Maximum PC benchmarking RSS Feed

Antec Skeleton

Nathan Edwards — Tue, 03 Feb 2009 10:00:00 -0600

We have to give Antec points for bucking the mainstream: The Skeleton is a seriously cool-looking case. The motherboard rests horizontally, so the case’s footprint is much wider than that of a standard tower chassis, but it’s also shorter. The open design and sliding component tray make it easy to swap parts in and out, and installation is quick and relatively painless. The Super Big Boy LED fan, which truly is both super and big, dominates the top of the case, keeping air moving over all your components, and the open design ensures that there’s plenty of airflow. A smaller fan mounts in front of the hard drive bay for additional cooling.

Removing two spring-loaded thumbscrews lets you slide out the component tray about three quarters of the way, providing easy access to the motherboard as well as the optical and hard drives. The side panels remove easily, and there’s a sliding bay for the power supply. But screwless this case ain’t. Your Phillips head screwdriver will be a constant companion.

The open design offers no protection against liquids, pets, children, or Chex Mix. The squat proportions also work against it—there are only two 5.25-inch and two 3.5-inch bays under the motherboard, so any additional hard drives need to be mounted outside of the case, for which Antec provides four brackets. And given the open-frame design, there aren’t many solutions for elegant cable routing. We had to do some finessing to connect all our parts.

Depending on your motherboard’s layout, you may have issues with the front-panel connectors; the HD Audio ports are in the rear of our EVGA 680i board, and the Skeleton’s front-panel audio wires barely reached them. And watch out for your CPU cooler—anything more than about four inches tall will bump up against the fan cowl and prevent you from moving the tray in and out, so avoid any of the monstrously huge coolers that are so popular these days. Stock fans will work fine, and fortunately, so will our current favorite cooler, the Thermaltake DuOrb. Due to the space constraints, any water-cooling setup will have to use an external reservoir—since the Skeleton has only two 5.25-inch bays, it’s impossible to mount one internally.

Still, we love the design and portability of the Skeleton. It’s especially useful for testers and other people who, like us, swap hardware frequently. If you have plenty of desk space and don’t need to worry about errant liquids or hairballs, the Skeleton could be the bold, new case you’ve been looking for.

Windows 7's WEI Gets a Boost

Mark Edward Soper — Fri, 23 Jan 2009 18:24:46 -0600

As noted by Gizmodo, Windows 7 has made quite a few tweaks to the Windows Experience Index (WEI) first introduced by Windows Vista. For those of you tuning in late, the WEI tests hardware performance of five subsystems (processor, memory, desktop graphics, 3D gaming graphics, and hard disk), calculates a score for each one, and uses the lowest subsystem score as your WEI base score.

Since just after Windows Vista shipped, users of high-performance components, especially graphics cards, have been complaining loudly about Vista's WEI top score being capped at 5.9. While the Minpaso database of Vista WEI scores calculates a "presumption score" to try to make allowances for today's faster hardware, there hasn't been an official move from Microsoft until now. The code jockeys in Redmond heard you, and the top WEI subsystem and base score in Windows 7 is 7.9.

Windows Vista's WEI scores reflect mid-2006 technologies, and a lot has happened since then. As discussed by our friends at the Engineering Windows 7 blog, rankings in the 6.0-7.9 range are designed to handle these new technologies:

As an example for gaming users, we expect systems with gaming graphics scores in the 6.0 to 6.9 range to support DX10 graphics and deliver good frames rates at typical screen resolutions (like 40-50 frames per second at 1280x1024).

In fact, you must also have a WDDM 1.1 driver and DirectX 10 GPU support or your graphics score is capped at 5.9.

That makes sense, but what has a lot of Windows 7 users who previously ran Windows Vista confused is the difference in how Windows 7 tests hard disk performance. Don't be surprised to see lower WEI hard disk subsystem scores on an identical system if you move from Windows Vista to Windows 7. Windows 7's WEI subsystem tests are designed to more accurately test how drives react to mixtures of typical operation than the tests used in Windows Vista's WEI.

As you might expect with a beta, there are still a few hiccups in the Windows 7 WEI, particularly in the hard disk test. As comments at the Engineering Windows 7 blog post point out, some hard disks get higher WEI subsystem scores if write caching is disabled (which should actually lower performance).

So, if you're running Windows 7 on the same hardware you used for Vista, how do your WEI subsystem and base scores compare? Hit Comment and tell us.

