beg for better vents, especially in the roof, but you can’t have a super-sleek design and have top-notch, unrestricted ventilation.

bit-tech’s Dark Base 900 Pro review


  • Performance
  • Temperature
  • Noise

Monitor settings

Before you decide on your computer specs, you need to figure out how demanding a task you are throwing at them. The foremost thing to consider are your monitor settings, mainly:

  1. Resolution
  2. Update rate

If you are like most people, you are, and will be, running your games at 1920×1080 at 60 Hz (and frames per second), ie 1080p60.

These days, running games at 1440p resolution (2560×1440) is quite popular. That’s a whole lot larger than 1920×1080. How much larger?

Resolution Pixels %
1920×1080 2,1M 100%
2560×1440 3,7M 178%

On top of this, people who play first-person-shooters in particular have taken to playing their games at 144 Hz (on monitors that support it), which prefers 144 frames per second—compared to the usual 60, a difference of 240%.

To bring it all together:

Resolution Pixels/s %
1080p60 124M 100%
1440p60 221M 178%
1080p144 299M 240%
1440p144 531M 427%

Bit of a difference, no?

Keep in mind that the 144 frames per second are mainly important for first-person shooters where the difference is significant—and desktop work where the idle load is low anyway.

If you have a *Sync monitor, running below the update rate of your monitor won’t hit you as hard as it would otherwise. I’m not going to get into the weeds of explaining (VESA) Adaptive-Sync, FreeSync, and *G-Sync, because I don’t plan on buying a monitor right now; *Sync monitors are wicked expensive right now and not even available with IPS panels.

If you’ll just be going with your usual 1080p60 setup, maybe you don’t need a beast of a GPU to shoulder your tasks.

Since I do plan on getting an IPS 1440p144 *Sync monitor eventually, I’ll be planning my computer accordingly.

  • [ ] 1080p60
  • [ ] 1440p60
  • [ ] 1080p144
  • [x] 1440p144

RIP my wallet.

NB: Make sure your monitor and video port can even handle this, especially if you add some kind of HDR. “Video Terms” has a helpful break-down of this.


There are two cooler types for graphics cards:

  1. Blower design
    • Exhausts air out the case
      • Bad when airflow available (big cases)
    • One fan
    • Loud, hot
    • Easier to mount custom cooler
  2. Turbine design
    • Exhausts air into the case
      • Bad when airflow unavailable (small cases, SLI)
    • Multiple fans depending on size
    • Quiet, cool
    • Harder to mount custom cooler

The turbine design is overwhelmingly preferable for most cases, and more important, for this one.

  • [ ] Blower cooler
  • [x] Turbine cooler

Graphics card

Next, we’ll have to have to narrow down the candidates for graphics cards.

Nvidia look like the solid winners of the current generation of graphics card with their new Pascal microarchitecture. “Microachitecture” is also colloquially referred to as “µarch” or “uarch”.

The Pascal microarchitecture is currently represented by four “GP” chip designs; “P” stands for “Pascal”:

  • GP100: Nvidia Tesla P100 GPU accelerator is targeted at GPGPU applications such as FP64 double precision compute & deep learning training that uses FP16. This GPU uses HBM2 memory.
  • GP102: This GPU is used in the Titan X. It has 12GB of GDDR5X memory, 384bit memory bus & 480GB/s of memory bandwidth.
  • GP104: This GPU is used in the GeForce GTX 1070 and the GTX 1080. The GTX 1070 has 1/4 of its shaders disabled and is connected to GDDR5 memory, while the GTX 1080 is a full chip and is connected to GDDR5X memory.
  • GP106: This GPU is used in the GeForce GTX 1060 with GDDR5 memory

*80 and *70 are the higher-end consumer graphics cards, while the *60 and *50 are in the middle to lower end of this generation of Nvidia cards. The Titan cards are for people with too much money for their own good.

If Nvidia follow their three-tiered model, they will eventually release a 1080 Ti, which serves as a “Pro” upgrade over the regular models, often timed in response to whatever AMD is cooking up with their Radeon series.

In other words, the options with Nvidia usually are:

  • *50
  • *60
  • *70
  • *80
  • *80 Ti
  • *80 Titan

There used to be a mobile “M” series, but Nvidia have done away with that with the 10 series.

As nice as it would be to get the 1080 over the 1070, the pricing is quite balmy:

Model US Price
GTX 1080 FE $699
GTX 1070 FE $449

These are the models sold by Nvidia, and the prices are even higher here in Europe.

Since we’ve decided on a 1440p144 setup, we’ll have to bite the bullet and get the 1080.

  • [ ] GTX 1070
  • [x] GTX 1080

But Nvidia aren’t the only ones who put out graphics cards; they also let third parties create so-called aftermarket cards that use Nvidia’s chips.

So how do we choose a 1080 model amongst these cards? Here are some criteria:

  • Size
  • Brand reputation
  • Temperature (cooling)
  • Noise
  • Watercooling support
  • Overclocking potential
  • VRAM—which can vary, but not for GTX 1070/80

(And cooler type, which we decided on earlier—go turbine.)

