Sign in with
Sign up | Sign in
AMD Radeon HD 7970 GHz Edition Review: Give Me Back That Crown!
By , Igor Wallossek,
1. Is An Overclocked Radeon HD 7970 Greater Than GeForce GTX 680?

Can you believe the Radeon HD 7970 was introduced six months ago? In those 180 days, we’ve seen AMD claim the single-GPU performance crown, flesh out an entire family of graphics cards based on its capable GCN architecture, and then lose its fastest-in-the-world title to Nvidia’s GeForce GTX 680. I’m working on a script for the movie.

Along the way, though, we’ve wondered about the company’s approach to pricing. Even after Nvidia launched a faster, less expensive card, AMD kept selling its Radeon HD 7970 for $550. It eventually shaved off $70, only to see the competition kick out an even cheaper board (GeForce GTX 670) able to rout the 7970 in most games.

At no point was AMD’s flagship ever a bad card, though. Its dominance was simply contested quickly—and frustratingly—by a competing piece of hardware suffering such poor availability that you had to sign up for notifications just to catch it in stock. Although 680s are in stock now, as recently as a couple of weeks ago they were mostly a threat on paper. But the GeForce GTX 670 has always been a lot more accessible at $400, putting the Radeon HD 7970 under significant pressure. Certain 7970s are even down as low as $450.

To the point, with the maturation of TSMC’s 28 nm manufacturing process, AMD is discovering that a greater number of its Tahiti GPUs are running stably at higher core clock rates. Now, we already knew that the Radeon HD 7900s overclocked well. But rather than leaving extra performance on the table for enthusiasts to exploit on their own time, the company is unveiling a Radeon HD 7970 GHz Edition card that purportedly one-ups Nvidia’s GTX 680 and gives AMD a reason to push prices back up.

Meet The Radeon HD 7970 GHz Edition

Physically, this is the Radeon HD 7970 you already know. Put AMD’s reference GHz Edition card next to the one launched six months ago and you can’t tell them apart. What few differences there are all materialize under the card’s 11”-long fan shroud.

Most obvious is a higher core clock rate. Officially, AMD lists it at 1000 MHz—a bump up from the original version’s 925 MHz frequency. But it also enables a higher 1050 MHz P-state that the GPU favors when thermal headroom allows. AMD is marketing this combination of clock rates as PowerTune with Boost.

If you’re not already familiar with what PowerTune is or how it works, I break it down in my Radeon HD 6970 and 6950 review. Basically, though, AMD confirmed for us that PowerTune with Boost is the same exact thing, plus the ability to dynamically increase voltage. The company says its 1 GHz clock is fixed, and altering Overdrive’s frequency slider only changes the maximum boost level.

But it seems like you could also describe the 7970 GHz Edition as a 1.05 GHz card that, subjected to a synthetic power load like FurMark, drops 50 MHz and some voltage to not violate its TDP. After all, that’s what PowerTune has done for a year and a half.

The other performance enhancement comes courtesy of faster memory. Back when AMD launched the Radeon HD 7970, it “only” had access to 1375 MHz GDDR5 modules. On a nice, wide 384-bit bus, they were good for 264 GB/s of aggregate bandwidth. Now it’s using 3 GB of 1500 MHz modules on the same bus to push 288 GB/s.


Radeon HD 7970 GHz Edition
Radeon HD 7970
GeForce GTX 680
Stream processors
204820481536
Texture Units
128128
128
Full Color ROPs
3232
32
Graphics Clock
1000 MHz Base / 1050 MHz Boost
925 MHz1006 MHz
Texture Fillrate
134.4 Gtex/s
118.4 Gtex/s
128.8 Gtex/s
Memory Clock
1500 MHz
1375 MHz1502 MHz
Memory Bus
384-bit
384-bit256-bit
Memory Bandwidth288 GB/s
264 GB/s
192.3 GB/s
Graphics RAM
3 GB GDDR53 GB GDDR5
2 GB GDDR5
Die Size
365 mm2365 mm2
294 mm2
Transistors (Billion)
4.314.31
3.54
Process Technology
28 nm28 nm28 nm
Power Connectors
1 x 8-pin, 1 x 6-pin1 x 8-pin, 1 x 6-pin2 x 6-pin
Maximum power (TDP)
250 W250 W
195 W
Price
$500 MSRP
$450 Street
~$520 Street


