The Benchmarking Process - How To Benchmark Your Computer
It's pretty easy to just dive into benchmarking a system with the tools laid-out above, but full component analysis requires a small bit of setup (mostly in the form of logging). Just a little bit, not a lot.
I'll assume that most of you will be wanting to run a standard "gaming benchmark," looking to test out your newly-built gaming rig and get some hard data on its performance. For anyone testing things other than gaming performance, leave a comment and I'll help you tweak the test sequence.
Setup: Video-Graphics Gaming Benchmark
We'll start with installing AFI Afterburner (found in our toolkit section) and configuring it to maintain logs of graphics card performance. With Afterburner installed, launch the program and do the following:
- Click Settings.
- Click the Monitoring tab.
- Set the polling period to 1000ms (1s).
- Check "Log History to File."
- Check "Stop Logging when Log File Exceeds" and set it to 10MB.
- Download and install Heaven by Unigine (in our toolkit section).
With that setup, the testing can begin! Keep in mind that some games have built-in FPS graphs or FPS overlays (often enabled through console commands), which can sometimes be more accurate than third-party applications; you may want to look into the capabilities of your games in this regard.
Video-Graphics Bench: Understanding The Results
The hardest part of any testing is understanding the results -- we can pool all the data on the planet, but without an understanding of what's considered good, bad, and in between, it's impossible to make any calls on hardware performance.
It all depends on the video card, so while I won't be able to give specific numbers-to-cards guidelines, I can give rudimentary information so that you can make your own decisions -- which is, really, what we're here for at the end of the day. For specific performance questions, please post in the comments and ask what we think should be normal.
A typical MSI Afterburner log will spit something out that looks like this:
Let's first define what results we need to be concerned with:
FPS - Frames Per Second: Many of you know this already, certainly, but just in case -- FPS is the measurement of performance by frames rendered to the screen per second. Anything beyond 60FPS is fairly unnoticeable (pro StarCraft 2 players will occasionally claim they can tell the difference, but for those of who are humans, it's unlikely). 30FPS is the magic limit (film is often shot at either 24 or ~30FPS) of what we mortals are comfortable playing with, so that should be your target minimum. These are often measured in minimum, average, and maximum FPS.
Temperature: Afterburner will also log your GPU's temperature to a file, with each temperature tick corresponding with the fan tachometer (RPM), FPS, GPU usage, and clock speeds for that same second. Acceptable GPU temperatures vary based on the platform, but try to keep things below 85-95C if possible; that's quite warm and teeters on the edge of 'dangerously hot' for the video card. Most GPUs should sit around 60-75C for an average gaming session, but overclocking or particularly abusive games will cause that to vary. Most video cards or motherboards will have safety triggers built-in that force a shutdown in the event of potentially overheating components. The temperature should remain stable or slow in its progression as fan speed increases or stabilizes.
GPU Usage: The percentage of load currently placed on the graphics processing unit. This can be used as an indicator to scope whether the GPU can handle the application at its current settings, as it gauges 'how hard' the GPU is working to render and process the commands sent to it by the software and CPU.
Fan Speed: The percentage of the GPU fan's maximum speed (in RPM).
Fan Tachometer: The fan's speed in raw RPM. Modern GPUs will automatically increment their fan's speed to help stabilize rising temperatures; faster speeds don't necessarily mean the component will actually drop in temperature, but they usually help the GPU to maintain control of its temperatures within a peak range of efficiency.
Clocks: The Core, Shader, and Memory clocks should only fluctuate if you have told your GPU (or if it does it automatically) to increase voltage/clock speeds as load increases. You can read more about what the core clock, shader clock, and memory clocks do in our GPU Dictionary.
Memory Usage: This unit, on the other hand, will fluctuate based on application performance requirements. Video cards have a set amount of memory on the chip that can be accessed by applications; as a game or tool polls for more memory, the GPU grants it to the application and can increase its relative "speed" and the amount of objects/data rendered to the screen.
For FPS counts, you'll want to be in the range of 25-45 for best the playability-to-graphics output that your card can offer; anything lower will be noticeably laggy during gameplay, anything higher will be - to the vast majority of users - completely unnoticeable and effectively be trimmed as 'invisible FPS,' so to speak. The most important factor is consistency of frames (v-sync can help with this, but will also impose an upper limit of FPS that is confined to the monitor's refresh rate - normally 60 or 120Hz).
Video Testing: Running Gaming Benchmarks
Every thorough stress test will include synthetic and real-world testing. Synthetic tests, as the name sort of implies, are generated purely through interfaces that are designed specifically to abuse hardware to its fullest potential; in this regard, they are not entirely realistic and are not necessarily demonstrative of real-world scenarios. Synthetic tests are often the most useful in determining the maximum capacity of hardware when under programmatically-optimized testing; we can use these tests to define the upper-limit and lower-limit of the system's hardware.
With Afterburner running in the background and with logging enabled, it's time to kick that GPU into gear with Heaven, one of the prettiest-looking benchmarkers available (and free).
