Configuring a Computer

I decided to write up a guide to configuring a computer.  I'll try to keep this accurate and edit it with any changes in the future.  For TL;DR, read this first bit and the look for the "→" in the component sections for a "quick" answer.

First, what is your intended use?

• Gaming (which games?) - Gaming is usually very graphics intensive, but some can put demands on other hardware.  If you want to know if a particular game works on a system of a certain configuration, do a web search for a benchmark of the game using a particular GPU (e.g. GTX 970, R9 280, etc.); multiple reviews benchmarking most popular games are out there and the reviews use different processors, so you can compare and contrast how the game performs on different set-ups.

• Light Use (Web Browsing, Word Processing, Watching Videos, Streaming (watching/download, NOT upload of a game stream), etc.) - This really doesn't require much out of computers these days, honest.  I think it is safe to say that all Digital Storm machines will outperform any mobile device like a tablet or smart phone.

• Video/Image Editing - Highly driven by memory performance and can usually take advantage of many CPU cores, multiple graphics cards, and pretty much everything you can throw at them these days.

• Software Development - This is more dependent on what you are developing and you will probably be the best judge for what you need here.

• High Performance Computing - Lots and lots of processing and/or memory (depending on the application; I've seen things that do fine on a dual socket high performance Xeon CPU, but required nearly half a TB of memory due to the nature of the computation).  Stability is also much more important and drives the need for error correction with ECC memory (requires a Xeon with Intel; some lower end AMD processors do support ECC) and ECC VRAM (Quadro for NVIDIA and FirePro for AMD).

• etc.

Next, what monitor(s) do you want to use?  Keep in mind:

• Number of Monitors

• Screen Resolution

• Refresh Rate

• Color Depth

• Dynamic Range

• 3D (stereo)

All of these play a role in how much bandwidth needs to go to each monitor; the larger the bandwidth, the more video processing the graphics card(s) need to have.  I think most people understand how the number of monitors, screen size (width multiplied by height), and stereo 3D (one image per eye) corresponds to the bandwidth.  Refresh rate...for games you would like to keep it above 30 frames per second (FPS) and preferably above 60 FPS.  Some monitors go higher to 120 (2x bandwidth) or 144 FPS (nearly 2.5x bandwidth) to reduce motion blur.  High demanding graphics in games will put a strain on the graphics cards, so this is where looking at the benchmarks counts.

There are proprietary refresh rate technologies: G-Sync for NVIDIA and FreeSync for AMD; these require the technology in the monitor as well as the respective brands graphics card. Color depth and dynamic range haven't had much emphasis on them yet...you'll see an occasional monitor advertising 10-bit color depth (requiring Quadro graphics card for NVIDIA, AMD supports 10-bit on all cards), but hasn't been pushed yet.  However, that will probably change with more UHD content coming out; check out the UHD Rollout Timetable for expected specifications and timetable.

Okay, so now that we've identified some of the most critical aspects of determining the use, we'll get into the components.

I'll touch briefly on the case/chassis right now but will get more into it near the end.  If looks or a size (form factor) are high priorities, take a look at what interests you and see what can fit in what you desire.  There can be some trade-offs that need to be made and will be a bigger balance for small form factor chassis.'

Processor (CPU)

With Intel's recent dominance in CPU performance, AMD has been somewhat relegated to being the budget offering.  AMD is still competitive at the low end and worth a look.  Also, if you are getting and AMD video card, there are some neat interactions between an AMD processor and graphics card that goes on.  AMD also happens to power both the X-Box One and PlayStation 4.

Intel on the other hand has a range of processors.  For the light use scenario above, Pentium's and Core i3 processors should work just fine.

When you get into gaming, you'll want to look at the Core i5 and Core i7 ranges.  The Core i5 will actually work quite well for most games, as many games are not optimized well enough to take advantage of the many threads offered by Core i7 processors, there are always exceptions: look at benchmarks to help make the decision.  The processors with a K or an X at the end of them are unlocked and may be overclocked.  Before I get to far, I will note that there are two varieties of Core i7 processors.  There are the lower end (and cheaper) consumer variety that fit on the same motherboards as the lower end processors; and there are the high end "enthusiast" processors that use a socket also used in the enterprise world.  The high end require different motherboards and will have, in addition to more processing power, a lot more communications bandwidth to other devices on the motherboard.  For the most part, gaming will be just fine on the lower end Core i7 processors (remember, Core i5 processors will work for many uses, too); it is only when you get into having 3 or 4 graphics cards to drive very high resolutions and/or frame refresh rates, is there a need to move to the high end processors.