Unfortunately, side-by-side recommendations are very hard to come by; the only decent one is a German GTX 1080 benchmark. (“Anno 2205” is a videogame used for testing performance during load, “Last”, while “Windows-Desktop” is the idle setting.)

  • [ ] ASUS
  • [ ] MSI
  • [ ] EVGA
  • [ ] Gainward
  • [ ] Palit

Going by the benchmark, the best card on the metrics of temperature and noise are the Gainward and Palit, both of which are from the same company with similar designs. And they also have the right cooler.

I can’t say I care about overclocking or watercooling1, and the VRAM is the same across all 1080 models.

Brand reputation is often very anecdotal, and I don’t know a good way to decide the winners and losers here in an evenhanded manner. I do see a lot of stories about poor quality control (QC) from ASUS, though.

Update 10/20 ‘16: EVGA has been busted in messing up their Pascal models with incompetent cooling according temperature scans discussed on the EVGA forum and reddit. Also, Google. Ouch.

Don’t buy an EVGA card.

  • [-] EVGA

As for size, this is a matter of what can fit in your case. To put it into perspective, compare one of the largest cards with one of the smallest, both of which use the GTX 1070 chip:

GTX 1070 GameRock Premium card
GTX 1070 GameRock Premium by Palit.
1070 Mini ITX OC card

Even if you have a large case, you can still run into problems with space and slots as evidenced by an unfortunate redditor who bought two 1080 GameRock PEs for SLI. Measure, measure, measure.

Gainward vs Palit

We’ve now narrowed it down to the Gainward and Palit GTX 1080.

Gainward’s GTX 1080 is available in a whole rainbow of versions according to their website:

The one featured in the benchmark was the GLH.

Comparing the specs pages makes it look like the only difference is the clock speed on the cards.

Palit is even more confusing:

The only difference I can tell is clock speed and memory bandwidth.

G-Panel is apparently a device for monitoring GFX info. Won’t be needing that.

From what little I can gather, between the JetStream and GameRock edition, the GameRock is the better one.

The card used in the benchmark was the GameRock.

This leaves us with the following graphics cards:

  • Gainward
    • Phoenix
    • Phoenix GS
    • Phoenix GLH
  • Palit
    • GameRock
    • GameRock Premium

When it comes to Palit and Gainward, it gets a little trickier for Americans, because availability differs between Europe and the US. Both are very popular manufacturers, but you somehow don’t hear a lot about them. I am buying from Denmark, so obviously my decisions aren’t applicable to all cases. Keep this in mind, if you’re deciding between the two; there’s no point in picking a card that’s not even available in your country. In my case however, both are.

The easiest way for me to decide is just to look at the price difference.


Card Price (DKK) Price (USD)
Phoenix 5,899 889
Phoenix GS 5,899 889
Phoenix GLH N/A N/A


Card Price (DKK) Price (USD)
GameRock 5,699 899
GameRock Premium 5,999 879

(US prices rounded to nearest 9.)

Finding the cards in US shops is pretty difficult, but from a cursory glance, it looks like the prices here in Denmark are around 50% higher than in the US. Swell.

Apparently, the price between the different cards is not something that varies much here in Denmark. With the Phoenix GLH unavailable, this mean it comes down to:

  • Gainward Phoenix GS
  • Palit GameRock Premium

If you can recall, the two cards in the benchmark were the Phoenix GLH and the GameRock Premium. In light of the unavailability of the GLH to compare to the GS and GameRock Premium, we’ll be going with the GameRock Premium, which performed the best in the benchmark anyway.

  • Gainward
    • [ ] Phoenix
    • [ ] Phoenix GS
    • [ ] Phoenix GLH
  • Palit
    • [ ] GameRock
    • [x] GameRock Premium
GTX 1080 GameRock Premium card
GTX 1080 GameRock Premium by Palit.

Remember that Palit and Gainward are the same company, and the cards are virtually identical, so brand doesn’t factor into the decision here.


There are only two CPU manufacturers in town: Intel and AMD. To keep it short, I’ll be going for Intel, for the main reason that this is a high-performance PC.

  • [x] Intel
  • [ ] AMD

<Update August 2017> Funny story! Turns out AMD released Ryzen which—after a lot of fixes after a shaky launch—turned the industry and CPU market upside-down, and they’re actually entirely competitive with Intel now!

