|How-To: Benchmark Your PC Like a Pro|
Putting a new gaming rig to the test is a rare-but-joyous task; like anything else, there's a certain intrinsic value to be had the first time a new system is burned-in for testing and functionality. Following-up with our "refurbished product burn-in" testing guide that we recommend for all used parts, this How-To guide for benchmarking your gaming PC will put you in a position to boast FPS -- whether on an optimized budget build or hardcore machine -- and troubleshoot for any problems that may arise. Find them early and often, then tweak for the future.
As this guide is intended to be a quintessential software toolkit for intermediate and new builders, we'll go over the basics of how to benchmark your PC and testing each major component, including memory, video cards, CPUs, and solid-state or storage drives. The ultimate goal is to create a guide that is representative of what our professional benchmarking team does here.
Benchmarking and system burn-ins have many practical and synthetic uses; rooting out problems, of course, is the most obvious one -- optimizing overclocking efforts, testing for stability, crunching numbers for tangible evaluations of PC performance, and checking for advertised performance are all very real uses for serious system benchmarking.
On the practical side, as mentioned briefly above, benchmarking serves many purposes that can potentially help save money, boost performance, check for advertised rates, and so on. By synthetically and realistically benchmarking new or used components and completed system builds, any potential issues can be unveiled and thoroughly squashed before warranty replacements, catastrophic failures, or generally 'bad things' happen.
There's an entire culture to system building and PC gaming, as we've discussed before, and benchmarking is a part of that. Being able to definitively classify a system by framerates or arbitrary points is a useful ability for comparative purposes, and can even lead to "arms races" between friends and forumites. This is all good. PC gaming is a culture. It's complex and is similar to that of an ecosystem; as such, benchmarking components can be the catalyst that leads to greater hardware knowledge, overclocking, and inclusion in communities that can help further the lifespan of each built system.
Sure, these are all practical as well, but there's something immediately different about them.
Once it's established that a bit of time should be spent on classifying an established or new system, the next logical step is to determine what, exactly, gets tested. There are about eight core components in a system (including an optical drive), but not all of them need abusive testing sequences. For the most part - at least, in a non-enterprise, gaming-focused PC - the most obvious components to test will look something like this (in order of most gaming-relevant to least):
The motherboard is tough to test - there aren't many comprehensive motherboard testing toolkits out there, but it's also not entirely necessary for the average gamer. Enthusiasts may want to install thermal probes to ensure any aftermarket modifications to heatsinks or bridges are functioning properly. Optical drives are not really necessary to test. If it doesn't read the disc, there might be a problem, but that's going to be the biggest indicator that something's wrong (and even then, there's not much worthwhile that can be done beyond returning it). The eight component is the case, which should be fairly straight-forward. The build process will 'test' this about as much as it needs to be, but definitely scout for airflow irregularities or other anomalies when the system is up-and-running.
Video card testing is the most immediately useful and visually noticeable -- there are very easy-to-read differences when a test goes well (or poorly) with graphics stress testing, just by nature of being visual. For gaming, of course, it's also very important to understand the capabilities and limitations of any GPU. Especially if overclocking is a consideration. Learn it before overclocking it.
Then there's the SSDs, CPU, PSU, and memory. SSD testing is straight-forward and primarily synthetic, but should give a good idea if software/firmware settings are appropriate (like AHCI-enabled, for example), the speeds of the drive, and the stability of the drive. CPU stressing can be mostly done through Prime95 (which will also test your RAM), but we'll talk about that below; testing a processor is, as with all components, advisable for stability scoping. Make sure it can stand on its own under pressure -- if not, analyze what went wrong, decide if failure was reasonable under the circumstances, then consider whether replacing it would fix the problem. Power Supply testing is really just a look at its load delivery and total functional capacity when the GPU is placed under max TDP. RAM testing is pretty clear-cut: If things are bad, you'll receive a failure message; if they're good, it says "OK." The main time memory testing would get more exhaustive is when voltages and/or frequencies are inaccurate or overclocking is involved.