By default, the settings of Heaven should look something like the above. You'll notice that its options scale greatly beyond these meager settings, though, and can eventually look more like this:
Don't get too aggressive, though. Start small. Go with the initial settings, launch the program, play around with the settings and explore the world a bit, then meander your way up to the "benchmark" button in the top left. Once started, Heaven will go through its motions and perform 26 different stages of testing, all the while logging its results internally (and with Afterburner extracting its own results); look over the numbers afterward. Use our above 'understanding' section as a guideline for what's acceptable.
Once Heaven goes through its first pass and has reasonable (read: 30FPS or higher) framerates, consider running the program a second time with higher settings. Repeat this process until the 'borderline' between a playable and unplayable framerate, then mark those settings somewhere. This gives an idea of the card's capacity.
If the card is found faltering unacceptably or to have significantly higher temperatures than average, perhaps it's time to consult our hardware warranty guide.
Metro 2033 has been used for benchmarking competitions (whether for fun, like this OCN one, or for work) - its complex graphics (motion blur, sub-scattering, shadow filtering, volumetric texturing, and much more) are the perfect test for real-world gaming scenarios. This test requires a copy of Metro 2033 and installing its official benchmarking tool.
Within reason, set the graphics to something like this:
- DirectX 11
- 1920x1080 (or 1680 x 1050) resolution
- Very High quality
- AAA Anti-aliasing
- AF 16x
- Use the "frontline" scene and set it to run for one pass.
Ensure MSI Afterburner is still running and logging. Keep the graph of MSI Afterburner open and available so you can screenshot it post-bench.
Click "Run Benchmark" and sit back and watch. Once the benchmark is over, screenshot your MSI Afterburner page (it changes, so do this fast) and save the accompanying Metro 2033 results. They should look something like Majin SSJ's screenshot, who did a fantastic job of organizing his results on the OCN forums:
Note: You may see an improvement as more and more passes are run with preset benchmarking settings. The program will be stored to memory and perform better with each pass of the benchmark, in theory, so a more "true" test will be the first pass.
Since we've covered Furmark extensively in the past, this section will be lifted from our hardware burn-in guide.
FurMark is one of the best burn-in testers out there for GPUs, and while Heaven is fantastic, FurMark is spec'd a bit more heavily toward the stress-testing side. FurMark is for serious benchmarking - we highly recommend running it to abuse any new or used GPU you pick up. Download Furmark here.
Be careful with this tool - it can be deadly to GPUs if handled carelessly.
Launch FurMark once it's downloaded. Enable the Xtreme burn-in option and Burn-in option; click "Settings" and enabled the GPU Temperature Alarm. Set your max GPU temperature to something reasonable. If you want to be safe, set it to something around 80C. More experienced users should be able to deduce the maximum temperature if it is higher than this.
Set the benchmark duration as necessary. I always recommend running a short benchmark first (you can abort at any time) and manually monitoring temperatures for a few minutes before deciding on extended test settings. Check the "Log GPU temperature" option while you're in here.
Let's start simple: Run the preset 1080 benchmark (this is one of the least abusive ones) and personally monitor it until completion or boredom sets in. If things are looking good and the temperatures aren't crazy, go for the burn-in benchmark, get some food, and return in fifteen minutes. Overnight tests can be run with either manual settings or the "BURN-IN test" button. Just don't overdo it. Always ask us in the comments for recommended settings if you're unsure.
If no catastrophic failures have occurred after all of this synthetic testing, we can hope that the product is stable and will not fail any time in the near future. The hope is, to re-iterate, that we initiate failures in the first 24-72 hours, rather than 31 days from purchase. A reliable GPU will not be damaged by this testing - it's just like playing graphics-intensive games (again, just don't overdo the settings - if the FPS is low and temperatures are high, that's a sign to stop and lower the settings).
CPU Testing & Benchmarking
CPU-Z, as in the toolkit page, is the first thing to get installed for CPU testing and understanding. Check that the specs match up as they should. The items to be most concerned with -- especially if any aftermarket modifications or overclocking have taken place -- would be: Core Voltage, Core Speed, and the Multiplier.
Let's start with gaming benchmarks and then move to Prime95 for synthetic testing.
Civilization V & Shogun 2
Civilization V's AI is extremely complex -- sure, its texture decompression algorithms make for interesting DirectCompute results from video cards, but the speed of the AI is a great benchmark for CPUs. Similarly, Shogun 2: Total War has massive amounts of units on the screen at any given time, tracks thousands of projectiles, dead units, individual damages, and more, and thus beats up on the CPU fairly harshly. Follow the same steps below for Shogun 2, with the only variance being in the actual benchmark that gets launched (obviously, you'd be launching the Shogun 2 benchmark, not the Civ V one - opt for Shogun's 1080 bench).
Fraps can be used to record framerate (using Fraps' built-in free bench utility), otherwise, the native Civ V benchmarks will provide a score at the end of testing. For best use, look up your components afterward to see if the score received (or FPS, using Fraps) seems to match with that others on similar hardware have received.