A-Data Sets New SuperPi Benchmarking Record

Paul Lilly — Fri, 23 Jan 2009 11:09:45 -0600

Last year it was Biostar -- and not Asus, DFI, or Gigabyte -- who set a frontside bus world record with its Biostar TPower I45 motherboard, and further blurring the lines between traditional enthusiast branding and companies better known for taking the budget end of the spectrum, A-Data -- not OCZ, Corsair, or Kingston -- has just broken a benchmarking record of its own.

"A-DATA® Technology Co., Ltd., a worldwide leading manufacturer in high performance memory products, announced today that its XPG™ DDR3 memory modules have broke a new world record on SuperPi 32m," A-Data stated in a press release. "The record was set by utilizing the DFI Lanparty UT X58 motherboard and XPG X Series v2.0 memory, the DDR3-2133X v2.0 2GBx3 triple-channel kit."

The new record now sits at 6min 40sec 360ms, which required overclocking A-Data's triple-channel DDR3-2133X v2.0 kit to 2237MHz with 8-7-7-21 latencies. A-Data didn't say how much voltage it took to reach that frequency, but if we had to guess, we'd say it ran high. The same kit comes rated at 2.05V-2.15V with 10-10-10-30 latencies at its stock frequency.

Image Credit: A-Data

How We Test: Displays

David Murphy — Tue, 26 Aug 2008 12:54:21 -0500

Testing a display is two parts science, one part art. It's difficult to measure the performance of a display the same way Maximum PC evaluates other products. There is no benchmark that we can just fire up and then report a score from. Nor can we even test displays in their out-of-box condition. A fair amount of tweaking and visual analysis is necessary to ensure fair display benchmarking. And at the end of the day, determining which monitor reigns supreme is a mix of qualitative testing and the editor's expertise.

That said, benchmarking displays isn't as simple as firing up a DVD and seeing which one looks better. When we pull a display out of its packaging, we hook it up to our standard test bed (Evga 680i motherboard, Evga GeForce 8800GTX, QX6700 processor, 2GB RAM) alongside another display. Comparing a new monitor against the best-in-class allows us to make judgments about the display's prowess based on the flaws that we observe. It's much more accurate to look at identical pictures across two screens than to run our suite of display evaluations on two displays sequentially and try to spot the differences from memory.

We hook both displays up to our videocard using identical dual-link DVI cables. We're running our benchmarks on Windows XP SP3, as we've found much more success cloning the picture for our multimedia tests on this OS versus Windows Vista. Both displays are each allowed an hour of warm-up prior to our initial round of tweaking.

Synthetic

We start our evaluations by firing up a batch of DisplayMate tests in an attempt to set the monitor's optimum brightness and contrast values. We look at a number of grayscales on the dark and light ends of the spectrum to discern the point at which adjusting a display's brightness or contrast stops improving the image. This is a tricky balance, as improving the grayscale range on one end of the spectrum--by adjusting the contrast, for example--can often poison or reduce the range on the other.

Once we've achieved an ideal balance, we begin running through the rest of the DisplayMate tests. On the color saturation test (shown below), we run through a number of hues to discern the display's ability to produce very light shades of color arrayed against a white background. A good monitor will be able to produce color all the way down to a two-percent level, whereas weaker displays can look whited-out at four or six-percent.

We then run through a series of gradients that help us discern a number of particulars about a display. Banding and color-tracking issues appear best in grayscales. The former refers to the dark bars of compression that appear in a display's smooth gradient; the latter refers to unnatural coloration that tints part of a black-to-white gradient.

We note these when they occur, including the exact number of gray values (128, 256, et cetera) that are on the screen at the first instance of either of these issues. The grayscale test as a whole also allows us to take a big-picture look at a display's overall range. For example, displays that are unable to reproduce a wide range of values at the dark end of the spectrum will show a huge black bar (or square) in the middle of our test patterns.

Note that you're looking at the same pattern of grayscales in the above pictures. The top image uses 256 shades of gray to create the black-to-white fade. The lower image uses fewer. It's a serious issue when color-tracking and banding issues appear in gradients with few gray values.

The two above tests are identical in scope: each present the display's grayscale performance using two different gradient patterns.

We also use a DisplayMate pattern to check a monitor's streaking. We haven't seen this issue pop up in recent displays we've tested. Nevertheless, streaking is what its name implies: when an image appears to have an aftereffect, or a "tail," as a result of a switch between extreme contrast values. On the test pattern below, this would show up as a horizontal block--of an opposite color--extending from the white or black bars into the gray right half of the screen.

Real-World

We run through a number of real-world test scenarios to discern just how the flaws (or successes) of a monitor's DisplayMate performance match up with typical use. Our testing is split into three categories: photography, movies, and gaming.