With video processing, as I mentioned above, lots of memory performance and pretty much everything else, especially when you are getting into UHD and 4K resolution video.  This points directly at the high end Core i7 processors.  More memory channels gives more bandwidth and more memory slots for a larger quantity of memory.  You also get more cores in the processor to be able to do more multi-threaded crunching.  Get what you can afford to get with the processor.

For software development, it depends on what you're doing.  A Pentium may suffice for basic web development, but for something like a Flash intensive development, it may require something a bit more substantial.  You'll know best.

High Performance Computing...you should be looking at the workstations.  Here you'll have options for Xeon server processors.  As mentioned above, they support ECC memory and have extra features to support robust operations.  They cannot overclock like the enthusiast processors; you should be prioritizing stability over clock speed and if you need more performance, you should scale out.

→ Bottom line: match your expected use to the relevant scenarios and benchmarks to determine the correct processor to get.

Motherboard

The motherboard will be narrowed down by the selection of the processor.  You should also have an idea if you need multiple graphics cards that a motherboard will need to support.

I'll start off by talking about the interfaces.  For example take this Gigabyte X99 board that supports the high end processors:

There is a memory bus on the right side of the processor.  High end processors will have 4 memory channels on them and low end processors will only have 2.  This effectively doubles the memory bandwidth, assuming all else is equal (which is usually not the case).

There are two kinds of PCI Express interfaces: version 3.0 coming directly off of the processor (there are 40 "lanes" on the high end processors and only 16 on the low end processors).  The second is the version 2.0 coming off of the X99 chipset that connects to the processor (there are 8 "lanes" on both X99 chipsets for high end processors and Z97/H97/Z87/H81/etc. chipsets for the low end processors).  Version 3.0 is essentially twice the bandwidth of version 2.0.

On the version 3.0 side, these are typically the connections for the graphics cards.  The graphics cards will have a x16 connector, meaning electrically, they can connect to up to 16 lanes of PCI-e 3.0.  However, it will still work with less, the communications bandwidth will just be that much less.  As you can see in the diagram above, there is a switch on the board that enables the electrical connection to be either configured as a single x16 (16 lane) slot or two x8 slots.  These slots will have the physical appearance of x16 slots to accommodate graphics cards, but when both slots have a graphics card (or other PCI-e card) installed in them, they only function as a x8 slot, electrically.  The switch will consume power and have a very slight impact on performance with the communication between the graphics card(s) and CPU (you can see a slight advantage in benchmarks where the lower end motherboards actually have an ever so slight advantage over the high end ones, as the high end motherboards are trying to pack in as many features which can be slightly detrimental to the performance of the motherboard, assuming all else is equal).

On the high end processors, CPUs with 40 lanes will be able to support x16/x16/x8 configurations while those with 28 lanes will only be able to do x16/x8/x4.  Check here to see how many lanes a particular processor has.  With the low end processors, they will only be able to support a x16 configuration.  Of course motherboard manufactures add all sorts of switches and timing circuits to enable multiple configurations of the PCI-e lanes, so check the specifications of the motherboards for all configurations supported.  (This is a little more complicated as you can make other configurations directly from the processor, but the motherboard manufacturers all support x16 layouts and will split out other configurations from there, so I won't get into it any more than letting you know there is more complexity, but nothing to concern yourself with)

USB and SATA connections will come off of the chipset.  See Chipsets to see how much natively is supported by the chipset.  There may be additional USB and SATA connectivity support through the use of third party controllers connected to the PCI-e 2.0 lanes on the chipset (this is the case with the Renesas controllers providing additional USB ports on the Gigabyte motherboard above); the connections off of the third party controllers will always be slower than those native to the chipset (the data has to go through both the chipset and the third party controller vs. just the chipset for the native connections).