AMD acknowledges that the Tahiti GPU itself is exactly the same. If you want the skinny on that, feel free to reference back to our Radeon HD 7970 launch coverage. Everyone else, let’s move on to a deeper analysis of PowerTune with Boost.

2. PowerTune With Boost: Is The Accelerator Stuck?

With a basic understanding of what PowerTune with Boost is, we wanted to explore its behavior in more depth.

According to AMD, the Radeon HD 7970 GHz Edition’s base clock is no longer end-user editable. It’s stuck at 1 GHz. When you move the slider in the Catalyst Control Center’s Overdrive tab, you’re actually moving what the company calls a boost state. And because the boost state is accompanied by a higher voltage, overclocking potential should be higher than modifying the base clock. From there, you’re supposed to be able to push Overdrive’s power control slider up even further to keep the card running at its higher P-state for longer without bouncing off of its protection circuitry and back to 1 GHz.

So, in theory, AMD is introducing an extra P-state, allowing the card to either operate at its base clock or a higher frequency. As long as the card stays within its TDP, it’s able to utilize the high clock, which can be adjusted through the Catalyst Control Center. If TDP is exceeded, the clock reverts to 1 GHz. The idea that AMD’s graphics hardware runs at one speed, selected for its applicability to real-world games, but then drops to another speed if some synthetic workload pushes it over the edge is not new at all. The existing Radeon HD 7970 could already do that. Getting more voltage to help augment a higher boost frequency would be a fresh concept, though.

Per AMD’s suggestion, we started bumping up the clock rate slider in Overdrive, hoping to see a higher voltage that’d help us hold onto a correspondingly higher overclock. What we saw instead was that altering the boost frequency by even 1 MHz up or down prevented the card from dynamically switching back to 1 GHz at all. Instead, the GPU remained at its dialed-in boost speed, temps kept going up, and we shut ‘er down to prevent a more serious problem.

We reproduced this behavior on two cards in two different countries running on four different platforms and multiple installations of Windows. Interestingly, we didn’t even need to adjust the power slider in Overdrive to maintain our overclocked boost state. We simply set a higher frequency, fired up FurMark, and watched the temp crest 90° in under a minute.

Our interest piqued, we decided to follow up by testing the card’s load in a number of scenarios. These included FurMark with boost operating as it should (between 1050 and 1000 MHz), as well as completely pegged at 1051 MHz. Bitmining serves as our compute workload, while the Aliens vs. Predator demo represents gaming. As you can see, several tests remained at full load with the card running at its maximum clock rate.

So, what are we looking at here? A protection mechanism that only works at its default settings, or a power-limiting feature asleep on the job? In order to answer our questions, we’ll look at power draw on the 12 V rail during each load scenario.

The Bitmining client and FurMark, both running at a constant 1051 MHz, pretty much max out the typical board power rating set forth by AMD. When the clock-limiting feature works properly, though, FurMark’s consumption drops notably. AvP and our OpenCL-based test are both able to run at full speed without any need to jump down to a lower frequency.

Now, if we take a look at the currents involved, we can see that at a partial load of less than 70 percent, the OpenCL test demonstrates some interesting spikes. Since the graph for the chip’s power draw, being a product of amperage and voltage, is a lot straighter, we can surmise that the GHz Edition card’s core voltage is constantly rising and falling.

It looks like we end up with a “persistent boost state.” Does that mean we can get by without any throttling thanks to built-in headroom? As of now, we still don’t know why overclocking confounds PowerTune.

