Radeon HD 7970 Graphics Card
For years now, the GPU market has been locked in a roughly annual cycle of product refreshes in a battle between AMD and Nvidia. Occasionally a completely new architecture comes along - this happened last in 2006/7 as a result of DX10 and Unified Shaders and there have been major upgrades such as VLIW4 and Fermi but now, at the end of 2011, AMD have launched their first GPU based on their Graphics Core Next (GCN) architecture.
It actually came as a bit of a surprise for us as we were not expecting samples until the new year so had to rush through testing. For months we've been watching rumours circulate the net about AMDs Southern Islands (as the 7000 series will be called) and GCN products. Many of them (such as RDRAM for the memory) were wildly off target and people could only speculate on what TSMCs new 28nm fabrication process would bring. Nvidia's Kepler range will not be with us for many months so AMD have once again taken a huge advantage as they did with the 5000 series Cayman launch in 2009.
The driving factor behind GCN seems to be general computing power and to regain the lead that Fermi products have when it comes to GPU processing tasks. 3D performance has not been overlooked but one AMD speaker after another at the secretive launch event in London this December stressed the importance of using GPU processing in everything from Video Editing to Financial Modelling. There are also a whole host of new features as we will see.
As usual AMD is launching its top end range first. Code named Tahiti, the 7970 (known as Tahiti XT) was launched a few days ago. Those expecting a 7950 to launch will be disappointed as AMD have, for reasons known only to themselves, decided not to launch a 7950 at the same time. We are not able to discuss any aspects any potential 7950 card so will focus on the 7970 in this article and update our readers when we can (AMD will let us know if and when such a card is to be launched).
Here are the main features of this card:
PCI-E 3 is to replace the aging PCI-E 2 standard (first implemented in the Radeon HD 3870) although most tests have shown only a minor (1-2%) benefit going from PCI-E x8 to x16 so a doubling of bandwidth seems unnecessary. The reality is all about computing power and the vast internal bandwidth which is great for games is not so useful for modelling climate change or massive parallel processing where a lot of information is transferred to and from system memory. We'll see applications such as AES encryption make good use of this.
Power management is a big thing for AMD and they always strive to have a lower idle power consumption with each generation. The 6970 and 6950 boasted 18W idle dissipation and AMD have gone further by claiming less than 3W in certain states. They do this by shutting down most of the GPU when it is not in use and springing to life instantly when needed.
We'll cover the other features later in this article.
Compared to the previous flagship we see 40% more transistors, a third more stream processors, a modest 5% clock speed bump, a third more Texture Units and most notably an extra Gigabyte of memory. The memory bus is now 384 bits wide instead of 256 and we are assured by AMD that there is a huge potential for overclocking to over 1GHz. This seems about right and we get the impression that there is a lot of room for improvement in the new 28nm process and in time we'll see speeds as well as yields rise.
One of the new features is the ability to have discrete sound channels for each video source and can be useful in video conferencing to tie up voices to video as well as in multi monitor setups.
We've already seen advances in Display Port for the 6000 series and now new HDMI standards are being embraced. This will allow for Stereo 3D effects by allowing two 60Hz signals, effectively doubling the HDMI bandwidth as well as supporting 4K by 2K displays and "deep colour" (48 bits per pixel).
One of the clever things to come from Intel's Sandy Bridge launch was QuickSync, their hardware H.264 encoder and surprised many with just how much a simple on chip GPU could do. AMD have come up with the Video Codec Engine (VCE). They go further than QuickSync by enabling a Hybrid Mode to utilise the massive parallel computing capabilities of the GPU and encode even faster. We don't currently have access to any software supporting this feature so will have to wait until a future date to test it.
Lastly, AMD have introduced Steady Video. We first touched upon this in our Llano review a few months ago and with version 2 has been improved and support added for interlaced content. The Quad Sum of Absolute Difference (QSAD) replaces the previous SAD and specialised instructions in GCN allow for over half a Trillion such operations per second.
GCN brings a radical shift in design::
The basic GPU building block is the GCN Compute Unit of which there are 32. Each of these Compute Units consists of 4 Texture Units and 4 SIMD Units. There are 2 Asybchronous Command Engines and 2 Geometry Engines.
Looking beyond the front end and shader cores, we’ve seen a very interesting reorganization of the rest of the GPU as opposed to Cayman. Keeping in mind that AMD’s diagrams are logical diagrams rather than physical diagrams, the fact that the ROPs on Tahiti are not located near the L2 cache and memory controllers in the diagram is not an error. The ROPs have in fact been partially decoupled from the L2 cache and memory controllers, which is also why there are 8 ROP partitions but only 6 memory controllers. Traditionally the ROPs, L2 cache, and memory controllers have all been tightly integrated as ROP operations are extremely bandwidth intensive, making this a very good design for AMD to use.
Here we can see the difference between GCN and VLIW. The former is much easier to program and allows for greater flexibility. There are a number of fundamental changes such as decoupling ROPs and only time will bring to light GCNs true potential.
