While AMD and NVIDIA are consistently revising their GPU architectures, for the most part the changes they make are just that: revisions. It’s only once in a great while that a GPU architecture is thrown out entirely, which makes the arrival of a new architecture a monumental occasion in the GPU industry. The last time we saw this happen was in 2006/2007, when unified shaders and DirectX 10 lead to AMD and NVIDIA developing brand new architectures for their GPUs. Since then there have been some important revisions such as AMD’s VLIW4 architecture and NVIDIA’s Fermi architecture, but so far nothing has quite compared to 2006/2007, until now.

At AMD’s Fusion Developer Summit 2011 AMD announced Graphics Core Next, their next-generation GPU architecture. GCN would be AMD’s Fermi moment, where AMD got serious about GPU computing and finally built an architecture that would serve as both a graphics workhorse and a computing workhorse. With the ever increasing costs of high-end GPU development it’s not enough to merely develop graphics GPUs, GPU developers must expand into GPU computing in order to capture the market share they need to live well into the future.

At the same time, by canceling their 32nm process TSMC has directed a lot of hype about future GPU development onto the 28nm process, where the next generation of GPUs would be developed. In an industry accustomed to rapid change and even more rapid improvement never before have GPU developers and their buyers had to wait a full 2 years for a new fabrication process to come online.

All of this has lead to a perfect storm of anticipation for what has become the Radeon HD 7970: not only is it the first video card based on a 28nm GPU, but it’s the first member of the Southern Islands and by extension the first video card to implement GCN. As a result the Radeon HD 7970 has a tough job to fill, as a gaming card it not only needs to deliver the next-generation performance gamers expect, but as the first GCN part it needs to prove that AMD’s GCN architecture is going to make them a competitor in the GPU computing space. Can the 7970 do all of these things and live up to the anticipation? Let’s find out…

AMD GPU Specification Comparison
  AMD Radeon HD 7970 AMD Radeon HD 6970 AMD Radeon HD 6870 AMD Radeon HD 5870
Stream Processors 2048 1536 1120 1600
Texture Units 128 96 56 80
ROPs 32 32 32 32
Core Clock 925MHz 880MHz 900MHz 850MHz
Memory Clock 1.375GHz (5.5GHz effective) GDDR5 1.375GHz (5.5GHz effective) GDDR5 1.05GHz (4.2GHz effective) GDDR5 1.2GHz (4.8GHz effective) GDDR5
Memory Bus Width 384-bit 256-bit 256-bit 256-bit
Frame Buffer 3GB 2GB 1GB 1GB
FP64 1/4 1/4 N/A 1/5
Transistor Count 4.31B 2.64B 1.7B 2.15B
Manufacturing Process TSMC 28nm TSMC 40nm TSMC 40nm TSMC 40nm
Price Point $549 $350 $160 -

The Radeon HD 7970 is a card of many firsts. It’s the first video card using a 28nm GPU. It’s the first card supporting Direct3D 11.1. It’s the first member of AMD’s new Southern Islands Family. And it’s the first video card implementing AMD’s Graphics Core Next architecture. All of these attributes combine to make the 7970 quite a different video card from any AMD video card before it.

Cutting right to the chase, the 7970 will serve as AMD’s flagship video card for the Southern Islands family. Based on a complete AMD Tahiti GPU, it has 2048 stream processors organized according to AMD’s new SIMD-based GCN architecture. With so many stream processors coupled with a 384bit GDDR5 memory bus, it’s no surprise that Tahiti is has the highest transistor count of any GPU yet: 4.31B transistors. Fabricated on TSMC’s new 28nm High-K process, this gives it a die size of 365mm2, making it only slightly smaller than AMD’s 40nm Cayman GPU at 389mm2.

Looking at specifications specific to the 7970, AMD will be clocking it at 925MHz, giving it 3.79TFLOPs of theoretical computing performance compared to 2.7TFLOPs under the much different VLIW4 architecture of the 6970. Meanwhile the wider 384bit GDDR5 memory bus for 7970 will be clocked at 1.375GHz (5.5GHz data rate), giving it 264GB/sec of memory bandwidth, a significant jump over the 176GB/sec of the 6970.

These functional units are joined by a number of other elements, including 8 ROP partitions that can process 32 ROPs per clock, 128 texture units divided up among 32 Compute Units (CUs), and a fixed function pipeline that contains a pair of AMD’s 9th generation geometry engines. Of course all of this hardware would normally take quite a bit of power to run, but thankfully power usage is kept in check by the advancements offered by TSMC’s 28nm process. AMD hasn’t provided us with an official typical board power, but we estimate it’s around 220W, with an absolute 250W PowerTune limit. Meanwhile idle power usage is looking particularly good, as thanks to AMD's further work on power savings their typical power consumption under idle is only 15W. And with AMD's new ZeroCore Power technology (more on that in a bit), idle power usage drops to an asbolutely miniscule 3W.

Overall for those of you looking for a quick summary of performance, the 7970 is quite powerful, but it may not be as powerful as you were expecting. Depending on the game being tested it’s anywhere between 5% and 35% faster than NVIDIA’s GeForce GTX 580, averaging 15% to 25% depending on the specific resolution in use. Furthermore thanks to TSMC’s 28nm process power usage is upwards of 50W lower than the GTX 580, but it’s still higher than the 6970 it replaces. As far as performance jumps go from new fabrication processes, this isn’t as big a leap as we’ve seen in the past.