In terms of photos, we run through a number of examples to evaluate a display's grayscale performance, coloration, and technical flaws. We check to see if skintones look too red or washed out. We check to see just how badly a display's banding or color-tracking will affect an image's gradients. We check to make sure that pictures we've shot look true to form as a general mix of coloration and contrast. We look to see what details get mashed together as a result of a display's poorer grayscale range.

We use these same questions on our movie and gaming tests. We also use these different forms of media to test a display's preset options. While we tweak our displays to perfection, we understand that not everyone is going to be as interested in fooling with a display's specific parameters. This is an important part of our evaluation, as we've been known to discover presets that actually ruin a display's picture quality. Also, movies--specifically, V for Vendetta--allow us to test fancier elements like a display's dynamic contrast feature. Likewise with our Bioshock gaming test, a scene of which is shown below. In this case, a display's deficiencies in coloration and grayscale reproduction will become apparent in its inability to produce detailed smoke or vibrant flames.

And there you have it! It sounds like a small list, but an average display review takes at least an hour or two to finish. We carefully analyze all the subtleties of a display, as oftentimes, the smallest differences will separate different models. Although it's not a benchmark per se, we also factor in a display's ergonomics and connection options into the final verdict mix. While it's easy to tell a 9-Kick Ass monitor from a 4 verdict, it's not quite as easy to fill the gaps in between. That's why we spend a great deal of benchmarking time across a variety of tasks--to ensure that the monitors we recommend will not fail you in any way you choose to use them.

How To: Benchmark Your PC without Breaking the Bank

David Murphy — Fri, 08 Feb 2008 18:12:56 -0600

We run benchmarks at Maximum PC because we have to; there’s no other way to determine the minute differences between systems without a repeatable standard of comparison. But you don’t have to be a reviewer to run a benchmark; in fact, regular benchmarking can give you valuable insight into the status of your system. For example, benchmarks are the best way to decipher whether the various performance-enhancing applications you’re running on your PC actually do anything or whether that latest batch of drivers hurt your gaming performance more than it helped.

The Maximum PC suite of benchmarks costs upwards of $1,000—a bit out of the price range of users who just want to see if their machines are up to snuff. But there are cheaper (and by that we mean free!) alternatives; we’ll show you how you can use them to test your rig in the comfort of your own home.

Time: 22 hours

What You Need

A PC
Cinebench
www.maxon.net
3DMark05
www.futuremark.com
Call of Juarez DirectX 10 Demo
http://tinyurl.com/2g6dhr
HD Tach
www.simplisoftware.com
Cosbi OpenSourceMark
www.sourceforge.net/projects/opensourcemark
Prime95
www.mersenne.org
Patience

Score Your CPU

We scoured the Internet and racked our brains to find the most appropriate (and most free) CPU test for your machine. And trust us, it wasn’t easy. Whittling down the list of options to just those that are free was difficult enough—there’s not much out there that will cost you absolutely nothing.

After picking through that small pile of programs, we discovered an important corollary that bears repeating: Just because a program claims to be a CPU test doesn’t necessarily mean the score it generates is a proper reflection of your CPU’s performance, particularly if you’re running a multicore machine. (Single-core users have a bit more leeway with their CPU benchmark choices, as anything that taxes the CPU is going to hit your one, lonely core.) The surest way to test a benchmark’s effectiveness is to pull up the Windows task manager while running a given CPU analyzer. A true CPU test will completely maximize the usage of all your cores.

Cinebench’s built-in database keeps track of all of your benchmark runs. Label everything correctly so you don’t forget what changes you’re testing!

Grab Cinebench and you’ll be pleasantly delighted by its absurd ease of use and applicable testing environment. The program runs on everything from single-core to 16-core machines. It’s a wonderfully future-proof little benchmark that gives you an overall performance score based on your computer’s ability to render a 3D image in as little time as possible. You can even record your results to a built-in database, a helpful way to keep track of your scores when modifying your rig. If you’re suffering any CPU performance loss as a result of your tweaking, Cinebench will let you know.

Test Your DirectX 9 Performance

One of the surest ways to test your videocard’s DirectX 9 performance is to—you guessed it—fire up a graphics-heavy game that includes a benchmark mode (like the FEAR benchmark we use in our Lab) and let ’er rip. But not every game tests your graphics card’s performance. There’s a reason we use Quake 4 and FEAR for our official benchmark runs: The former is an OpenGL-based game that’s far more dependent on your CPU than your videocard, whereas the latter is a better demonstration of GPU-based prowess.

If you have no acceptable games to test your rig’s performance, the next best thing is a free solution from Futuremark. Head over to the site and grab yourself the demo of 3DMark05. You might be tempted to download a later version for upgradeability’s sake—don’t. We’ve found that 3DMark05 pushes your graphics card more than later versions, which test the CPU a bit more.