Moving on to the card itself:.
Size is about the same as the 6970 (just under 11") but the circular cooler is evidently larger. The shroud in the reference design is sealed to improve cooling but will no doubt come in a variety of shapes and sozes from the usual manufacturers. Gone is the backplate to save space and this will improve airflow in Crossfire settings.
The exhaust grille now runs the full length of one slot, improving cooling but also reducing the space available for connectors. The result is that one DVI port is sacrificed but AMD say that an HDMI to DVI adaptor will be included with all Tahiti cards. Power requirements are an 8-pin and 6-pin connector to be on the safe side for overclockers. The dual BIOS is still a feature as can be seen from the microswitch but AMD will not be allowing a 7970 BIOS to be flashed onto a 7950 card as was the case for the 6970 and 6950.
Looking under the hood, the larger blower has larger fins as well to improve aiflow and the heatsink is similar to that of the 6970. AMD did point out that a special thermal compound had been used of a phase change variety and that if the heatsink were ever removed then such a perfect seal could not be regained. This sounds very similar to the material used on the 6990.
In operation our usual GPU-Z screenshot gives a clear overview of the cards characteristics including DX11.1 support and massive 3GB RAM. Had we more time we could have used Powertune and Overdrive to explore how far this card will go.
We have two main test systems at present and recently upgraded our Intel one to an i7-2600K we received from Intel for the Sandy Bridge launch in January. The AMD system has the most powerful processor currently available in the AMD range. All cards are tested at their default speed but in the case of the Radeon HD6990 we are able to get it to run at full 2x 6970 speeds by flicking a switch on the card. Therefore we will show results as HD6990 and HD6990 OC to distinguish between them.
Firstly we will start with DX10 testing with DX11 tests to follow.
3DMark Vantage is a full DirectX 10 compliant synthetic benchmark. It tests out all of the DirectX 10 abilities of each card to give the user an idea of where they rank with their graphic card in a collective pool of many users all over the internet.
The testing was done with the default performance setting for all of the cards to give an idea where each one stands in the GPU line up.
Ignoring the green bars (which should all be fairly similar) we see that the Radeon HD 7970 is faster than the GTX580 and almost on a par with the Radeon HD 6990.
Setting aside the middle bars (the CPU performance is fairly level as the only variable during testing is the graphics card) it can be seen that the performance of the high end cards scales with their product positioning. The Radeon HD 5850 which has served us so well for over a year is still capable of putting up a good fight but is now end of line and is about the same price as the better performing Radeon HD 6870. The 6970 is slightly behind its more expensive Nvidia rival. The 7970, the GTX580 and the HD6990 all manage to reach around the 30000 mark which is no easy thing. Graphically, the 7970 is almost as fast as the 6990.
Far Cry 2 is still a firm favourite and was taxing on any system. It's now broken by the HD6990 and is CPU limited at all resolutions. Having said that, the 7970 nudges ahead at 1024x768 and is otherwise on par with the GTX580.
HAWX seems to be easy for most systems to handle with playable rates from all contendors at all resolutions. We'll have to see how that changes when we move to HAWX 2. The 7970 is slightly behind its Nvidia rival here.
The current darling of horror games provides lots of eye candy and furious action. As with Far Cry 2 performance is very good but only the top end series of cards can provide triple digit frame rates at all resolutions. The 7970 wins hands down at all resolutions.
We're in the process of changing our benchmarks to more demanding ones and appreciate any feedback suggestions.
Because of the longevity of our standard tests we have introduced a host of new benchmarks. We wanted to add Crysis2 as well but DX11 issues with that title mean we will have to wait until a patch is released.
Aliens VS Predator
As with many of the already released DirectX 11 benchmarks, the Aliens vs. Predator DirectX 11 benchmark leverages DirectX 11 hardware to provide an immersive game play experience through the use of DirectX 11 Tessellation and DirectX 11 Advanced Shadow features. In Aliens vs. Predator, DirectX 11 Geometry Tessellation is applied in an effective manner to enhance and more accurately depict HR Giger’s famous Alien design. Through the use of a variety of adaptive schemes, applying tessellation when and where it is necessary, the perfect blend of performance and visual fidelity is achieved with at most a 4% change in performance.
All cards provide playable framerates with the 7970 being above 80fps at 1920x1080, comfortably beating the GTX580.
Stalker: Call of Pripyat
This is a special version of S.T.A.L.K.E.R.: Call of Pripyat Having a PC tested with this benchmark will enable the gathering of detailed information about its performance on various graphical modes and finds out how well the system is balanced for gaming and S.T.A.L.K.E.R.: Call of Pripyat in particular.
This benchmark is based on one of the ingame locations – Pripyat . The testing process includes four stages, those utilizing various weather and time of the day settings (day, night, rain, bright sun). In order to test the system thoroughly the benchmark is provided with a number of presets and options including different versions of DirectX (9.0, 10.0, 11), screen resolutions, antialiasing etc.