Intel still leads in single-thread performance, but developers are learning to optimize for more than four cores (ie Ryzen), so who the hell knows which you should get by the time you’re reading this. </End of update>

Just like there’s the Pascal microarchitecture of 10_0 and the former 9_0, and so on, the nomenclature for Intel’s is:

uarch Nomenclature
Sandy Bridge 2000
Ivy Bridge 3000
Haswell 4000
Broadwell 5000
Skylake 6000—current one
Kaby Lake 7000—upcoming one
Coffee Lake 8000
Cannonlake 9000

Pay attention to the chips’ motherboard socket and memory compatibilities:

uarch Socket Memory
Sandy Bridge 1155 DDR3
Ivy Bridge 1155 DDR3
Haswell 1150 DDR3
Broadwell 1150 DDR3
Skylake 1151 DDR4/DDR3L
Kaby Lake 1151 DDR4/DDR3L
Coffee Lake TBA DDR4
Cannonlake TBA DDR4

This will determine the kind of motherboard and memory you buy; if getting an older CPU microarchitecture now seems sensible to you, remember that you will have to replace the compatible motherboard with a one with a newer socket. Same goes for your RAM. This is very important with regards to futureproofing. In light of this, it’d be dumb not to go with Skylake—or something newer:

  • [ ] Haswell
  • [ ] Broadwell
  • [x] Skylake

The same applies to going with a processor that supports DDR4 memory.

<Update ahead of Coffee Lake launch, August 2017> Intel’s been feeling their oats after AMD released Ryzen, resulting in a lot of unforced errors. While Coffee Lake supports LGA 1151 on paper, Coffee Lake only compatible with the 300-series motherboards.

In other words, you can’t use the nominal socket type to determine CPU-motherboard compatibility anymore—or it’d be stupid to trust it anymore at least. </End of update>

Each Intel microarchitecture is divided into five subcategories:

  1. Celeron
  2. Pentium
  3. Core
    • i3
    • i5
    • i7

Think of it the way you do the Nvidia’s GTX 1060, 1070, 1080 (etc).

When it comes to high-performing computers, it comes down to either of the last two:

  • Core i5
  • Core i7

This is entirely true; these are the consumer-brand categories. There’s also the Xeon intended for servers, data centres, and people who don’t pay enough in taxes. They cost thousands and thousands of dollars, but if you’re merely upgrading an old computer, you might be able to snag some discounted Xeons for a bargain after said data centres replace their hardware. They also to overclock better than anything else.

Fancy-pants CPUs are generally referred to as HEDT, High-End Desktop.

This comes in handy when companies like Intel launch CPUs with nonsense names with -E and -X affixes or i9- prefixes. The E stands for “Extreme”, you see. “But w—”, you ask before giving up. I know. Intel clearly don’t.


Our choice of processor will have a name like this:

Core iN-6XX0(A)

Where N is the number 3 or 5 and XX comes down to the individual model, which we’ll figure out later on. But what is A?

Because things aren’t complicated enough, Intel also slaps on a letter to add another processor classification.

The two salient classes, from what I’ve researched, are K and X.

I’ll let Ars Technica’s Andrew Cunningham explain the K series:

K-series: This relatively rare suffix denotes a multiplier-unlocked CPU that can be overclocked when paired with a high-end Intel Z170 chipset. The chips also have a higher 91W TDP, relative to the standard 65W for a quad-core CPU.

“Pentium? Core i5? Core i7? Making sense of Intel’s convoluted CPU lineup”

The K series is particularly relevant for overclockers, given Intel’s habit of locking down non-K chips through BIOS updates, which may be a result of how competitive overclocked older generations of can be Intel processors. The K series is also known as the “unlocked” version.

You can read what overclocking support means for your choice of motherboard in the “Chipset” subsection under “Mobo”.

I recommend Ars Technica’s Intel processor guide, if you’re still confused about Intel processors. Can’t say I blame you.

There’s also Intel’s own guide to their confusing naming. At least they seem to realize they have a problem.

While we may not want to overclock now, overclocking can stay the replacement for a while, when or if it should come a performance bottleneck. In light of this, we will need a K-series processor; the price difference between a regular processor, 6XX0, and its K counterpart, 6XX0K, is practically nothing, so there’s no reason to not go with the K processor.

Going through the list2 of available Intel processors, the only available K processors for i5 and i7 are:

  • i5-6600K
  • i7-6700K

Let’s compare the two:

Model Cores HT Base Clock Boost Cache Price
i5-6600K  4 4 3.5GHz 3.9GHz 6MB $219
i7-6700K 4 8 4.0GHz 4.2GHz 8MB $319

So the differences are:

  • Hyper-threading
    • Only i7 has it, hence the 8 virtual cores vs its 4 physical cores
  • Clock speed
  • Cache
  • Price

Here is a primer on hyper-threading:

<Update August 21 2017; Coffee Lake is presented> O.K., so once again, thanks to the new competition from Ryzen, Intel decided to make it even more impossible for me to explain the differences between i3, i5, and i7.

In the months leading up to the Coffee Lake presentation, Ryzen, after months of post-release optimizations, became the best (only) option for people who don’t want to buy an i7-7700K (Kaby Lake). Literally: get an i7-7700K or something by AMD, be it of equal, lower or higher parity. Because Intel’s response to Ryzen/Threadripper with Kaby Lake-S/i9 was a colossal disaster, AMD also became the preferable CPU brand for people who wanted to shell out more money for something other than i7-7700K. And if you want to go below, 1600 or lower is the smart choice. It’s either 7700K or AMD.