Other uses of the PCI-e 2.0 on the chipset are SATA Express (2 lanes each), Ethernet (1 lane for each gigabit controller), Wi-fi (usually 1 lane), and M.2 (1, 2, or 4 lanes, but some M.2 interfaces connect directly through the PCI-e 3.0 for more bandwidth).

An important thing to keep in mind is that ALL of the data going through the chipset must go through the DMI 2.0 interface, which is equivalent to a PCI-e 2.0 x4 connection.  The SATA drives, USB connections, Ethernet, Wi-fi, etc. will all be competing for the same bandwidth over that connection.

→ So, what to look for in a motherboard:

• It will obviously support the CPU

• It can handle the memory speeds and amount of memory sticks you want

•   Some boards, like ASUS' TUF-series, can only handle limited memory speeds

• It will support how ever many graphics cards you need/want

•   Not all boards support SLI, so check to see if the motherboard supports it if you really want it

•   Crossfire is supported in many more boards, but be careful as some of those connection may be through the PCI-e 2.0 connections through the chipset, which as previously noted, are highly constrained interfaces

• It either has included or can expand to support desired features like:

•   Wi-fi (on-board, mini-PCIe card, M.2 card, PCI-e 2.0 x1 card, or USB)

•   Game streaming co-processor

•   Headphone sound card

• Manufacturers warranty

• If you've already decided on a chassis, then that it will fit it

• If you haven't chosen a chassis, but desire a smaller machine ultimately, look to smaller form factors like Mini-ITX and μATX boards.  They have nearly as many features as full ATX boards with many of the features that would be gotten with PCI-e cards now integrated directly on the motherboard.

Memory

While the headline for memory is capacity and speed, that is not the whole story of the performance of the memory.  Latency is also a big performance factor...speed determines the cadence (the "beat," if you will) to which everything moves, but the latency determines how many beats of that cadence each thing actually happens.  You want lower latencies and faster speeds, but you usually trade off one for the other, so you have to look at the cumulative effect.  There are multiple latencies factors, if you want to learn more about it, look it up -- I don't want to drag.

The use of DDR3 or DDR4 is dictated by the processor and consequently the motherboard how much of that memory can be used.  The speed and latency ratings of DDR3 and DDR4 are pretty much an apples to apples comparison, performance wise; the only noticeable difference between DDR3 and DDR4 is that DDR4 has better burst performance, so the first instant memory is accessed, DDR4 will have a slight advantage there.  DDR4 also operates a low voltage so it will consume about 20% (around 1 Watt) less per stick than DDR3.

→ Most use cases for casual use and gaming should be good with 8GB of memory right now.  If new features in games start using more memory, it could be useful to have 16GB.  Also, if you multitask a lot (like having many browser tabs open, multiple virtual machines running, etc.) it would be desirable to have more memory to support this.  As I mentioned above, applications like video editing require very large amounts of high performance memory and you should look to get as much as you can afford to get.

As for selecting brands - memory is very finicky and fickle.  Brands built their reputation on quality and Corsair, Crucial, and Kingston are up at the top.  Other brands have been improving quality as well, and product reviews around the internet can give a good amount of insight as to their current product quality.

Optical Disk Drive(s) (ODD)

With the increases in internet connectivity and expansion of available bandwidth, physical media is becoming more rare.

There are some advantages to retaining an ODD or two.  First, media being streamed is highly compressed.  Netflix or any other video streaming service highly compresses the video to fit in the available bandwidth...and what is available may not even be determined by your connection to your Internet Service Provider (just search for Netflix's conflicts and deals with Comcast, Verizon, and other ISP's to see the issues further upstream, beyond your control).  So, the choppy or blocky video feed...that is the result of the high compression of the stream.  Optical disks do not have any of this bottle neck, beyond being able to fit the file on the media.  So, this results in a much high quality video than can be had by streaming.

Furthermore, no streaming service will stream UHD (4K) video feeds to a computer; they only do it to devices dedicated to streaming and browsing.  This is for security as they are trying to control piracy of ultra-high resolution content.  So, for the foreseeable future, it appears as though the only way in which UHD movies will be playable on computers is with optical media.  While no official announce of what media will be used for UHD content, news from up the product chain in disk manufacturing indicates that BD-XL disks (three and four layer Blu-Ray) will be used.