We use the Ultra (highest details) setting and DX11 mode. The only switching we did was between 1680x1050 and 1920x1080 resolutions and to switch the DX11 Hardware Tesselation and Contact Hardening Shadows.
The Radeon HD 7970 is the best performer here.
Once we turn on the DX11 eye candy the performance hit is low while providing a huge visual improvement. The 7970 beats even the dual GPU Radeon HD 6990.
While Street Fighter IV features models and backgrounds rendered in 3D, the gameplay remains on a traditional 2D plane, with the camera having freedom to move in 3D at certain times during fights, for dramatic effect. Producer Yoshinori Ono has stated that he wanted to keep the game closer to Street Fighter II. A new system called Focus Attacks (Saving Attack for the Japanese version) has been introduced, as well as Ultra Moves. The traditional six-button control scheme returns, with new features and Special Moves integrated into the input system, mixing classic gameplay with additional innovations.
This game will be fine on any graphics card but the 7970 provides some ridiculously high frame rates.
The chronic oversaturation of the mafia in our international media has taught us much. Mafia II is an attempt to chronicle these teachings in game form. Fact number one: mafia men do lots of killing. Fact number two: they like suits. Fact number three: mafiosa don’t call each other mafiosa; they use the term ‘wiseguys’.
Some titles hit a plateau and this is one of them with a Radeon HD 5850 being adequate (roll on Mafia 3). The 7970 here is not much better than the 6970 it replaces.
The primary purpose of Test A is to give an
indication of typical game play performance of the PC running Lost Planet 2.
(i.e. if you can run Mode A smoothly, the game will be playable at a similar
condition). In this test, the character’s motion is randomized to give a
slightly different outcome each time.
The Radeon HD 7970 leaves everything else in the dust and is the only card able to provide over 60fps in both tests.
We cannot give exact decibel levels due to lack of sufficiently sensitive equipment but can confirm that the ASUS EAH6970 is relatively quiet. Thanks to the fan spinning at lower than normal speeds most of the time it is inaudible at idle performance levels unless we turn off all the fans in our test rig (we use water cooling so the CPU is safe).
Power consumption is a big issue these days with GPUs sometimes consuming 2 or 3 times as much as a CPU. Performance returns never seem to scale anything like power increases and heavy users can find that the cost of increased electricity mounts up over time. The idle loads shown are all perfectly acceptable but under load (running Resident Evil 5 fixed benchmark) the HD6970 jumps to around 300W - about the same as the GTX580. Using FurMark the HD6970 uses 340W. That compares to the 400W of the GTX580. The 7970 consumes less power than the 6970.
Temperatures determine how long a component will last and can affect other components in a case. Remarkably, the 7970 produces lower temperatures than not only the 6970 but also those of the 3-slot design of the GTX580 DCU2 - a real testament to AMDs improved cooling. The only caveat we have is that the cooler on the 7970 is slightly louder than that on the 6970 it replaces.
Having the latest graphics card is not just about speed but also other benefits. One of these is hardware acceleration in video playback and this is illustrated with Adobe's Flash player with support for hardware acceleration.
The difference is startling and shows how important this feature can be if widely implemented. Let's hope that people keep pestering the relevant software developers until they all have this level of support by default.
To see the benefits of DX11 that are unique to the DX11 series cards we need to dig further. It’s tricky to test to see how much of a speed difference DX11 makes as there are few reliable DX11 benchmarks but as a showcase of DX11 quality the situation is much better. We used Unigine’s Heaven Benchmark and can only convey the quality to our readers via YouTube.
The improvement in quality is apparent for all to see.
This next video shows the effects of Hardware Tessellation. The technical white paper on this feature is enough to send the most avid technophile to sleep but there's no denying the benefits of the end result.
If you have a DirectX 11 card then you can download the demo for yourself at http://unigine.com/download/
The AMD Radeon HD 7970 has given us a lot of food for thought. A new architecture has surfaced that we can only begin to touch upon. We get the impression that this card can do a lot with the right software and from the Compute side of things there seems to be enormous potential.
Some will be disappointed that the die shrink to 28nm hasn't resulted in a doubling of specs compared to the 6970 but this is unfair as performance in some cases matches that of the dual GPU Radeon HD 6990 while using less power than the Radeon HD 6970.
A plethora of new features is pushing forward the frontiers in a number of areas and opening up exciting possibilities in many applications.
Without a doubt AMD has regained the single GPU performance crown from Nvidia and has a window of many months before the release of Nvidia's Kepler architecture. The estimated street price of $549 (or about £450 here in the UK when VAT is taken into account) is on par and will be snapped up by enthusiasts. As the 28nm process matures we will no doubt see a fall in prices - particularly as we approach the Kepler launch.
Ultimately, the Rdeon HD 7970 is a great performer now and offers the promise of a lot more when developers start to become profficient with the capabilities of AMDs GCN architecture. We give it our Gold Award.
All trademarks are the property of their respective owners.