In other words, there was basically no reason to get an i5—and definitely not an i3.

As a result, i3 will now also have 4 cores instead of 2, and i5 will go from 4 to 6 like the i7, but still without hyper-threading. To summarize:

  • i3: 4C/4T vs 2C/2T
  • i5: 6C/6T vs 4C/4T
  • i5: 6C/12T vs 4C/8T

Thanks to AMD for lighting a fire under Intel’s ass, but also “thanks a lot” for making all the pre-Coffee Lake advice in this guide a pain in the ass to update and rewrite. </End of update>

Is the $100 premium worth it to you? I’ll take it.

  • [ ] i5-6600K
  • [x] i7-6700K

If you’re reading this piece in 2017, you’ll want the Kaby Lake version instead, ie the i7-7700K3. Note that Kaby Lake is planned to only work on Windows 10 (and Linux):

Intel spokesman Scott Massey told us, “per Microsoft’s support policy, they made the decision that Windows 10 would be the only Windows OS supported on 7th-gen Intel Core” processors. He added: “The Microsoft support change obviously doesn’t impact other operating systems.”

Linus’s spreadsheet

This part is completely optional, so feel free to skip it.

Linus just made a video on picking an Intel CPU; it’s not too straightforward, but it expands on what you’ve just learnt:

What’s really handy, though, is the spreadsheet of Intel processors attached to the video.

I downloaded the .xlsx spreadsheet, uploaded it to Google Docs, converted it to a CSV file, and removed the warning rows. I’ve checked to make sure it passes validation in csvlint. This way, you won’t need something proprietary like Excel to view and manipulate it4. And we can perform a lot of simple Bash operations, as you will see below.

With a CSV file, I can now type the following in Bash:

$ cat linus.csv | grep -i lga1151 | grep -i Core | grep -v -i i3- > linus-matches.csv

What this does is:

  1. Output all of linus.csv
  2. Pipe the output to grep and return the lines that fit a case-insensitive search for “lga1151”
  3. Return all of those lines that contain “Core”
  4. Return all of those lines that do not contain “i3-“
  5. Save the result as a new file named linus-matches.csv

Even shorter:

$ cat linus.csv | grep -iE "lga1151|Core" | grep -vi i3- > linus-matches.csv

This will take us from 87 CPUs to 36 to 18 to 12. Do note that this will remove the first line of headers, so just manually copy-paste this from linus.csv to linus-matches.csv.

There are also a lot of columns in the spreadsheet, and we won’t need all of them. It also turns out that all our remaining CPUs have the Skylake microarchitecture, so we won’t need that column either.

  1. Order
  2. Model
  3. Cores
  4. HT
  5. Base Clock
  6. Boost
  7. Cache
  8. TDP
  9. Price Low
  10. Price High
  11. Amazon Price
  12. Amazon Link
  13. Reviews
  14. Socket
  15. Onboard Graphics
  16. Mobo price factor
  17. uarch
  18. Corrected Price
  19. Notes

In other words, we want delete columns 9–18. To do this, we type:

$ cut -d, -f9-18 --complement linus-matches.csv > linus-clean.csv

What this does is:

  1. Using , as delimiter
  2. Take columns 9–18
  3. And return the other columns
    • In other words, remove the columns that are not returned, ie 9–18
  4. Save to linus-clean.csv

You may get the error cut: illegal operation -- -. The tl;dr of this is that you have a version of cut that doesn’t support the --complement method. I got this on macOS, but Bash (ie Ubuntu) for Windows 10 works.

Instead, you’ll have to ask for the columns you want to keep:

$ cut -d, -f1-8,19 linus-matches.csv > linus-clean.csv
Order Model Cores HT Base Clock Boost Cache TDP Notes
4-1 i5-6400T  4 4 2.2GHz 2.8GHz 6MB 35W
4-2 i5-6400  4 4 2.7GHz 3.3GHz 6MB 65W
4-3 i5-6402P  4 4 2.8GHz 3.4GHz 6MB 65W
4-4 i5-6500T  4 4 2.5GHz 3.1GHz 6MB 35W Tray only
4-5 i5-6500  4 4 3.2GHz 3.6GHz 6MB 65W
4-7 i5-6600T  4 4 2.7GHz 3.5GHz 6MB 35W 3rd party sellers tray only
4-10 i5-6600  4 4 3.3GHz 3.9GHz 6MB 65W
4-8 i5-6600K  4 4 3.5GHz 3.9GHz 6MB 91W
5-2 i7-6700T 4 8 2.8GHz 3.6GHz 8MB 35W Tray
5-1 i7-6700 4 8 3.4GHz 4.0GHz 8MB 65W
5-3 i7-6700K 4 8 4GHz 4.2GHz 8MB 91W Amazon #1 best seller
5-4 i7-6785R 4 8 3.3GHz 3.9GHz 8MB 65W wtf
Intel CPUs supporting LGA 1151 sockets.
Intel CPU data courtesy of Linus Tech Tips.