If you would like to copy CD's, DVD's, or Blu-Ray's for back-ups, a second drive could be useful if you do a lot of it.  One drive can be a read-only drive that reads the disk to be copied, and the second drive would be a writing drive (CD-R/RW, DVD-R/RW, or BD-R/RW) that writes the back-up as the other drive reads.

Storage

NAND (or NOR) has a couple major features to it: node size, which is how small the features etched in it are; and the number of bits-per-cell.  Both of these are being pushed to provide greater densities, however lately increased manufacturing costs are starting to make gains moving to smaller node sizes neglegable and moving to the greater bits-per-cell are significantly reducing the life and performance.  The life is measured in number or writes and rewrites (erases don't count because you are only removing an index...the data is still present until the drive overwrites it with other data or in a garbage clean-up operation).

For bits-per-cell, there are a few varieties: SIngle-level Cell (SLC), Multi-level Cell (MLC; really 2 levels), & TLC (Tri-level Cell).  This describes how many bits in a cell is represented by the number of voltage levels each cell may possess (SLC has 2 voltage levels, MLC has 4, and TLC has 8).  Basically, the more voltage levels you have, the harder it is to distinguish between the levels and you will see slower response from extra digital signal processing to determine the voltage, as well as cells wearing out faster as tolerances are much tighter.  The obvious benefit of moving to more bits-per-cell is that you get that much more storage density.  You may also see V-NAND or something to do with 3D.  This is moving the planar lay-out into the 3rd dimension...think of it as having a ream of paper to write on instead of a sheet.  Currently Samsung is the only vendor with V-NAND out on the market with their 850-series drives (32 layers), but Intel has announced it will be joining later on this year with 2.5" drives exceeding 10TB of storage (presumably for enterprise drives ($$$$); Samsung just announced some 3.84TB enterprise drives).  With the vertical nature of the V-NAND, Samsung has stepped back to a larger node size, giving greater performance and a greater lifespan, compared to competitors products at around the same arial density.  This is why the 850 Pro drives have such good life, compared to all of the other MLC drives out there.  SanDisk has an interesting technology where they aportion some of the MLC or TLC NAND as a pseudo-SLC cache to take advantage of the SLC-like performance.

The complexity of controllers has dramtically increased since Intel jump started the SSD market with the X25-M drives.  I mentioned digital signal processing earlier: this along with other techniques like real-time data compression and deduplication have come a long way to combat the reduced life of smaller node sizes and more bits-per-cell to make SSDs much more affordable.  One of the areas of differentiation between the SSDs out there is the amount of verification and validation (V&V) of the controllers firmware as many of them use their own.  Being that the complexity has grown, the chance of a failure in a drive controller is that much greater, necessitating greater checks for issues with them.  Intel has been known ever since the X25-M drives to have some of the most extensive and complete V&V of the controllers and have the best reliability records out there.

The controllerw will have a number of memory channels.  For the lower capacity SSDs, not all of these channels will be populated; hence why you will notice substantially reduced performance on the smaller drives.  Once all of the channels have been populated, an increased amount of NAND is added to each channel.  There are slight performance gains with additional NAND as there is usually more free space to the controller to work with while writing data (it doesn't have to manage as many re-writes...this gets into the weeds of SSD performance)

As for interfaces with the SSDs, there are a few.  You will have the SATA connections, which are the same as used by HDDs.  These are bottlenecks in the performance of SSDs and a new interface called M.2 is emerging.  M.2 uses up to 4 PCI-e lanes, either 2.0 or 3.0, and new controllers taking advantage of the additional bandwidth are being introduced.  The only issue of using a PCI-e interface versus the traditional SATA interface is the configuration during boot and getting the operating system to appropriately recognize the drive.

I think I've probably gone into more than sufficient detail.  If you want to learn more about factors, such as garbage collection, write-amplification, and over-provisioning, I'd highly recommend looking at the information posted on Anandtech, as they have pretty much changed the controller industry by pointing out things like performance consistancy with their home-brewed benchmarks, and have done some of the most comprehensive tests of SSDs around.