We also wondered why only applications like FurMark and OCCT were able to trigger the feature, and only when we left the boost clock at its default value of 1050 MHz. Pure compute-oriented workloads able to generate similar load levels didn’t trigger the same reaction.

Digging deeper still, we’re going back to re-test everything on the original Radeon HD 7970. After all, AMD promises that this new card offers better performance at the same power levels using the same GPU. That’s be great, if it pans out.

3. Radeon HD 7970 Vs. Radeon HD 7970 GHz Edition

We start off by testing both cards at 1050 and 925 MHz. Since our samples are completely stable at those frequencies, we didn’t have to touch the PowerTune slider at all. The new card didn’t throttle, either, yielding an ideal comparison. As before, we logged power consumption for 50 seconds, using a gaming workload this time.

The dotted lines represent one card running at the emulated clock speeds of the other. And the final analysis yields an interesting result: the older and supposedly less-refined card draws marginally less power at 1050 MHz. It does even better at 925 MHz, coming in almost 5 W under the GHz Edition board. Perhaps this is a result of AMD’s voltage-adding mechanism designed to keep Tahiti more stable at its boost frequency.

However, we’re still not applying a full load to either card. Our next test does just that by applying a compute workload that doesn’t trigger throttling.

Power draw is pretty similar between the two boards. The new card might do its job under the TDP ceiling defined for the original 7970, but AMD’s GHz Edition board definitely doesn’t offer more performance at the same power levels as its predecessor. If you want more speed, you have to use more power.

4. Overclocking With PowerTune

We chose Crysis 2 for our overclocking tests. In our experience, this title is extremely sensitive to variations in GPU frequency, reacting negatively when a graphics card is near its edge (even more so than FurMark).

Both the Radeon HD 7970 and 7970 GHz Edition are set to 1050, 1100, 1150, and 1175 MHz. We were unable to reach anything higher than 1175 MHz on either of our GHz Edition samples without encountering serious visual artifacts.

First, let’s compare power consumption across the four runs at each frequency setting.

As we can see, increased power consumption corresponds to a reduction in GPU voltage. We would have expected the opposite.

Starting at 1175 MHz, we began to see minor visual artifacts (missing texture transparencies and polygon errors) that we were able to remedy by increasing VDDC. Apparently, PowerTune is stepping in very early on to turn voltage down, way before power consumption actually reaches a critical level.

The results of this intervention are quite obvious, manifesting as sporadically-dropped frames, texture errors, absent light sources, and other missing effects. Strangely, although the average and maximum frame rates continually increased, the minimum frame rates started to go down with each clock rate increase starting at 1150 MHz.

Our Radeon HD 7970 GHz Edition cards aren’t any more capable than the original Radeon HD 7970 overclocked. In our opinion, you reach the best results at 1150 MHz using AMD’s own tools. Push any harder and you’ll find the card dropping frames (even if its average frame rates seem to suggest better performance).

5. Will Your Old 7970 Take A GHz Edition Firmware?

Back in the day of the Radeon HD 6900-series cards, extracting more performance was as simple as taking your Radeon HD 6950 and flashing the 6970’s firmware over the top of it. Since much of what we've already seen from the Radeon HD 7970 GHz Edition already seems like a simple BIOS update, we wanted to see if the original 7970 would take this new card's firmware. Because the cards use different subsystem IDs (SSIDs), our first attempt involved forcing the update tool to overwrite the original BIOS.

Our system booted into Windows without incident after this little bit of cosmetic surgery. However, the driver wouldn't recognize a supported card, as witnessed in the following GPU-Z screenshot:

Clearly, something's not right. Just look at the memory capacity. Then there's the GPU clock rate and missing support for OpenCL/DirectCompute. In the end, we’re forced to concede that upgrading an old board to the new BIOS just isn't as easy as it was in the past.