I included the Amazon links in case it might earn the Linus people some money, even though we removed it from our initial dataset and I prefer people avoid using Amazon whenever they can.

I also removed some superfluous text like

  • “Intel”
  • “Core”
  • “®” and “™”

This can also be done from the terminal:

$ cat linus-clean.csv | sed "s/Intel\|Core\|[®™]//g" > linus-clean.csv


$ cat linus-clean.csv | sed -r "s/Intel|Core|[®™]//g" > linus-clean.csv

You can download the spreadsheets below:

The performance and hubris of Intel CPUs

You should not upgrade your Intel CPU frequently, because for whatever reason, Intel have become complacent with their lead in the high-end CPU market and only improve their CPUs incrementally; sometimes, the newer microarchitecture does worse than the old one in some areas.

The main improvements have to do with power use at idle load and multi-core performance—which many videogames aren’t optimized for. Linus has made a great video to show when you should and shouldn’t upgrade your CPU well worth checking out:


This section goes over how to pick a motherboard for an Intel CPU. As mentioned before, this is a no-AMD zone, be it graphics cards or CPUs.


In the previous CPU section, we made a choice establishing that our motherboard support socket 1151.


When you pick a CPU that supports manual overclocking, ie K and X, you need a Z-type motherboard chipset.

If you go with a standard processor that doesn’t support manual overclocking, you should get an H-type motherboard chipset. Z-type chipsets are generally for overclocking, but they’re also the bells-and-whistles motherboard with additional features.

Two other chipset types are B for business and P for pure overclocking without the feature set of Z.

For an overview comparing the different chipsets, read the wiki on LGA 1151 socket.


We want support for DDR4 and at least 3000 MHz clock speed. Better to do DDR 4 than 3 when our CPU supports it, and 3200 MHz is the clock speed where the prices on RAM is still reasonable. See the RAM section for more.


Here are some of the things you want in a motherboard:

  • Good ethernet controller
    • 5G speed (802.3bz)
  • 4+ RAM slots (DIMM)
  • Quad/dual channel support (more on this in the RAM section)
  • USB 3.0/USB 3.1/Thunderbolt 3
    • USB 3.0 headers (to enable USB ports in your case)
  • 5+ rear USB ports
  • Good fan control, ie fan headers and software
    • 3+ chassis fan headers (CHA_FAN) with 4-pin PWM
      • CPU fan
      • Front chassis fans
      • Rear chassis fan
      • Bottom chassis fans
      • Top chassis fans
      • Side chassis fans
    • Either that or voltage to support y-splitter cables
  • M.2 support
    • So-called “M.2 shield”, which works as a heatsink
  • Fast PCIe lanes (3.0 ×16)
  • Solid overclocking support

A quick table of PCIe speeds that double each generation:

Version Per lane (MB) 16 lanes (MB) Per lane (GT) 16 lanes (GT)
v1.x 250 MB/s 4.00 GB/s 2.50 GT/s 40 GT/s
v2.x 500 MB/s 8.00 GB/s 5.00 GT/s 80 GT/s
v3.0 985 MB/s 15.75 GB/s 8.00 GT/s 120 GT/s
v4.0 1969 MB/s 31.51 GB/s 16.00 GT/s 256 GT/s

Here are the things others might be interested in that I don’t care about:

  • Good on-board audio
  • (Good) on-board wi-fi
  • Watercooling monitoring support
  • Diagnostic LEDs
  • Cosmetic LEDs/RGB stuff (oh god)


<Update since Kaby Lake was released>

For the Kaby Lake processor, you’ll want a _270 motherboard, which is getting launched at the same time as Kaby Lake. The compatibility of _170 with Kaby Lake is a little unclear, except that they use the same socket type.

To quote bit-tech in the excellent Z270 overview:

While Intel’s new CPUs don’t offer much by way of upgrade allure, we do at least have plenty to look forward to from motherboard manufacturers.

“Z270 Motherboard Preview Roundup”

Should you get a Z270 motherboard, if you’re still going with a Skylake CPU? The motherboards are a big improvement, but they also come with a price premium, so weigh the pros and cons and decide for yourself.

Presumably, a Z370 will be released along with the next Intel microarchitecture, which might use a different, incompatible socket type.