Key things about SSD performance:

• NAND (or NOR) manufacturing size decreases will reduce cost, but also reduce the life span

• The number of bits-per-cell (SLC = 1, MLC = 2, TLC = 3) will reduce the cost the more you have, but also reduces the performance and life span, too

• Vertical-NAND stacking (V-NAND in Samsung terminology) will increase the density, but they are still working to get a price benefit

• The larger the capacity, the life span will increase (wear leveling) and, usually, the performance will increase (filling out the number of memory channels up to ~256GB, then its just the sheer amount of NAND to utilize)

• Verification and validation of the firmware is very improtant for reliability - Intel is the very best at this with their SSDs

For HDDs, with 4TB and below the differences will be in capacity, spindle speed (which directly impacts performance, noise, and heat), buffer size (primarily impacts performance), and quality assurance (for enterprise drives).  When you get into 5+TB drives, you start getting into more exotic implementations like shingled magnetic recording (which increases density, but adversely affects performance) and helium filled drives.

→ You will be the best judge of how much storage you will need or want.  It is highly recommended that you use an SSD as a boot drive to host your operating system, as well as programs and games that you use on a regular basis; so take a look at the required size needed for the programs you're going to load, take a guess at what you may need/want in the future, and calculate the necessary size.  For saving all of the media and other files you have where you don't need/want the performance of an SSD, HDDs still offer the best price-to-capacity for online (always available) storage.

Sound Card

Modern on-board (on the motherboard) sound cards are quite excellent these days and most recently they have begun implementing power filtering to the 5V power distribution to reduce power noise in the system, too.  There is not much reason to install any sound card in addition to the on-board chip.  The only scenario I can think of is if anyone wants a 1/4" headphone jack for high-end headphones, then you can maybe look at getting either of the ASUS Xonar Essence STX/ST sound cards.  There are also a number of USB DAC/amplifiers out there that will serve this function (albeit, outside of the case).

However, NO motherboard nor sound card will be able to decode Dolby TrueHD or DTS-Master Audio, which are the high resolution audio files on Blu-Ray movies.  Only A/V receivers and audio pre-processors can decode it and it must go through an HDMI connection.  This all has to do with security and there is no way around it.  If you want to experience the full soundtrack of your movies, than you must connect your computer via an HDMI to an A/V receiver or pre-processor that can decode the Dolby TrueHD and DTS-MA.  Future immersive audio formats like Dolby Atmos and Auro3D (IOSONO) will also necessitate the use of dedicated audio equipment, as well.

→ Bottom line for sound: skip the sound card and use the on-board audio.  If you want to get the absolute best, look for an A/V solution to hook your computer up to.

Power Supply

The are a few things to understand when figuring out what power supply to get.  A LOT of people out there in the enthusiast community will completely over-specify the wattage for a given system.  It is definitely much better to have more power than too little power; you can ruin your components with too little power.  But, if you go overboard: a) you are spending unnecessary money on a power supply that is too large and b) you will have the power supply operating in less efficient regions of its power band, wasting more power and putting more heat into your room.  A power supply is generally most efficient in region of about 50% to 70-80%.  There are significant drops in efficiency below 10-20% and above 90-95%.  So, when your computer is at idle, it is using significantly less power than its maximum draw.  Ideally, you'd like this to be above 20% of what the power supply is rated for.  A lot of manufacturers report the TDP - Typical Design Power.  While this isn't the absolute maximum power the component can draw, it is representative of what you should expect to be used for pretty much all situations (you need something that acts as a power virus to get to absolute maximum power draw, it is extremely rare for this condition to occur, let alone happening in multiple components simultaneously).  I'd suggest aiming to get the TDP in the 70-80% range to take advantage of the efficiencies there, account for the small degradation to power supplies over their life, and allow for the rare occurrences when a component does draw more than TDP. There is also SDP - Scenario Design Power, but this is used more in mobile applications for battery life calculations (but could be used informatively for figuring what your computer uses in lighter workloads).  Please keep in mind that overclocking will increase power draw, especially if voltage levels are being tweaked (If you care about power consumption, I'd advise you to get better components rather than overclocking as the performance-to-watt ratio will be better...overclocking allows you to get a little bit better performance for not too much cost, although lifetime costs can be higher as components will wear little faster...much faster if the voltage is cranked up).