In a significant departure from the launch of the Radeon HD 5870 and 4870, AMD will not be pricing the 7970 nearly as aggressively as those cards with its launch. The MSRP for the 7970 will be $550, a premium price befitting a premium card, but a price based almost exclusively on the competition (e.g. the GTX 580) rather than one that takes advantage of cheaper manufacturing costs to aggressively undercuts the competition. In time AMD needs to bring down the price of the card, but for the time being they will be charging a price premium reflecting the card’s status as the single-GPU king.

For those of you trying to decide whether to get a 7970, you will have some time to decide. This is a soft launch; AMD will not make the 7970 available until January 9th (the day before the Consumer Electronics Show), nearly 3 weeks from now. We don’t have any idea what the launch quantities will be like, but from what we hear TSMC’s 28nm process has finally reached reasonable yields, so AMD should be in a better position than the 5870 launch. The price premium on the card will also help taper demand side some, though even at $550 this won’t rule out the first batch of cards selling out.

Beyond January 9th, AMD as an entire family of Southern Islands video cards still to launch. AMD will reveal more about those in due time, but as with the Evergreen and Northern Islands families AMD has a plan to introduce a number of video cards over the next year. So 7970 is just the beginning.

Winter 2011 GPU Pricing Comparison
  $750 GeForce GTX 590
Radeon HD 6990 $700  
Radeon HD 7970 $549  
  $500 GeForce GTX 580
Radeon HD 6970 $350 GeForce GTX 570
Radeon HD 6950 2GB $250  
  $240 GeForce GTX 560 Ti
Radeon HD 6870 $160  


A Quick Refresher: Graphics Core Next


View All Comments

  • mczak - Thursday, December 22, 2011 - link

    Oh yes _for this test_ certainly 32 ROPs are sufficient (FWIW it uses FP16 render target with alpha blend). But these things have caches (which they'll never hit in the vantage fill test, but certainly not everything will have zero cache hits), and even more important than color output are the z tests ROPs are doing (which also consume bandwidth, but z buffers are highly compressed these days).
    You can't really say if 32 ROPs are sufficient, nor if they are somehow more efficient judged by this vantage test (as just about ANY card from nvidia or amd hits bandwidth constraints in that particular test long before hitting ROP limits).
    Typically it would make sense to scale ROPs along with memory bandwidth, since even while it doesn't need to be as bad as in the color fill test they are indeed a major bandwidth eater. But apparently AMD disagreed and felt 32 ROPs are enough (well for compute that's certainly true...)
  • cactusdog - Thursday, December 22, 2011 - link

    The card looks great, undisputed win for AMD. Fan noise is the only negative, I was hoping for better performance out the new gen cooler but theres always non-reference models for silent gaming.

    Temps are good too so theres probably room to turn the fan speed down a little.
  • rimscrimley - Thursday, December 22, 2011 - link

    Terrific review. Very excited about the new test. I'm happy this card pushes the envelope, but doesn't make me regret my recent 580 purchase. As long as AMD is producing competitive cards -- and when the price settles on this to parity with the 580, this will be the market winner -- the technology benefits. Cheers! Reply
  • nerfed08 - Thursday, December 22, 2011 - link

    Good read. By the way there is a typo in final words.

    faster and cooler al at once
  • Anand Lal Shimpi - Thursday, December 22, 2011 - link

    Fixed, thank you :)

    Take care,
  • hechacker1 - Thursday, December 22, 2011 - link

    I think most telling is the minimum FPS results. The 7970 is 30-45% ahead of the previous generation; in a "worse case" situation were the GPU can't keep up or the program is poorly coded.

    Of course they are catching up with Nvidia's already pretty good minimum FPS, but I am glad to see the improvement, because nothing is worse than stuttering during a fasted pace FPS. I can live with 60fps, or even 30fps, as long as it's consistent.

    So I bet the micro-stutter problem will also be improved in SLI with this architecture.
  • jgarcows - Thursday, December 22, 2011 - link

    While I know the bitcoin craze has died down, I would be interested to see it included in the compute benchmarks. In the past, AMD has consistently outperformed nVidia in bitcoin work, it would also be interesting to see Anandtech's take as to why, and to see if the new architecture changes that. Reply
  • dcollins - Thursday, December 22, 2011 - link

    This architecture will most likely be a step backwards in terms of bitcoin mining performance. In the GCN architecture article, Anand mentioned that buteforce hashing was one area where a VLIW style architecture had an advantage over a SIMD based chip. Bitcoin mining is based on algorithms mathematically equivalent to password hashing. With GCN, AMD is changing the very thing that made their card better miners than Nvidia's chips.

    The old architecture is superior for "pure," mathematically well defined code while GCN is targeted at "messy," more practical and thus widely applicable code.
  • wifiwolf - Thursday, December 22, 2011 - link

    a bit less than expected, but not really an issue:

  • dcollins - Thursday, December 22, 2011 - link

    You're looking at a 5% increase in performance for a whole new generation with 35% more compute hardware, increased clock speed and increased power consumption: that's not an improvement, it's a regression. I don't fault AMD for this because Bitcoin mining is a very niche use case, but Crossfire 68x0 cards offer much better performance/watt and performance/$. Reply

Log in

Don't have an account? Sign up now