</End of update>

Let’s see which compatible motherboards are available for this build according to PCPartPicker:

Motherboard Form factor RAM slots Max RAM Price
Asus Z170 PRO GAMING/AURA ATX 4 64GB $164.99
Asus P10S WS ATX 4 64GB $211.99
Supermicro C7Z170-OCE ATX 4 64GB $296.13
ASRock B150M Pro4V Micro ATX 4 64GB $72.99
Asus H110M-E/M.2 Micro ATX 2 32GB $59.98
Gigabyte GA-H110M-S2H Micro ATX 2 32GB  

We only want the ATX motherboards, since we have a case to accommodate it:

Motherboard Form factor RAM slots Max RAM Price
Asus Z170 PRO GAMING/AURA ATX 4 64GB $164.99
Asus P10S WS ATX 4 64GB $211.99
Supermicro C7Z170-OCE ATX 4 64GB $296.13

If you can recall, we picked a K-series Intel processor which requires a Z170 chipset—on top of the additional features a Z170 motherboard brings. Z-type motherboards may also increase compatibility with future CPU microarchitectures.

By removing non-Z170 motherboards we get:

Motherboard Form factor RAM slots Max RAM Price
Asus Z170 PRO GAMING/AURA ATX 4 64GB $164.99
Supermicro C7Z170-OCE ATX 4 64GB $296.13

Supermicro is too obscure a brand, and there’s no point in the $130 premium. That leaves us with just one candidate:

ASUS Z170 Pro Gaming/Aura motherboard Rear ports on the motherboard
Z170 Pro Gaming/Aura motherboard by ASUS.

Let’s look at some of the specs:

  • Memory: 3466 MHz (OC)
  • USB:
    • 3.0: 4×rear, 2×in-board,
    • 2.0: 2×rear, 6×in-board
    • 3.1: 1×rear
    • USB-C: 1×rear
  • SATA: 6×6 GB/s
  • M.2 (socket 3)
  • PCIe ×4 (Gen 3.0 ×16)
  • Intel Gigabit ethernet controller

And much more.

  • [x] ASUS Z170 Pro Gaming/Aura


Picking RAM is fairly boring, but we could use some boring for a change.


Because our processor supports DDR3L/DDR4, we are going with the futureproof DDR4 option. Which brings us the the next thing to figure out.

Clock speed

Basically just get as high a clock as possible without getting into a crazy price tier. The price doesn’t tend to go up by much so you may as well get as much out of it as possible. The prices seem reasonable around DDR4-3200 right now. Keep in mind that your motherboard’s RAM support only goes up to a certain clock speed, so make sure you get a motherboard that supports 3000 or better.

Check out this benchmark video by Hardware Unboxed to see the difference clock speed—sometimes—makes around the 3:11 mark:

Check out the Fallout 4 results at 5:22 to see one of the cases where clock speed makes a world of difference.

And a more recent one that starts around the 3:31 mark:

Update 04/21 ‘17: Check out this new Linus video, too:

CAS latency

CAS latency, sometimes abridged “CL”, is the opposite of clock speed; the lower the better. Given the choice, go with the lowest option.

If you see something like “3200C14D” in the SKU for some RAM, the “3200” is for the clock speed, and the “C14” means a CAS latency of 14.


You’ve for four options for memory:

  1. Single
  2. Double
  3. Triple
  4. Quad

Here is Linus explaining memory channels:

As it stands now, Skylake and LGA 1151 don’t support quad channel:

It’s also important to note that, while Haswell-E and X99 supported quad-channel operation, Skylake and Sunrise Point will only do dual-channel. Presumably Skylake-E will again bring back a quad-channel integrated memory controller.

Ars Technica’s i7-6700K review

So we’ll be going with dual channel. These are the RAM sticks that come in twos—unsurprisingly.

Update 03/08 ‘18: As for performance, the improvement of quad channel in games is practically zero for typical use cases.

This also serves us fine, since people never buy four sticks of RAM for their initial build. Instead, we’ll just buy them in pairs of two.

Putting it all together

How many RAM do we need, by the way? 16 GB should be more than enough, unless you’ve got some very specific needs. If we wanted to go 32 GB, we could either go 2×16 GB or 4×8 GB, but since I don’t ever plan on going 64 GB, I’ll just go 2×8 GB and leave the door ajar for going 4×8 GB by buying two more identical sticks at a later point.

We should now have all our criteria:

  • DDR4
  • ≥ 3000 MHz clock speed
  • Dual channel
  • 16 GB

Some make a note out of the different profile for some models, meaning some reach a bit higher, which may or may not complicate your setup.

These leave us with a list of memory candidates with some manufacturers I’ve never heard of removed.

Ultimately, RAM is just RAM, so get the highest clock speed you can for the best price possible for the products that are available in your country. Remember that your motherboard may only support up to around 3200, however. Ours supports 3466.

Another thing you can filter the results by is the CAS latency. If we keep increasing the “CAS” filter, we end up with four candidates with a CAS of 14, compared to the original 19:

Product DDR type Price
G.Skill Trident Z Series DDR4-3200 $109.99
G.Skill Trident Z Series DDR4-3200 $109.99
G.Skill Ripjaws V Series DDR4-3200 $102.99
G.Skill Ripjaws V Series DDR4-3200 $99.99

Judging by this G.Skill press release, the Trident Z series have a much better design as well as superior timings compared to their Ripjaws V sibling.