6. Test Setup And Benchmarks
Test Hardware
Processors
Intel Core i7-3960X (Sandy Bridge-E) 3.3 GHz at 4.2 GHz (42 * 100 MHz), LGA 2011, 15 MB Shared L3, Hyper-Threading enabled, Power-savings enabled
Motherboard
Gigabyte X79-UD5 (LGA 2011) X79 Express Chipset, BIOS F10
Memory
G.Skill 16 GB (4 x 4 GB) DDR3-1600, F3-12800CL9Q2-32GBZL @ 9-9-9-24 and 1.5 V
Hard Drive
Intel SSDSC2MH250A2 250 GB SATA 6Gb/s
Graphics
AMD Radeon HD 7970 GHz Edition 3 GB (@ 1000/1500 MHz)

AMD Radeon HD 7970 3 GB (@ 925/1375 MHz)

AMD Radeon HD 7950 3 GB

Nvidia GeForce GTX 680 2 GB

Nvidia GeForce GTX 670 2 GB

Nvidia GeForce GTX 690 4 GB
Power Supply
Cooler Master UCP-1000 W
System Software And Drivers
Operating System
Windows 7 Ultimate 64-bit
DirectX
DirectX 11
Graphics DriverAMD Catalyst 12.7 (Beta) For HD 7970 GHz Edition, 7970, and 7950

Nvidia GeForce Release 304.48 (Beta) For GTX 680 and 670

Nvidia GeForce Release 301.33 For GTX 690


You'll notice we're missing some comparison hardware that was included in our most recent GeForce GTX 670 review. That's by design. AMD's Catalyst 12.7 driver release purportedly includes a number of important improvements. Thus, we felt it necessary to retest the Radeon HD 7970 and 7950, omitting any 6900-series board due to time constraints. Similarly, Nvidia just this week made public its 304.48 driver, which was supposed to improve performance in a number of games as well. Unfortunately, it didn't really affect any of the titles we tested, so we wasted some cycles there. But at least we have the GTX 680 and 670 benchmarked anew using the latest from Nvidia, too.

Games
Battlefield 3
Ultra Quality Settings, No AA / 16x AF, 4x MSAA / 16x AF, v-sync off, 1680x1050 / 1920x1080 / 2560x1600, DirectX 11, Going Hunting, 90-second playback, Fraps
Crysis 2
DirectX 9 / DirectX 11, Ultra System Spec, v-sync off, 1680x1050 / 1920x1080 / 2560x1600, No AA / No AF, Central Park, High-Resolution Textures: On
Metro 2033
High Quality Settings, AAA / 4x AF, 4x MSAA / 16x AF, 1680x1050 / 1920x1080 / 2560x1600, Built-in Benchmark, Depth of Field filter Disabled, Steam version
DiRT 3
Ultra High Settings, No AA / No AF, 8x AA / No AF, 1680x1050 / 1920x1080 / 2560x1600, Steam version, Built-In Benchmark Sequence, DX 11
The Elder Scrolls V: Skyrim
High Quality (8x AA / 8x AF) / Ultra Quality (8x AA, 16x AF) Settings, FXAA enabled, v-sync off, 1680x1050 / 1920x1080 / 2560x1600, 25-second playback, Fraps
3DMark 11
Version 1.03, Extreme Preset
HAWX 2
Highest Quality Settings, 8x AA, 1920x1200, Retail Version, Built-in Benchmark, Tessellation on/off
World of Warcraft: Cataclysm
Ultra Quality Settings, No AA / 16x AF, 8x AA / 16x AF, From Crushblow to The Krazzworks, 1680x1050 / 1920x1080 / 2560x1600, Fraps, DirectX 11 Rendering, x64 Client
SiSoftware Sandra 2012
Sandra Tech Support (Engineer) 2012.SP4, GP Processing and GP Bandwidth Modules
Arsoft MediaConverter 7.5
449 MB MPEG-2 1080i Video Sample to Apple iPad 2 Profile (1024x768), 148 MB H.264 1080i Video Sample to Apple iPad 2 Profile (1024x768)
LuxMark 2.0
64-bit Binary, Version 1.0, Classroom Scene


If you're one of the folks who likes to see our tessellation scaling results in HAWX 2, you'll find them below. In essence, though, the Radeon HD 7970 GHz Edition scales exactly like the 7970 that came before it.

