Learn About Computer Hardware – Ultimate Guide 1
Things you need to know to become a PC hardware expert
Knowledge is power, and when it comes to PCs and computer hardware that’s especially true, because only by knowing how your PC components’ specs actually affect performance can you get the maximum power you need for the type of computing you do—and avoid being seduced by features that sound impressive on the box but won’t do squat to improve your experience.
Knowing your stuff has other benefits, too. An in-depth understanding of what makes all your parts tick enables you to better troubleshoot problems, upgrade in ways that make sense, and converse with other nerds in your own secret language. Continue reading to begin your crash course in PC spec-speak.
Just how many cores and how much cache do you need? We’ll help you answer those questions and others with cool confidence
There are two kinds of buyers: Those who will never upgrade a CPU and those who actively plan for it. For the former, even a CPU welded to the motherboard won’t matter, but upgraders who want to use a system for years need to pay attention to the socket, as it’s one of the primary factors limiting your upgrade options.
On Intel , there are three sockets to choose from: LGA2011 , LGA1155 , and the new LGA1150 . Of the three, LGA1155 has the least amount of life left in it, as it will be slowly phased out in favor of the new LGA1150 platform.
We know from Intel roadmaps that LGA1150 and LGA2011 are good for at least another couple of years. On AMD ,AM3+ offers a superb assortment, from budget dual-cores all the way to eight-core chips, with the company’s new Piledriver chip even slotting into this old socket. The company’s FM line isn’t quite as stable.
FM1 didn’t go very far, but the company’s FM2 looks like it might have longer legs. The thing is, FM2 processors—or rather, APUs—aren’t aimed at the type of user who upgrades every year.
We suspect that most FM2 buyers will use the platform for a couple years and then buy a new system instead of upgrading. For long-haulers, we recommend AM3+, LGA2011, and LGA1150. If you don’t care about doing an upgrade, go with whatever CPU you want.
Core count is the new clock speed. That’s because as consumers have been trained not to look at megahertz anymore as a defining factor, vendors have turned to core count as an emotional trigger. Two is better than one, four is better than two, and six is better than four.
Here’s the deal, though: More cores are indeed better—but only if you truly use them, and really only when compared within the same family of chips. For example, to assume that an eight-core AMD FX part is faster than a six-core Intel Core i7 part would be flat-out wrong.
Likewise, to assume that a PC with a six-core Intel Core i7 will be faster at gaming than a quad-core Core i7 is also likely wrong. To make things more complicated, Intel uses a virtual CPU technology called Hyper-Threading to push its CPUs. Some chips have it, some don’t.
So, how do you figure out what you want? First, look at your workloads. If you’re primarily a gamer who browses, does some photo editing, and word processing, we think the sweet spot is a quad-core chip. Those who encode video, model 3D, or use other multithreaded apps, or even many apps simultaneously, should consider getting as many cores as possible because you can never have enough for these workloads. A good bridge for folks who encode video only occasionally, though, is a quad-core chip with Hyper-Threading.
Your CPU choice should be based on your workload and not what you read about.
Remember the Megahertz Myth? It’s what we alluded to above. It arose from the understanding that clock speed didn’t matter, because a 2GHz Pentium 4 was barely faster, if at all, than a 1.6GHz Athlon XP. Years later, that generally remains true.
You really can’t say a 4.1GHz FX-8350 is going to smoke a 3.5GHz Core i7-3770K because in a hell of a lot of workloads the 3.5GHz Core i7 is going to dominate. Nevertheless, we have issues when someone dismisses megahertz outright as an important metric.
We don’t think it’s handy when looking at AMD vs. Intel, but when you’re looking within the same family, it’s very telling. A 3.5GHz Intel chip will indeed be faster than a 2.8GHz Intel chip. The same applies among AMD chips. So, consider clock speeds wisely.
When vendors start looking for ways to separate your cash from your pocket, clock speed and core count are their first line of attack. If those features don’t get you, we’ve noticed that the amount of cache is the next spec dangled in your face.
Choices these days run from 8MB to 3MB or less. First, you should know that in many cases, the chips themselves are often the same. When validating chips, AMD and Intel will weed out defective chips. If a chip has, say, 8MB of L2 cache and a bit of it is bad, it’s sold as a chip with 6MB of L2 cache, or 4MB of L2 cache. This isn’t always true, as some chips have the cache turned off or removed to save on building costs.
Does cache matter in performance? Yes and no. Let’s just say that a large cache rarely hinders performance, but you quickly get to diminishing returns, so for many apps, a chip with 8MB of L2 could offer the same performance as one with 3MB of L2. We’ve seen cache matter most in some bandwidth-sensitive tasks such as media encoding or compression, but for the most part, don’t sweat the difference between a chip with 4MB of L2 vs. one with one 3MB of L2.