I wager the only differene between the two Tridents is their colour. Here’s an overview of the Tridents and Ripjaws that meet our criteria. You can always match the exact product IDs with the one in the link to determine the colouring. I’m partial to the F4-3200C14D-16GTZKW.

G.Skill Trident Z Series DDR4-3200 RAM
Trident Z Series F4-3200C14D-16GTZKW by G.Skill.
  • [x] G.Skill Trident Z Series DDR4-3200
    • F4-3200C14D-16GTZKW

Update: These memory sticks are now available with built-in RGB lighting. They only cost slightly more than their unlit counterparts. Here’s a video of them in action:

Update: GLooks like G.Skill’s RGB RAM have some serious issues with at least Ryzen CPUs. It messes with the SPD for some reason, and your only recourse of action is basically to either return it or try out some obscure software to fiddle with the SPD yourself.

You can watch a YouTube trip report about it below:

Picking computer components is and will always be an absolute pain.

The final build

This is what we ended up with:

Component Model Notes
Case beQuiet! Dark Base 900 Door panel issues
GFX Palit GTX 1080 GameRock Premium Unavailable in NA
CPU Intel i7-6700K Get i7-7700K if available
Mobo ASUS Z170 Pro Gaming/Aura Get Z270 if available
RAM G.Skill Trident Z Series DDR4-3200 (2×8 GB) F4-3200C14D-16GTZKW

You can inspect and re-use the build on PCPP.

Because I already have my CPU cooler, power supply, Windows license, and storage drives, I won’t be picking these components, leaving you free to fill in the blanks.

Since the graphics card is hard if not impossible to find outside Europe—which is why there are no price nor wattage for it on PCPP—you might want to replace it as well. Recall the original 1080 benchmark to guide your decision.

One last thing you may want to do is hit up the SKU (Stock Keeping Unit) of each of your components and enter those into whatever shop you end up going with. The SKU is a unique product ID, and the reason you should use it here is that some components tend to come in weird variations that may not be the one you thought you ordered—and you’ll usually only realize this when it’s too late. So try to get the SKU of the component you’re sure is the right one and enter that rather than the product name into the search field for your shop of choice.

Failing that, you can go with the item model number like I did with my memory sticks: F4-3200C14D-16GTZKW. This also conveniently includes the frequency, latency, and colour information within it to make it harder for me to pick the wrong model when ordering.

Other types of identification codes include:

  • UPC: Universal Product Code (US)
  • EAN: The European Article Number
  • ASIN: Amazon Standard Identification Numbers
  • ISBN: The International Standard Book Number


Given that my current PSU is six years old as of October 2016, I should probably replace it to make sure it doesn’t take down the new rig with it.

Reliability and reputation

You probably don’t want your PSU to explode, but how do you assess who’s good and who’s bad?

You can do worse than Tom’s Hardware’s PSU tier list.

My general impression that the most reputable brands, in no particular order, are:

  • beQuiet
  • Seasonic
  • Antec
  • Corsair


There are three types of modularity:

  1. Non-modular: all cables come pre-attached
  2. Semi-modular: essential cables pre-attached
  3. Full-modular: no cables pre-attached

The more modular, the more expensive and simpler the setup.

Go for the most modular option you can.

80 Plus Tiers

There are six tiers of power effeciency:

  1. Standard
  2. Bronze
  3. Silver
  4. Gold
  5. Platinum
  6. Titanium

I could give you the in-depth explainer, but Corsair’s Jonny Gerow has already done it in his 80 Plus explainer—which doesn’t include the rarer Titanium tier.

Basically, the higher the tier, the more you’ll save in electricity. The higher the tier, the higher the price, but with our kind of use, the highest tier will definitely save us the most money.

PSUs tend to last 5–7 years, so replace yours if it’s older than that, and check for a 5-year warranty on it, if it kicks the bucket before.

Total wattage

The main takeaway here is that high wattage supplies are a racket, as even gaming computers will usually draw 400-something under heavy load.

And the newer your components are, the less power they are likely to draw. The newest generation of Nvidia GFXes, the 10_0, don’t even support more than two cards in SLI anymore.

Keep in mind how the power efficiency described in the previous section affects the wattage you need.

Also remember that overclocking increases the power usage. And that underclocknig decreases it.

In “Types of fan control” below, you will see that some PSUs have fan control depending on the “load percentage”. This means that it’s one thing to fill the power consumption need and another to exceed it. One example is Seasonic’s “Hybrid” with three tiers of fan speed for 30, 50, and >50% of load. Let’s just preview it here:

Seasonic’s three phases of fan control depending on load.
Seasonic’s three phases of fan control depending on load.

This means that if you have a 1,000W PSU, you can’t exceed 300W of power consumption for fanless mode, 500W for silent mode, and that your PSU fan speed gradually increases for anything above 500W.


The PSUs usually come with a built-in fan and an a grill on the top. The fans should be a part of your decision-making in picking a PSU.