7. Benchmark Results: 3DMark 11

We don’t make any recommendations based on synthetic results; if you can’t play it, you wouldn’t buy a graphics card for it.

In theory, though, 3DMark should allow us to compare AMD’s technology to Nvidia’s without the influence of developer bias resulting from the help that both companies provide in certain titles. If Futuremark is doing its job, the variation you see from one game to another should be largely mitigated here.

Whether or not that’s actually the case, AMD’s Radeon HD 7970 GHz Edition leapfrogs the GeForce GTX 670 and settles in just behind the GeForce GTX 680.

I don’t want to spoil the rest of the results, but you’re going to see that 3DMark doesn’t reflect the majority of our real-world benchmarks. In fact, there’s a trend that I’ll zero in on as we flip through three different resolutions—that is, the 7970 does increasingly well as you ask it to render more pixels and turn up quality settings. Considering that our 3DMark 11 benchmark employs the Extreme preset, I really would have expected it to place AMD’s latest ahead of Nvidia’s GeForce GTX 680.

8. Benchmark Results: Battlefield 3 (DX 11)

The Radeon HD 7970 GHz Edition effectively ties the GeForce GTX 680 at 1920x1080. At 2560x1600, it edges ahead, succumbing only to Nvidia’s GeForce GTX 690.

That these cards all finish so close to each other is what’s most remarkable to me, though. When the GeForce GTX 670 first emerged, it averaged better performance than the Radeon HD 7970. But AMD’s beta Catalyst 12.7 driver alters the landscape in our very first real-world benchmark, Battlefield 3.

When you combine the benefits of Catalyst 12.7 with the 7970 GHz Edition’s clock rate boost, this metric sets the tone for the rest of our testing. Even a Radeon HD 7950 at $360 or $370 doesn’t looks like such a bad deal any more.

9. Benchmark Results: Crysis 2 (DX 9/11)

AMD and Nvidia trade blows in our DirectX 9-based Crysis 2 testing, the GeForce GTX 690 always placing first, the 7970 GHz Edition taking second, and the less expensive boards mixing it up, depending on resolution.

Nvidia fares better under DirectX 11, though, which is what most enthusiasts are probably going to want to use. The GeForce GTX 680 is faster than the new GHz Edition board in all three charts, but the difference isn’t significant.

10. Benchmark Results: The Elder Scrolls V: Skyrim (DX 9)

AMD’s driver team is to thank for the reshuffling of results in Skyrim. Across the board a Radeon HD 7970 GHz Edition beats Nvidia’s GeForce GTX 680. In fact, once you hit 2560x1600, even the 7950 gets its licks in, outperforming the 680 in the High and Ultra quality presets.

The Radeon HD 7970 GHz Edition, specifically, finishes just ahead of the older 7970 at all three resolutions, in all six test scenarios. Is the delta between both cards worth $50? Decidedly not—the 7970 has you covered all the way through 2560x1600, and at no point does the GHz card’s extra performance help it make or break playability.

It’s also worth noting that our Radeon HD 7970 overclocks right up to the 1125 MHz limit imposed by AMD’s Overdrive software. Coupled with a 1500 MHz memory clock, you can quite easily exceed the performance of the stock 7970 GHz Edition (and come very close to the same card overclocked) for $50 less.

11. Benchmark Results: DiRT 3 (DX 11)

Previously, Nvidia had little trouble running the table in DiRT 3. But again, AMD’s driver team helps kick these Tahiti-based cards into gear. In all three of our tested resolutions, the Radeon HD 7970 GHz Edition outperforms the GeForce GTX 680 with and without anti-aliasing enabled. Given much more memory bandwidth, it’s not surprising to see the anti-aliased numbers favor AMD more prominently.