Integrated graphics are likely one of the biggest advances in CPUs in the last few years. Yes, for gamers, a discrete graphics card is going to be faster 105 percent of the time, but for budget machines, ultra-thin notebooks, and all-in-ones, integrated graphics are usually all you get, and there’s a world of difference between them.
Generally, AMD’s integrated graphics chips lead the way over Intel’s older generation of Ivy Bridge and Sandy Bridge chips. It’s like, well, AMD is the Intel of integrated graphics and Intel is the AMD. Intel’s latest Haswell chips make it far more interesting, though, as the graphics performance has increased greatly. Then again, AMD has also recently released its new APUs with Radeon HD 7000 graphics. The spec that matters most on integrated graphics is the number of graphics execution units and clock speed. More EUs mean better performance, as does higher clock speeds.
When to Run Aftermarket Cooling
Let’s get it out in the open: Stock CPU coolers really aren’t as bad as people make them out to be. Sure, we all scoff at them, but the truth is that Intel and AMD spend considerable money on the design and certify them to work with their CPUs in all types of environments. For the vast majority of people, the stock cooler is just fine.
The Cooler Master Hyper 212 Evo is a low-cost, worthy upgrade over stock—if you need it.
But you’re not the vast majority of people. Sadly, today, if you can even open up the case, you’re an enthusiast. Sure, there are applications for the stock cooler, such as an HTPC or a small box that won’t be overclocked, but we like to think of the stock cooler as the minimum spec you should run. It’s fine, but it can be greatly improved upon.
Obviously, if you’re an overclocker, a beefier heatsink is a foregone conclusion, as heat is one of the worst enemies of a successful overclock. Swapping out the stock cooler for an aftermarket model is almost guaranteed to net higher or more stable overclocks than you can hit with the stock cooler.
Even if you don’t overclock, an aftermarket cooler can be a worthwhile addition. Since they can dissipate more heat than a stock cooler, and the fans are typically larger, the fan RPMs are usually lower, thus quieter.
Closed-loop liquid coolers are also a good option, as they require zero maintenance and the risk of a leak is extremely low. Liquid coolers are also quite affordable today and easily outstrip the vast majority of air coolers. One thing you’ll need to keep in mind is that closed-loop liquid coolers aren’t always the quietest option out there, though.
Knowing your way around a motherboard is a distinguishing characteristic of a PC nerd. Let us help orient you
The form factor of a motherboard is its physical dimensions. The most popular today is the 18-year-old ATX form factor. The two other popular sizes are the smaller microATX and Mini-ITX . Intel tried and failed to replace ATX with BTX .
Two additional form factors are the wider Extended-ATX and XL-ATX. XL-ATX is not an official spec but generally denotes a longer board to support more expansion slots. For an enthusiast, ATX will cover about 90 percent of your needs. Besides offering the most flexibility in expansion, it’s also where you the get the widest range of selection.
You can get budget all the way to the kitchen sink in ATX. MicroATX is usually reserved for budget boards, but there are a few high-end boards in this form factor these days. Mini-ITX is exciting, but the limited board space makes for few high-end options in this mini size.
As we said in our CPU write-up, your motherboard’s socket dictates all that the board will ever be. If, for example, you buy a discontinued socket such as LGA1156, your choice of CPU is greatly limited. The most modern sockets today are LGA1155, LGA1150, LGA2011 for Intel, and AM3+ and FM2 for AMD. For Intel, LGA2011 and LGA1150 have the longest legs. Though still useable, the sun is now setting on LGA1155 boards. AMD is actively supporting AM3+ and FM2, but there is talk of a new socket to replace FM2.
The chipset on a motherboard refers to the “core logic” and used to entail multiple chips doing several jobs. These days, the core-logic chipset is down to one or two chips, with much of the functionality moved into the CPU. Chipsets manage basic functions such as USB, PCIe, and SATA ports, and board makers throw on additional controllers to add even more functions.
You should pay special attention to the chipset if you’re looking for certain functionality, some of which is only possible on newer chipsets. The P67 chipset, for example, did not support Intel’s SSD caching, but the Z68 did. Current high-end chipsets from Intel include the Z77 , Z87 , X79 ; from AMD you have the A85X, 990X, and 990FX.
The vast majority of gamers never run more than one video card, but it’s always nice to know you have the option. AMD’s multicard solution is CrossFire for two boards, and CrossFireX for more than two. For its part, Nvidia has SLI for two-card setups, tri-SLI for three cards, and four-way SLI for four cards. We won’t judge the relative merits of each system, as this isn’t the place for it. Most boards that offer one, also offer the other, but don’t assume a CrossFire board will support SLI. Read the specs ahead of time if you plan to run multiple cards.