Benchmarking is easy, but it's not necessary to survive, or anything like that. Anyone who has recently built a system, is troubleshooting a suspect component in a dysfunctional computer, has purchased refurbished equipment, or is trying to establish a tangible estimate of a gaming rig's capabilities (going forward) should be interested in benching their rig. If you're the type that likes to tinker, it's a fun task to undertake and can give very real numbers that are comparable (relatively) to the numbers that friends may have. This ultimately creates a healthier gaming environment, understanding of the PC, and building mentality.
I guess the easy answer is: For diagnosis, troubleshooting, reliability, and entertainment.
There are a lot of different words that can generally be distilled-down to mean "benchmarking," but each phrase typically has its own specific meaning. Let's drill through some of the system testing and reliability terminology that gets tossed around the most.
Burn-in Testing: Again, as discussed in our "how to burn-in your components" guide, burning a component in is the general process of getting it prepped for prolonged use. Hardware manufacturers do this on a grand level (Kingston explained their process to us recently) and put silicon chips under serious loads to determine what that specific chip should be "binned-out" as (or packaged as). The process of "burning-in" hardware can be seen as prolonged, extensive testing with the specific intent of checking for reliability and sustainability.
Stress-Testing: System testing features all walks of performance benching, and "stress-testing" is a part of that group. To "stress-test" a component is to put it under as much load as physically possible with the intent of determining a point-of-failure or other limitations. Stress-testing is normally the most successful in utilizing a component to its fullest potential.
Benchmarking: A more general term. Benchmarking often includes testing a component with a specific grouping of settings, hoping to optimize for usage and performance.
The hardest part about getting started is, simply, knowing where to start. Since we are focusing this article on the gaming aspect of systems, let's make things easy and start with video/graphics stress testing and benchmarking. We'll move around from there and sectionalize each component in a clearly-marked fashion.
Before diving into actual component analysis, though, it's important to get all the right tools setup and ready for use. It's even more important, of course, to understand the core concepts of what, exactly, defines a "good" result from a test and a "bad" result, otherwise everything turns into a game of internalized checking between relative results, and gives no hard data or comprehension thereof. We strongly encourage checking out some of our "dictionary" articles, which go into details on the specific terminology used for their respective components. Here are some links to those:
The tools themselves are primarily composed of pre-packaged software kits that do all the hard work for you (or games, in some instances, which have benchmarking sequences pre-routed), so the task of actually benching a rig is quite easy. The below list was taken from our Burn-in Guide (the one linked a few times, already) and adapted for the specific purposes of this article.
We'll go over various benchmarking and stress-testing/burn-in utilities for the different components below; many are free or have free versions that are quite useful, but a few of our suggestions may be paid utilities or games (Metro 2033 is an example -- it's a fantastic game for benchmarking).
It is key to mention the importance of logging throughout the entire process; linked below, MSI's Afterburner will do a great job of logging the stats of the GPU, including min/max/avg FPS, temperatures, and clock speeds. Logging is hugely important to the troubleshooting aspect of benchmarking, as it allows the tester to determine at what point(s)-in-time the suspect devices failed, their triggers, and then analyze if heat, instability of clocks, or other issues were the catalyst.
Make sure you log everything!
That said, here's the basic toolkit.
Video card testing is one of the most expansive areas of hardware benchmarking out there, luckily for us gamers. Tools like FutureMark give us relative numbers for comparative purposes, while Heaven, Furmark, and games can give more immediately understandable outputs.
Let's delve into some of the better options out there:
MSI Afterburner - MSI Afterburner is great for overclocking, but it's also great for logging (and EVGA users can enjoy EVGA PrecisionX). Go to Afterburner's settings and enable logging, then specify the preferred level of detail.
GPU-Z - Similar to CPU-Z, GPU-Z runs through everything in our GPU Dictionary and checks the numbers; it's up to you to ensure those numbers match the advertised specs. Use this when overclocking a GPU to ensure stability and reliability.
Heaven - Unigine's Heaven benchmarker is one of the best for stress testing in a pretty fashion. Run this alongside MSI Afterburner's logging for full analysis of temperature fluctuations and FPS or clock variance.