12. Benchmark Results: World Of Warcraft: Cataclysm (DX 11)

Although the Radeon HD 7970 appears faster than the GHz Edition board at 1680x1050 and 1920x1080, that’s only because we’re sorting by non-anti-aliased numbers. With the greater demands of 8x AA, the GHz Edition card is quicker. And of course, at 2560x1600, it outmaneuvers its predecessor with and without AA.

But, as we’ve seen many times before, AMD struggles against Nvidia in WoW. Even at 2560x1600, the 7970 GHz Edition can’t catch a GeForce GTX 670. An average of 68 FPS is most definitely ample at our highest tested resolution with 8x AA enabled, but we’d still like to know why multiple generations of Radeon cards haven’t done well in this popular MMORPG.

13. Benchmark Results: Metro 2033 (DX 11)



In five of the six combinations of resolution, anti-aliasing, and anisotropic filtering, AMD’s Radeon HD 7970 GHz Edition beats Nvidia’s GeForce GTX 680. Half of the time, even the Radeon HD 7970 beats the GTX 680 (the half of the time when MSAA is applied, that is).

14. Benchmark Results: GPU Compute

Last week, I got my first look at AMD’s Trinity-based APU. The story kicked off with x86-based benchmarks. The architecture’s Piledriver-based cores didn’t disappoint, but they won’t be Trinity’s biggest strength, either. Then we hit graphics—an AMD forte, no question. Finally, we dipped our toes into OpenCL with WinZip 16.5 and witnessed the potential of GPU acceleration in a productivity-oriented title.

Or so it seemed.

That's WinZip 16.5 with OpenCL turned off...That's WinZip 16.5 with OpenCL turned off...

...and with OpenCL acceleration enabled. Core i7-3960X-likey!...and with OpenCL acceleration enabled. Core i7-3960X-likey!

Reader mayankleoboy1 asked me to look into the resource utilization of WinZip 16.5, and it turns out that enabling OpenCL puts minimal load on a Radeon HD 7970. In the screenshots below, our Core i7-3960X is only at a 14% duty cycle with OpenCL disabled and activity on six of its available threads. Turn the feature on, though, and much more of our host processor is utilized. While I managed to catch GPU-Z showing 5% utilization on the Radeon HD 7970, it spent far more time at 0 and 1%.

With that said, you’re only able to turn OpenCL support on with an AMD graphics card, so it remains an exclusive feature for now. And it is effective, cutting our workload in more than half. It just doesn't seem like a very "GPU-accelerated" capability.

The Radeon HD 7970’s good result is improved upon with the GHz Edition card in LuxMark 2.0, which centers on the LuxRender ray tracing engine. The new board nearly doubles the performance of Nvidia’s dual-GPU GeForce GTX 690. Although WinZip doesn’t seem to owe its speed-up to AMD graphics, LuxMark certainly does.

So does MuseMage, a photo editing application with a number of OpenCL-accelerated filters. With scaling that corresponds to shader processors, clock rate, and, clearly, architecture, the new GHz Edition board finishes in first place.

There were a number of additional tests we wanted to run, including Photoshop CS6, a beta build of HandBrake, and a beta build of GIMP. However, retesting all of our cards using the latest drivers in games took precedence. But the expanding list of OpenCL-optimized titles makes it clear that, finally, all of the talk about heterogeneous computing is giving way to real software we can test. More important than that self-serving state of affairs, apps that people actually use on a daily basis are being affected.

Consequently, we’ll be including a lot more compute-oriented testing in our graphics card reviews moving forward. Depending on how Nvidia responds, this could remain a real bastion for AMD, which is already betting so much of its success on the Heterogeneous System Architecture as Nvidia largely ignores client workloads in favor of its Quadro/Tesla business.

As developers get better about identifying the aspects of their workloads able to benefit from a GPU’s parallelized architecture, and then maximize the performance they’re able to extract from the hardware, Sandra helps demonstrate that the potential of AMD’s GCN far outstrips Kepler as it is implemented in GK104. And with GK110 not expected on the desktop until 2013, our placing isn’t expected to change this year.