Start CPU-Z, HWMonitor, run Afterburner just in case, and keep the Windows Task Manager open to the 'performance' tab for the best spread of numbers. Keep in mind that the task manager will read-out numbers for the entire system, not just Civ V.
Start up Civilization V and venture to the graphics / video options. Set the screen resolution to something appropriate (probably 1920x1080) and boost up everything else to reasonable levels. Start low if necessary.
Launch Civ V's native benchmark, hit the hotkey to enable Fraps' benchmark, let it run through, record the score it spits out, then check the FPS rates that Fraps gives and the CPU usage (check the threaded utilization) that Task Manager shows. For those using Turbo Boost-enabled Intel CPUs, CPU-Z should show fluctuation as Turbo Boost kicks into gear.
This program has been around for a number of years and has certainly seen its share of use in enterprise and corporate environments, but as hardware becomes more widely accessible by the general gaming public, it's found its place in gaming analysis and benchmarking, too.
Prime95 tests the CPU and memory exhaustively and installs with three presets for testing -- for CPU stressing, we'll use Small FFTs (Fast Fourier Transformation, a complex mathematical approach to computationally pressuring the CPU) and Large FFTs for now. The blend test is great to throw in if time permits - and it's also a good way to check up on the RAM.
Large FFTs should stress the CPU the most, you should see a screen that looks like this when P95 is launched:
The Large FFTs test will take a bit longer to run, but should give the best stability results.
Initiating the test will introduce a screen that looks something like this:
Let the test run through - if any anomalies are spotted or crashing occurs, you may have a problem with instability or unstable cores. Try running the test on fewer cores to see if that helps (additionally, unlocked cores may cause this problem - try disabling the unlocked core temporarily).
SSD/HDD Reliability Benchmarking
SSDs have to be tested almost exclusively with synthetic benchmarks. It's possible, of course, to test with real-world data transfer scenarios and actively time game load times (which we did in our Kingston HyperX 3K Review), but synthetic tests will be more numerically conclusive.
So, aside from getting out a stopwatch and checking your load times against an HDD, let's get to the actual benchmarking procedure for solid-state drives, starting with HD-Tune and AS-SSD, borrowed again from our hardware burn-in guide:
We don't need to run all the testing tools for drives - a select few will get the job done admirably. HD Tune is one of the best options out there (and has a pro version, which is quite extensive) and will do what we need done; go grab HD Tune before we start (the free version is fine).
Launch HD Tune. Everything should already be good to go for testing -- select the proper drive in the top drop-down menu and hit "Start" for the read test. Let it run until completion. Once finalized, you should be presented with a graph that looks something like this:
Now, of course, the numbers of each drive will vary based on its abilities, but that's the general idea. If a ludicrous amount of spiking is present or if the drive fails to complete, you may have a problem. The other testers will verify this for us.
Despite its name, this little program can be applied to both HDDs and SSDs for testing. AS-SSD offers a few different synthetic benchmarks (gaming, compression, and copy) that will give an idea of your drive's performance. Download AS-SSD for free before we begin.
The program will detect your drives at first launch - go ahead and wait for it to do so. Once open, select your drive in the top drop-down box. We'll proceed to run a few quick tests on it, in addition to the above HD Tune test. Check to make sure no errors are displayed in the top-left mini-box; if "PCIIDE - BAD" is displayed, you may need to change your BIOS settings for your drive from IDE to AHCI. For the first iteration, go ahead and hit "Start" - the default test will do well enough. Double-check those numbers against the advertised rates, if it's an SSD, and make sure no odd noises are emitted by the drive.
For our second test, go to Tools -> Copy Benchmark, check all three options, and hit "Start" again. As before, make sure speeds are reasonable (ask us below if you require assistance) and ensure no clunking noises are produced.
The final test is just as easy. Return to the main window, go to Tools -> Compression Benchmark, and start it as-is.
If your system is still running after these, you should be in good shape to continue on.
Conclusion & Results
What's the point of all this? To re-iterate the intro paragraphs, there are a lot of reasons to benchmark - both practical and cultural. Sharing results and system specs is a fun way to learn more about computers, tweaking, and how to achieve the best results for the least (or most, in some cases) amount of money. Working off of one of our PC Builds as a template, following the above steps should help lead to a personally-tweaked (on a software- and hardware-level) system that performs admirably for your specific purposes. That's half the point of all of this.
The rest of it is the culture of hardware enthusiasts and PC gaming. There's a level of respect that is shared for builders who take the time to really hammer-out the machine they'll be working on for several years; the gaming rig is the lifeblood of PC gamers, so optimizing its performance to fit personal gaming preferences just makes sense.
Hopefully the results are easy enough to understand; for the most part, it's all comparative. Matching your video card or SSD's results up against other cards or SSDs is the best way to get a grip on the relative power of components. If you have any questions at all - however specific or vague - please feel free to ask in the comments below or post on our forums.
-Steve "Lelldorianx" Burke.