Furmark - Furmark's not quite as beautiful as Heaven, but it does what it wants to: Stresses the GPU.
3D Mark Vantage / 3D Mark 11 / 3D Mark 12- One of the best-known video-graphics testing applications out there, 3D Mark utilities are best known for their point-based ranking systems, which helps easily identify the maximum throughput of video cards based upon other publicly-ranked cards of equal or similar caliber. Unfortunately, the current iteration of the free version isn't quite as stressing as a number of other benchmark tools, so for best results, get the full version (or wait for the new one, which is forthcoming).
EVGA Precision - EVGA's video card read-out utility is one of the best available for owners of EVGA video cards -- with Android applications available for best portability, Precision helps tweak EVGA video cards to perfection.
(Game) Metro 2033 - Metro 2033 has proven to be one of the industry's most-abusive real-world graphics testers out there; its ambient occlusion, DirectX 11 usage, performance anti-aliasing, and complex shader utilization make it a true test for gaming cards.
Checking the specs of your CPU to ensure it's actually running at the advertised rates is important -- especially if you bought a refurbished laptop.
CPU-Z - Probably one of the more widely known utilities, CPU-Z interprets the specs of your CPU (this is also great for verifying overclocking), spits them out, and lets you decide if you like them.
Prime95 - As mentioned in the memory section above, Prime95 is a great stresser for CPUs.
(Game) Civilization V - Civilization V's graphical complexities are rapidly being lost in a sea of more recent games, but its streamlined multi-threaded processing stands as a powerful test-bed for gaming-grade CPUs.
(Game) Shogun 2: Total War - The Total War series has served our testers for years; with its capacity to render tens of thousands of units on the screen simultaneously -- a procedure that is split heavily between the CPU and GPU -- it acts as a fantastic real-world application test, since it can abuse the GPU and CPU at the same time.
The most note-worthy (and free) consumer memory burn-in testers can be found here.
Memtest86+ - We'll be detailing this one below; it's one of the most powerful and reliable memory testers, and can normally yield failure-or-success notifications with unbelievable accuracy and rapidity.
Prime95 - While this is primarily to be used for CPU punishment, Prime95 offers a 'torture test' that stresses both RAM and the CPU with several customization options. Its customizability and level of torture makes it a prime subject for extensive testing on refurbished systems, RAM, or CPUs.
BIOS - It sounds silly, but we always recommend a simple check of BIOS' system information screen to ensure your components are running at the frequencies advertised. The system information screen in BIOS is normally the most accurate, so check up on CPU speeds, memory frequency and voltage, and anything else that may be important to your specs.
Fortunately, there are several drive testing utilities of various power levels. We also recommend performing a full drive wipe before using someone else's recycled crap, since Windows doesn't actually delete files. We used all the below tools in our Kingston 3K SSD review, for those wanting examples of what they do.
ATTO - This simple utility will give you an idea of your drive's read/write capabilities within a user-defined range of file sizes and synthetic transfers. The speed of the transfer should positively correlate with file size.
AS-SSD - This one's a bit prettier and easier to read and is slightly more powerful than ATTO (but it's always good to check an SSD with multiple utilities to eliminate utility error). It has compression benchmarks, copy benchmarks, and a scoring system.
HD-Tune - My personal favorite drive tester and the most powerful of the lot. HD Tune will give you the most detailed information on the read/write speeds of your drives.
CrystalDiskMark - This one's probably the simplest of the bunch to use. Download it, run it, watch it go. Check the rates to make sure they correlate with your SSD's advertised speeds (if you have one) and if you have a spindle/magnetic drive, listen to check for clunking.
This one doesn't really belong somewhere special since it records all relevant hardware settings, but it is a very important utility to any enthusiast: CPUID's HWMonitor reads out the system's general health status, voltages, temperatures, RPMs, and performance in a single, easy-to-use interface. You should run this in addition to MSI Afterburner during any significant testing.
Continue to Page 2 for the actual benchmarking procedures.