Interesting also is that Nvidia's latest drivers still don't allow for PCI Express 3.0 support on Sandy Bridge-E platforms, resulting in less interface bandwidth than what AMD's cards achieve. However, you can now manually force it on through a patch from the company.

15. Benchmark Results: MediaConverter 7.5

At long last, the fixed-function Video Codec Engine is ready for testing, six months after its introduction in the Radeon HD 7970!

AMD sent us a copy of Arcsoft’s MediaConverter 7.5, specially optimized to exploit VCE. We eagerly got it installed, anxious to see how the company’s multi-stream H.264 encoder improved performance.

Unfortunately, it had the opposite effect, slowing down our MPEG-2 and H.264 source files compared to our overclocked Core i7-3960X working on its own.

Of course, very few people have their own $1000 processor running at 4.2 GHz, so we asked AMD what it’d take to turn the tables and see the VCE-enabled result on top. The company admitted that VCE will play a more assistive role in lower-end platforms armed with Radeon HD 7800- or 7700-series cards. To that, we’d add desktops with Trinity-class APUs in them.

Perhaps the most ironic data points come from the GeForce GTX 680 and 670, though. The same AMD-supplied, AMD-optimized build of MediaConverter also supports CUDA, demonstrating that not all graphics cards get outperformed by fast CPUs in these workloads.

16. Temperature And Noise

The Radeon HD 7970 GHz Edition is only slightly warmer and a skosh louder than its predecessor at idle. But we’re not concerned about idle. The last time we looked at Radeon HD 7970, we were most worried about its bad behavior under a load.

Although we’re not worried about an 84 degree load temperature, it is nine degrees higher than the retail Radeon HD 7970 we have in the lab (and the hottest-running single-GPU card we have).

Much more troubling is the noise generated by AMD’s problematic reference cooler. I even had to go back and re-test because it seemed inconceivable that the company would ship out cards that were even louder than the Radeon HD 6990.

There’s a silver lining on this one, though. Ahead of this review, I let AMD know about our acoustic concerns and the company claims that most partner boards will employ third-party cooling, not its reference configuration. Just a little earlier this week we saw in Radeon HD 7950 3 GB: Six Cards, Benchmarked And Reviewed that new heat sinks and fans can work wonders on Tahiti-based boards. Fingers crossed, then, that the Radeon HD 7970 GHz Editions that show up on store shelves don’t sound like our sample.

Of course, you don't have to wait for board partners to work their magic...

17. Radeon HD 7970 GHz Edition Gets Our Aftermarket Cooling Treatment

We're not going to recommend against buying a card with AMD's reference cooler without trying to come up with alternatives.

Considering that our sample spun up to 84 degrees C in an air-conditioned room in a relatively modest looping test that doesn't apply a constant load, we don't even want to know how it'd sound (or perform) during an extended gaming session.

The following chart shows the temperature curve of our reference card running a far more demanding FurMark workload at four different clock speeds over a period of four minutes.

Regardless of how noisy AMD's solution is, can an aftermarket card ameliorate this thermal situation? We decided to go with a cooler in the $50 range, which still ends up pricey when you add it to the card's $500 cost. Spending anything more just to counter a problem with the reference design, we think, is wasteful.

Is Gelid's Icy Vision-A A Good Compromise Between Performance And Price?

Gelid’s Icy Vision isn’t really a particularly new design, but it's now officially validated for the Radeon HD 7970. How well does it handle almost 250 W of heat dissipation under full load?

After we finished installing the new cooler, we re-ran our tests and compared the results to AMD’s reference design.

Gelid’s Icy Vision-A doesn't deliver the same cooling efficiency as what we've seen from companies like MSI, HIS, or Gigabyte, but the improvement is both quantifiable in the above chart, and in our acoustic testing. In light of its moderate price, we believe this cooler is a reasonable choice that helps complement AMD's hardware. This is the treatment we're hoping to see from board partners.