One of the main differences between a high-end board and a low-end board is the ports. High-end boards tend to have ports galore, with FireWire, additional USB 3.0, digital audio, eSATA, and Thunderbolt added on to convince you that board B is better than board C.
How many ports, and what type, do you need? That is something only you can answer. If you still run an older DV cam that needs FireWire, having the port on the board for “free” is always nice. Thunderbolt is also an incredibly cool, forward-looking feature, but is very pricey.
If you never use it, you will have paid for nothing. These days, we say eSATA and FireWire aren’t needed. What we want, mostly, is a ton of USB 3.0 ports. The ultimate board today might be one with nothing but USB 3.0 ports, if you ask us.
If you see a board with tons of those long PCIe slots, don’t assume they’re all hot. PCIe slots can be physically x16 in length (that means 16 lanes) but only x8 or x4 electrically (which means the data is limited to x4 or x8 bandwidth). Cheaper boards may even disable some onboard devices when run in multi-GPU modes, while pricier boards use additional chips to spread the available bandwidth around and keep the devices running.
AMD’s 990FX and Intel’s X79 don’t have the limited bandwidth of the Z77 or Z87 chipsets, so if you need lots of slots, you’ll want to opt for those chipsets. Unfortunately, Z77 and Z87 are where you find more PCIe 3.0 support. PCIe 3.0 doubles the effective bandwidth over PCIe 2.0, but it’s still not officially supported on X79, and only newer 990FX boards support it now. Confused? Our advice is that if you really need to run high-bandwidth add-in boards for video capture or RAID applications, ask the manufacturer what motherboards they have certified for it first.
There are degrees of enthusiast computing and motherboards to accommodate all scenarios.
This is a tiny segmented LED on the board that displays the POST code of the motherboard while booting. It may seem trivial, but POST LEDs are a godsend when things go sideways on a machine. If all other things were equal, we’d take a board with the POST LED over one without it.
A backup BIOS stores a duplicate BIOS on the motherboard that can be restored should the BIOS get corrupted. We think it’s a nice feature but a corrupt BIOS is pretty rare. Nevertheless, it’s probably better to have a backup BIOS and not need it than to need it and not have it.
Wireless, premium sound, fan controls, and headers galore are the special features board vendors use to hook you. You might dismiss them as unnecessary features, but so are the power windows and multi-speaker setup in your car. Certainly some extras aren’t needed, such as onboard Wi-Fi on a desktop box that will live on Ethernet, but fan control, such as Asus’s excellent FanXpert II, is worthwhile, as are premium audio circuits.
Budget vs. Premium: Is It Worth It?
In a given chipset family—say, Z77—it’s easy to find a motherboard costing $110 as well one running $379. Both use the same chipset, so are they the same? It depends.
If you intend to socket in a non-overclocked Core i7-3770K, run one GPU, and a sound card, you’d probably be hard-pressed to tell the difference, but don’t assume that premium boards are just a gimmick to rip you off. High-end motherboards aren’t just anodized a different color and slapped a higher price.
The $110 board will be pretty much a strippo option, with no multicard support, minimal ports and slots, and a design that’s not made for high overclocks. Yes, you might be able to overclock the budget board, but the voltage regulator modules and chipset cooling are likely to limit you. High-end overclocking boards are truly designed for the sport, with direct voltage readout hard points. And yes, fancy new technology such as Thunderbolt, additional USB 3.0, and SATA controllers cost more money. Even the software suite on the budget board will be pretty stripped down.
Still, the truth is that most of us will neither be overclocking with liquid nitrogen nor going ultra-budget. That’s why board vendors offer a dizzying array of selections between the rock-bottom and high-end. We think the $175 range gets you a pretty decent board, generally.
SSDs have a lot of complicated technology inside their waifish 2.5-inch shells, so follow along as we demystify it for you
The controller is the brains of the SSD , and what governs performance for the most part (along with the type of NAND flash used). The controller uses parallel channels to read and write data to the NAND, and also helps optimize the drive via the Trim command, as well as performing routine garbage collection.
Though some companies might license a third-party controller, they always use custom firmware that they have created in order to define the performance of the drive, so two SSDs that use the same controller will still have varying levels of performance in different workload scenarios. While the SSD world used to be somewhat ruled by the LSI SandForce controller, those days have long passed, and we are now seeing the rise of in-house controllers by companies like Samsung.
Over-provisioning is a spec you will rarely see explicitly mentioned on a product box, but its presence, or lack thereof, is evident by a drive’s capacity. Over-provisioning is simply space taken out of the drive’s general capacity and reserved for drive maintenance.
So if you see a drive with 256GB of capacity, there’s no space reserved, but a drive listed as 240GB has 16GB reserved for over-provisioning. In exchange for that space you get increased endurance, as it gives the SSD controller a lot of NAND flash to use for drive optimization and management. The provisioned NAND can be compared to a swap file used by a mechanical hard drive and operating system, in that it is space reserved to manage the files on the SSD.