Noise and control

Silent PC Review have a great list of recommended silent PSUs. Seasonic and beQuiet seem to do well.

While noise for a single component is important, keep in mind that a silent PSU is not going do too much for you, if your GPU or CPU sound like a lawnmower.

One way to approach is like what Seasonic are doing with one of their PSU’s below by creating three tiers of load and adjusting the fans accordingly:

  • Idle
  • Low load
  • High load

During idle and low load, you’ll want your computer to be silent and your fans to adjust accordingly. Whether you want to swing for the fences and go for a computer that’s silent during high load is up to you, but I think we should definitely go for a PSU that’s silent during idle and inaudible during low load.

Keeping the computer silent during high load will require a fan setup beyond the scope of this guide.


650W will be enough in 99% of cases. I always seem to run into edge cases, so I might opt for a 750W just because.

The relevant product series appear to be:

  • Seasonic
    • X
    • Platinum
    • Prime Titanium
  • Corsair
    • AXi
  • beQuiet
    • Dark Power Pro

The latest version the Dark Power Pro is currently the Dark Power Pro 11. Before that, we will look at Seasonic’s Platinum 660 and Corsair’s AX760i (because a 660 version isn’t available in Denmark).

The Seasonic Platinum from behind, modular with a Hybrid/Normal switch.
Note the Hybrid/Normal switch on the modular Platinum.
Seasonic’s three phases of fan control depending on load.
Seasonic’s three phases of fan control depending on load.

Here is what the graph looks like for the Prime Titanium series:

Similar to the other Hybrid graph, but the middle phase is “Hysteresis” instead of a dedicated silent mode
Similar to the other Hybrid graph, but the middle phase is “Hysteresis” instead of a dedicated silent mode to triage between the fanless and cooling mode. Notice the slightly gradual increase in fan noise within all modes.
Connectors ordered by category.
The Prime Titanium has better ordering of connectors behind; the Hybrid button is on the front with the power switch.
The Corsair AX760i.
The Corsair AX760i.

Corsair’s AXi series lets you control your PSU manually through a software interface, as opposed to Seasonic’s built-in on-and-off system that only requires you to flip a switch on the PSU itself.

The “Link” interface for the Corsair PSU which monitors and controls power, fan speed and temperature.
The “Link” interface for the Corsair PSU which monitors and controls power, fan speed and temperature after connecting the PSU to a USB header on the motherboard.

The beQuiet does not, from what I can surmise from their god-awful product page, seem to have any significant fan control.

The beQuiet Dark Power Pro 11 with clearly ordered and labelled connector ports.
The beQuiet Dark Power Pro 11 with clearly ordered and labelled connector ports. Notice the dedicated fan controllers. ("OCK" stands for "Overclocking-Key").

The Dark Power Pro 11 page also reads:

  • Variable fan speed is optimally governed to strike the best balance between proper cooling and deep quietness

  • Up to four case fans can be connected to, and regulated by the PSU, thus reducing overall system noise even further

There doesn’t even appear to be a “Hybrid” button here.

Sometimes, price can make the decision. Let’s look at the price chart:

Model Manufacturer Price (DKK)
Prime Titanium 650 Seasonic 1,700
Platinum 660 Seasonic 1,700
AX760i Corsair 1,500
Dark Power Pro 11 650 beQuiet 1,300
X-650 Seasonic 1,195

These are through my go-to shops in Denmark. Newegg’s pricing is all over the place, so I don’t have a good way to compare the prices internationally.

The Prime Titanium looks like a step up from Platinum, so at price parity, I might as well go for the Prime Titanium.

One minor hiccup: PSU labelling isn’t super helpful when you turn the PSU upside-down (fans-down), which is the preferred orientation in cases that support it. It’s nice that they’re grouped by category, though.

Types of fan control

There are three approaches to fan control in the candidates:

  1. Automatic (Seasonic, beQuiet)
  2. Manual (Corsair)
  3. None (not applicable here)

I’d like some kind of control depending on load, as that would bring down the noise, so whether I go with #1 or #2 depends on when exactly the automatic fan control triggers for Seasonic and beQuiet’s PSUs.

Final PSU thoughts

This has honestly been the most difficult section for me to write, and, unlike other sections, this one is far less objective and probably comes down to taste, gut feeling, and randomness more than any other component. By the end of the day, it’s all going to be fine. I’m gonna go with the Prime Titanium 750.

  1. Jay does a great job of explaining whether watercooling is worth it it to you↩︎

  2. The selection of our processor from Intel’s list of products may have happened a little fast for you, and if you want more details, read the next, optional, subsection called “Linus’s spreadsheet”. ↩︎

  3. Kaby Lake isn’t a huge improvement over Skylake; the main differences are:

    • HEVC (x265) and 10-bit support
    • Supports Netflix 4K on Windows
    • Performs about the same as Skylake in most instances


  4. Note that some values like price didn’t make it for whatever reason, so make sure to import those if they matter to you. ↩︎