Of course, the bars on a chart don’t really tell you much about the actual noise a specific cooler produces. That’s where our videos come in, allowing our readers to compare both cooling solutions directly.

First, our Radeon HD 7970 GHz Edition with Gelid's Icy Vision-A:

Then, our Radeon HD 7970 GHz Edition with AMD's reference cooler:

What a difference. The bottom line is that $50 bucks absolutely gets you a better cooling solution. In our opinion, you're better off saving the $50 AMD is trying to make on its GHz Edition card. Instead, buy the original Radeon HD 7970 and spend the leftover money on a cooler like this. You'll achieve similar performance at a similar price, but get less noise and better thermals. The only sacrifice is a loss of warranty.

18. Power Consumption

Idle power consumption remains one of AMD’s strengths, and its latest effort tops our charts by using less than 100 W of system power sitting on the Windows desktop.

ZeroCore technology favors AMD even more prominently, though the GHz Edition card doesn’t quite drop as low as the already-available Radeon HD 7900-series boards. Even still, that’s an additional 10 W savings when Windows shuts off your monitor.

The Radeon HD 7970 GHz Edition uses more power than a vanilla Radeon HD 7970—not surprising when you consider its faster GPU, higher-clocked memory, and ability to dynamically increase voltage.

Although AMD says the GHz Edition board stays under the original Radeon HD 7970’s TDP, it still averages about 13 W higher system power consumption in our 3DMark demo. That puts it 73 W higher than a Radeon HD 7950 and 43 W higher than a GeForce GTX 680.

19. New Drivers Deliver; Radeon HD 7970 Claims A Symbolic Win

AMD’s driver team deserves to take this weekend off. Its beta Catalyst 12.7 build does stellar things to the performance of several games in our benchmark suite. No longer is Nvidia’s GeForce GTX 670 faster than the Radeon HD 7970. And, in fact, these new drivers are largely responsible for the Radeon HD 7970 GHz Edition card outmaneuvering GeForce GTX 680 in most of the tests we ran.

Suddenly, the vanilla Radeon HD 7970 looks a lot more attractive priced between $460 and $480. So, where does that leave this new GHz Edition card?

Rather than simply replacing the original Radeon HD 7970, AMD sees its 1000+ MHz model coexisting. Prices are going to start at $500, the company says, and we’re assuming that covers the reference-class board. Cards with aftermarket cooling will almost certainly cost more. Expect to start seeing availability next week, with broader supply rolling in the week after.

Again, AMD’s factory cooler is disturbingly loud, so avoid that. But if board partners can tack on more interesting solutions, like the ones we saw in our Radeon HD 7950 round-up, and still manage to keep prices close to $500, the Radeon HD 7970 GHz Edition will take its place in front of the GeForce GTX 680.

Here's the rub, though. In a world without GeForce GTX 670, Nvidia’s 680 might still be on our radar, and we’d have a proper grudge match on our hands in the high-end space. That’s simply not the case anymore, though. Today, there are other high-end GPUs that offer nearly as much performance at much more attractive prices, dissuading us from either company’s single-GPU flagship.  

We already know that GeForce GTX 670 performs within a few percentage points of 680, and for $100 less. Moreover, once Catalyst 12.7 goes public and puts the older 7970 in the same league as GeForce GTX 680, it’d simply make more sense to save the $50 and do a little overclocking of your own. Shoot, I have two retail 7970s here that both hit the core frequency limiter in Overdrive at 1125 MHz, and have no troubling achieving the same 1500 MHz memory settings as the GHz Edition card.  

There’s no guarantee that 7970s will continue overclocking as well as they have been, particularly now that AMD is saving the top-binned ASICs for this new model. However, we’d rather save some money and come close. For that reason, the original Radeon HD 7970 and GeForce GTX 670 remain our favorite gaming graphics cards.