All SSDs use this type of memory, as it’s non-volatile, meaning you can cut off power to it and the data remains in place (mid-data-transfer is another story, though). The opposite is DRAM , which is volatile, so once you shut down your PC, it is deleted.
There are several manufacturers of NAND flash, including ONFI/Micro, Samsung, Toshiba, and SanDisk, and all the SSD vendors use them, so while a Samsung SSD obviously uses Samsung NAND, so does the new Seagate SSD, for example, since Seagate doesn’t own a NAND fab. Corsair SSDs use Toshiba NAND, and so forth.
There’s no answer to the question of “who makes the best NAND?” as they all have varying performance characteristics, and it’s typically the controller and its firmware that play the biggest role in determining a drive’s performance. Good NAND with a crap controller equals crap, so keep that in mind when shopping for an SSD.
MLC, SLC, TLC NAND
All modern NAND flash is either SLC, MLC, or TLC, which stands for single-, multi-, and triple-level cell, which indicates how many values it can hold in a cell at one time. The most secure, and precise, is SLC, which holds a single value in each cell. Obviously, this is a bit inefficient, but also very accurate, and has high endurance, making SLC NAND ridiculously expensive, and not for consumers (it’s for enterprise).
Next up is MLC, which stands for multi-level cell, as each cell can hold two values at a time. MLC is used on the majority of SSDs you can buy, as it strikes a fine balance between cost and capacity. TLC flash, which stands for triple-level cell, holds—you guessed it—three values per cell, giving it the lowest endurance of any drive available, with the caveat that it still allows years of usage. Only the Samsung 840 and Intel 335 use TLC NAND flash; the rest of the consumer SSDs available today use MLC NAND.
Here we see the main components of an SSD: NAND flash, controller chip, DRAM, printed circuit board, and SATA connectors.
Even though SSDs are the cool kids, we still need hard drives for our “multimedia” collections. Here are all the terms you need to know to sound like a pro
Spindle speed is the rotational velocity of the platters expressed in rotations per minute (rpm). Faster spinning platters result in lower seek times and improved performance. The most common desktop drives spin at 7,200rpm, but there are also 5,400–5,900rpm desktop drives, which we recommend only for backup purposes given their reduced performance relative to a 7,200rpm drive. There are 10Krpm drives as well, but the rise of much-faster SSDs have largely made them irrelevant in today’s market.
Every hard drive stores data on platters made of glass alloy, with data retained on both sides that’s accessed by read and write heads hovering on each side of the platter. The number of platters is something to pay attention to when shopping for a drive, as it dictates area density, or how much data is stored per platter. Right now, 1TB is the maximum platter density available, and it offers improved performance compared to a 750GB platter, all other things being equal. Since the platter has more data on it, the read/write heads have to move around less to pick up data, so we’ve seen significantly improved performance from drives bearing these super-dense platters.
All hard drives have a bit of onboard memory referred to as cache, and the market has mostly settled on 64MB being the standard. The cache is used as a buffer, in that data is sent to it before being written to the disk. Whatever was last written or read will usually still be in the buffer should you need it again, so it improves performance by making recently accessed data available instantly.
This practice of fetching data from the onboard cache is referred to as “bursting” in benchmarks, but in practice it rarely happens, so don’t use this number to determine a drive’s overall performance. Spindle speed is a much better indicator of hard drive performance compared to cache size.
This stands for Native Command Queuing and is technology that helps the drive prioritize data requests so that it can process them in an efficient fashion. For example, if a drive receives a command to go all the way out to the outer perimeter to fetch some data, but then receives a request for data that is closer to its current location, with NCQ enabled, it would fetch the data in the order of closest bits to furthest bits, resulting in faster data transfer.
A drive without NCQ would simply fulfill the requests in the order received, which is highly inefficient. NCQ only shows significant gains in a heavily queued workload, however, which typically doesn’t exist for home users, but does occur on a web server or some other high-traffic application.
A hard drive uses magnets (lower left) to move the read/write heads (the pointy things), which are both above and below the data platters.
The Scoop on SSD Caching
We all want the speed of an SSD but with the price and capacity of a mechanical hard drive. Obviously that’s not possible. However, there is a middle ground, which is using a small SSD as a caching drive for a mechanical hard drive. This allows your most frequently used files (including your OS and boot files) to be cached to the SSD for fast access to them, while less frequently accessed files reside on your hard drive.
This actually works quite well in our testing, and to set one up you’ll need to either run it off your existing motherboard with any SSD you have lying around, or buy a caching SSD and use the included software to set up the caching array. For Intel users, Z68 and Z77 boards include caching support natively via Intel Smart Response Technology, but users of other chipsets will need to BYO to the party.
Next : Part – 2