→ I'd suggest using a power supply calculator (it would be nice if they showed what the expected idle power draw is, but no such luck) to determine how big a power supply is needed.  If you plan on adding another video card or extending your computer in other ways, you should take that into account, as well.  Warranty from the manufacturer is also a factor to consider when selecting a PSU.

For instance, I calculated what a loaded Velox would need and I selected: High end desktop motherboard, Intel (LGA 2011-3) i7-5960X (90% TDP) overclocked to 4200 MHz at 1.3V.  I selected 4 x 8GB DDR4 and 4 NVIDIA GTX 980 in SLI, 1 7200 RPM SATA, 1 Flash SSD, 1 BD-RE drive, 2 USB devices, a fan controller, a front bay card reader, 7 120mm fans, 1 pump and 1 relay in the custom water loop, a system load of 90%, and capacitor aging of 30%.  This came out to be 1206W, which is definitely reasonable for a system like this, so a 1200W PSU would work well for this application.  As noted on the PSU calculator site, the +12V power rail is really the most important factor, rather that the overall power output; you can locate the specifications for the PSU in question and there will be one or more +12V power rails on the PSU.  You want to make sure that large power components on can be handled on the power rail they are connected to (divide their rated power consumption by 12V to give the amperage).

**I will finish this up later**

Edit - also not sure why the bullets are formatting the text as white...??  I changed the text to dark grey, but the formatting and bullets are still missing.

Edited by  - 16 Apr 2015 at 11:36pm

.......

Looking good.        I would see if DS can put a sticky to it so it doesn't get lost with added posts. 
Or, maybe in the Support Guides but keeping it here is better.     

(moved up to main write-up)


Edited by  - 16 Apr 2015 at 11:38pm

I thought I should clarify. Fastest clock doesn't necessarily mean fastest processor, but if you are comparing CPUs with a similar architecture, it should usually come close to being true.

Originally posted by michaeljhuman I found this sentence confusing - "Before I get to far, I will note that there are two varieties of Core i7 processors. There are the lower end (and cheaper) consumer variety that fit on the same motherboards as the lower end processors; and there are the high end "enthusiast" processors that use a socket also used in the enterprise world." I don't look at it that way. I don't see it as a low/high end thing. In today's market, you are likely looking at two options. A perfectly good i5/i7 with an 1150 socket, OR more than 4 cores. It's not that the processors which use the 1150 are low end, unless you just mean cheaper by low end ( sometimes it's not clear.) But the processors which have more than 4 cores, e.g. the 5920, need a different socket So it comes down to whether a user is going to benefit from more cores and/or needs a server. A gamer, and not saying everyone here is a gamer, benefits the most from the fastest clock. A gamer who overclocks therefore benefits the most from a CPU they can overclock the most. I admit I don't know which CPU that is :)

I wasn't trying to demean the processors on the LGA 1150 socket, just note of the clear differences between the two and that the LGA 2011 processors are treated as a much higher-end component (and priced accordingly).  I did say that the LGA 1150-based processors are very capable for gaming and would probably be the better choice if you don't have other things that need and can utilize the additional functionality of the higher-end parts.

Originally posted by michaeljhuman I thought I should clarify. Fastest clock doesn't necessarily mean fastest processor, but if you are comparing CPUs with a similar architecture, it should usually come close to being true.

Yes, but the major gains are in select functions in the instruction set architecture extensions (like fused multiply-add or wider vectors), which must be taken advantage of in the software to see the difference.   Software usually lags these advancements for a while as they do not want to alienate their userbase that hasn't adopted the latest and greatest components.  The microarchitecture on Haswell and Haswell-E is the same, and there will be no advantages of one over the other there.

Originally posted by michaeljhuman In terms of sound card, a person could also buy an external audio interface. IMO, it's mainly useful if you work on music/external audio. For music work you need a low latency audio interface. I learned that the hard way when I tried to record using onboard sound and had many dropouts. Note that external audio interfaces typically have 1/4" outputs and/or XLR.

That is why I had throw in the exception for USB DACs as well as A/V receivers and audio pre-processors.  The better USB DACs renegotiate the USB protocol upon connection and use a very stable clock of their own to time everything.  You still need to have compatible media player software, however.