The 3DMark05 official score throws your CPU into the mix, but you can get adequate FPS results from the app’s graphics-only tests.

Since the program’s a demo, you won’t get to edit any settings—you can’t adjust antialiasing, the resolution, or anything else. However, 3DMark05 will scale depending on the power of your graphics card, and there are numerous websites and forums you can visit to compare your demo score to the scores other rigs achieve!

Test Your DirectX 10 Performance

So you’ve plunked down big bucks for that fancy DirectX 10 card and you’re curious whether all the different drivers, tweaks, and overclocks have had any effect. The best free benchmark we’ve found is a DirectX 10 demo from Call of Juarez. It runs through a series of in-engine settings that test everything from particle effects to HDR antialiasing to shadows.

To squeeze more frames out of your DX10 card, reduce antialiasing. Your images will get a little jaggier, but you’ll see frame rates rise.

As with any benchmark, you’ll want to run multiple iterations of the graphical test to account for any errors or extraneous factors during the run. That said, the scores should be consistent, if not identical, across all three runs. If they aren’t, double-check to see if there’s anything eating up your computer’s resources in the background!

Benchmark Your Hard Drive

If you’re looking for the source of slowdowns in your system’s storage performance, the free HD Tach benchmarking utility is a must-have. With one click of a button, the application tests burst speeds, CPU utilization, random access speeds, and sequential read speeds.

The program gives you a ton of numbers once it’s finished. The most important of these is the average read speed of your drive—it takes less time to pull data from the inside layer of a platter than the outer, hence the “average” in the calculation. On the whole, this number is a good measure of your drive’s general performance.

If you have two identical hard drives in your PC, a large disparity in benchmark results could indicate a faulty drive. Back up now!

HD Tach’s burst speed measurement represents your drive’s ability to transfer data from its onboard cache to your CPU. Higher numbers indicate faster file transfers. The random access measurement indicates the time it takes the drive to access a random sampling of data from all over the drive. In this case, a lower number is better.

There’s not much you can do to improve the performance of a subpar drive. Check your BIOS to make sure you’re running at the fastest interface speed possible—SATA 3.0 instead of SATA 1.5, for example. Defragmenting the drive might help, but performance degradation over the life of a drive might indicate hardware failure.

Measure Your Overall System Performance

The open-source program COSBI OpenSourceMark attempts to replicate real-world benchmark scripts, similar to SysMark’s and PCMark’s. We’ve found that OpenSourceMark, which uses a number of real-world operations, is one of the better ways to analyze your computer. Install the program and click the “official run” button to start the tests—which include file compression, audio encoding, spreadsheet calculations, and image-editing activities. The program detects multiple cores and automatically reconfigures the benchmarks to take full advantage of your rig’s hardware. And if you just want to test a particular subset of performance—say, file encoding—just select the “custom run” option and handpick your benchmark suites.

OpenSourceMark lets you save information about your CPU utilization to a text file.

OpenSourceMark is a great way to test whether your computer tweaking is actually having a measurable effect on your system’s performance. Do you really need to defragment your drive 12 times a week? How much does your antispyware program actually slow down your PC? What’s the hard benefit of all that extra overclocking?

Test Your Rig's Stability

Prime95 runs your PC at full loads until one of two things happen: You’re content with your testing or your rig shuts down.

Whether you’ve been overclocking an old rig to wring out more performance or you just purchased a new overclocked machine, stress testing your computer’s stability should be high on your priority list. (Stock-clock users can join in the fun too, but it’s not as critical. You can test whether a beta driver you downloaded mucks up your system in some capacity, but for the most part, a stock-clock machine should be inherently stable hardware-wise.)

An overclock can push a rig past safe (or stable) operation. You might not notice this instability or Windows might crash once an hour. Either way, one sure way to determine whether you’ve gone too far is to run your computer like a madman, and if it survives the rite of passage, you’re golden.

We use Prime95 for stress testing in the Lab. In a nutshell, the program calculates new Mersenne prime numbers and taxes the heck out of your processor and RAM in doing so. If you’re on a single-core machine, all you have to do is fire up Prime95 and select the Torture Test from the options menu. Run the test for 10 hours on small FFTs, which nails your CPU, before switching to large FFTs for the RAM.

Owners of multicore machines will want to download the .zip version of Prime95 and extract its contents to a new folder for each core of your machine. Run the program out of each folder, which will open up one instance of Prime95 per core. Click “Affinity” on the program’s advanced menu and set each instance to run on a different CPU core. Dual-core owners should run a small FFT on one core and a large FFT on the other; just double that equation if you’re rocking a quad-core PC.