Unpacking the XFX HD 7950 Black Edition caused a bit of déjà vu. The card bears a strong resemblance to its big brother, the HD 7970 Black Edition (reviewed April 2012), clad in svelte brushed aluminum. If graphics cards dressed up for black tie galas, then the XFX Black Edition is ready to attend.
XFX pushes the reference clocks higher than stock, hitting 900MHz for the core clock and 1,375MHz for the memory clock. The additional memory cycles translate to a peak memory bandwidth of 5.5 gigabytes per second—the same as the HD 7970, and higher than the 5GB/s of the stock 7950. The question is: Can the GPU keep up? There’s always a balance between memory bandwidth and how much of that bandwidth the GPU cores can actually use. Plus, as more games become shader- and tessellation-intensive, bandwidth isn’t as big a part of the equation.
AMD’s reference HD 7950 board sets its core clock at 800MHz and memory at 1,250MHz, using the default cooling system. Sapphire takes this reference board, adds dual 12cm fans, and juices the core clock to 900MHz. The memory clock remains at 1,250MHz—but that’s 3GB of 1,250MHz GDDR5. Priced at around $480, it’s worth seeing how the card compares with Nvidia’s GTX 580. Note that we’ve also included results from the XFX Radeon HD 7970 for your reference, but excluded that card from the direct comparisons.
For direct comparison we turned to two different versions of the GTX 580—the slightly overclocked EVGA GTX 580 SC with 1.5GB of GDDR5 and the ultra-beefed-up EVGA GTX 580 Classified with 3GB of video RAM—as well as the XFX Radeon HD 7950 Black Edition reviewed next.
The Sapphire card ships with a slight memory bandwidth disadvantage compared with the XFX card (5GB per second versus 5.5GB/s) due to running its GDDR5 frame buffer at the reference clock speed. This shows up in a few benchmarks, where the Sapphire card places just a little behind the XFX card, but the differences are pretty small.
‘We don’t need no stinkin’ reference design!’ says XFX
It's not unusual to see factory-overclocked videocards ship with custom cooling solutions a few months after a GPU launches. But XFX didn’t waste any time with its Radeon HD 7970 Black Edition—a factory-overclocked card with a custom cooling solution that aims to take the performance crown. Based on what we’ve seen to date, XFX has delivered the fastest single-GPU card on the planet.
The Radeon HD 7970 is AMD’s latest GPU, with support for DirectX 11.1 and OpenCL 1.2. It’s a brand-new architecture—completely different from past AMD GPUs—built on TSMC’s 28nm manufacturing process and sporting a staggering 4.3 billion transistors. In AMD’s reference design, the 7970’s core runs at 925MHz, and its GDDR5 memory is clocked at 1,375MHz. XFX ups the ante significantly, pushing the core clock speed to a whopping 1GHz and running its 3GB of memory at 1,425MHz.
As you might imagine, the results are nothing short of amazing. We’re seeing genuine performance milestones here, including a 3DMark 11 performance score higher than 8,000 (for a single GPU), Far Cry 2 hitting 100fps at 2560x1600 with 4x AA, and Batman: Arkham City heading north of 50fps at the same resolution and AA settings. On top of that, the idle system power is just 124 watts, and a dark idle (when Windows 7 blanks the screen) draws 110 watts. Push the card and you’ll see system power consumption climb to 349 watts, but that merely puts its overall power draw into Fermi territory. XFX’s Radeon HD 7970 Black Edition is substantially faster than EVGA’s super-overclocked 3GB GeForce GTX 580 Classified, and it’s outfitted with just two PCIe power connectors (one 8-pin and one 6-pin). EVGA’s card requires three power connectors.
ASUS HAS GOTTEN a lot of mileage out of its beefy DirectCU II GPU-cooling technology. It has brought some serious overclocking chops to the GeForce GTX 580 in the form of Asus’s Matrix-branded edition, for example. The DirectCU II versions of the GeForce GTX 560 Ti and the Radeon HD 6870 also sport serious overclocks, and those cards perform well in their respective classes. What’s even better is that the company doesn’t charge much of a price premium for its best cooling tech on cards below the Matrix GTX 580.
We’re scratching our heads, however, over Asus’s decision to offer this GTX 570 card in a three-slot configuration similar to its Matrix GTX 580, but running at Nvidia’s reference clock speeds. The beefy cooler delivers plenty of DIY overclocking potential, but you must assume all the risk. Since we review cards based on out-of-the-box performance, we had to benchmark this one with its 742MHz core clock and 3,800MHz (effective) memory clock.
One good thing the new cooler does provide is fewer decibels. This card isn’t whisper-quiet under load, but it generates much less noise than many of the cards in its class—particularly the Radeon HD 6970, which can get fairly loud under heavy loads.
We found the Asus Matrix GTX 580 Platinum that we reviewed in the November 2011 issue to be pretty badass: It’s a solid, factory-overclocked card that’s impressively easy to push even harder. But it’s also three slots wide and requires two 8-pin PCIe power connectors. Gigabyte’s GTX 580 Super Overclock (model GV-N580SO-15L) takes Nvidia’s GPU even further, pumping the core from a stock 772MHz all the way to 855MHz, and the card’s 1.5GB of GDDR5 memory from a stock 1,002MHz to 1,025MHz (the Matrix GTX 580 comes out of the box with its GPU running at 816MHz and its memory at 1,002MHz). And the Gigabyte takes up only two slots and uses just a single 8-pin power connector.
It’s not unusual to see factory-overclocked videocards ship with custom cooling solutions—a few months after the GPU launches. This time, XFX hits the ground running with their Radeon HD 7970 Black Edition. This is a factory-overclocked card with a custom cooling solution that aims to take the performance crown. Based on what we’ve seen to date, XFX has delivered the fastest single-GPU card on the planet.
When is a GTX 560 Ti not really a GTX 560 Ti? When it’s almost a GTX 570.
Nvidia’s latest GPU, the GTX 560 Ti 448 is really a GTX 580 (originally dubbed the GF110) with two functional blocks disabled, reducing its CUDA Core count from 512 to 448. The GTX 570 is a GF110 with one functional block disabled, endowing it with 480 CUDA Cores. The original GTX 560 Ti is a completely different chip, with different power requirements, but all 384 of its cores are fully functional.
Imagine a graphics card weighing 5.25 pounds with three (yes, three) 8-pin PCI Express power connectors. Now imagine this card taking up three PCI Express slots and almost sucking the life out of an 850W power supply.
That may be one reason Asus named this card after the Roman god of war. It's probably the most powerful single graphics card we've tested, but that power comes at a substantial cost. You'll need the right type of motherboard and case, too—one where you can install a three-slot-wide card that's 12.25 inches long and 5 inches tall.
The Asus Matrix GTX 580 Platinum is quiet, fast, and really, really easy to overclock. It's also massive.
How massive? When we got the box, we thought Asus had shipped us a motherboard by mistake because the box was so large.
The size of the pacage is a clue to the size of the card itself. Asus builds a variant of its DirectCU II dual-fan technology onto the GTX 580, resulting in a card that's fully three expansion slots wide.
The Radeon HD 6950 often gets overlooked, because it falls into an in-between netherworld of pricing. Typical cards cost anywhere from $240-$300, but most seem to hover around the $270 mark. This MSI overclocked card, built using the company's Twin Frozr III dual-fan cooler, sits at around $280. So high-end buyers overlook this price category and budget buyers feel like it's a little too much.