
When [Otellini] joined the company in 1974, most people didn't even know what a PC was - From the Wall St Journal 11-11-2004
WITH ITS LATEST PARTS, the last major launch of the 40nm GPU generation, AMD has now presented its answer to the recently announced and seemingly quite well performing Nvidia chippery, the GTX580. The Radeon HD6970 and its lower-cost sibling, the HD6950, fare as well as expected in the benchmarks, and provide a nice feature boost too.
The twin graphics engines for much faster geometry and rasterisation plus massively sped up tessellation, Powertune real-time performance and power adjustment, as well as higher double precision floating point performance, are just some of the benefits. Image quality wise, there's Enhanced Quality Anti-Aliasing (EQAA) and morphological anti-aliasing (MLAA), plus support for deep colour and other new pixel formats.
To display all that, the twin Displayport 1.2 and HDMI 1.4a ports together with dual DVI outputs support up to six monitors from a single card. Finally, the approximately 3TFlops single precision and nearly 1Tflops double precision floating-point performance will help in compute jobs too, aided by the doubled memory capacity for locally executed tasks without hopping non-stop over the slow PCIe connection.

How about the performance boost versus the best of the past generation, the HD5870? In this case, besides a couple of benchmarks comparing the old reference HD5870, I took along what is probably the best ever HD5870 card: The Asus Matrix 5870 Platinum, a custom design in everything from the PCB with the best componentry and massive voltage tweaks, to the much more powerful yet less noisy cooling, and, yes, a full 2GB of GDDR5 memory just like the new cards - it does help in multi-monitor Eyefinity setups, too. You can see it here inside the same test setup:

Why is it interesting? Well, such a card can now be had at quite a discount due to AMD having launched the new cards, and, on top of 6 per cent higher processing and memory speed than the reference, its doubled memory can offer a more equal apples-to-apples comparison versus the new cards in detail-intensive and other memory capacity-dependent benchmarks even beyond, say, 3Dmark11.
Coming back to the naming, I still think that the cards should have rightfully been called the HD6870 and HD6850, as the up to 30 per cent speedup from the HD5870 and HD5850 kind of justifies that. What is now called HD6870 would then have been HD6770, as presumably the original plans were. So, what would have been HD6970 dual GPU card to succeed the HD5970, will now be called HD6990 early next year.
The HD6970 and HD6950 are just as long if not longer than the HD5870, so beware the case space requirements when purchasing. Otherwise, I think the card design, even though boxy, is just as modern as the HD5870 generation. I squeezed them in a very compact Gigabyte chassis in which our test platform for the HD6870 review last month already was - the uber-reliable Asus Sabertooth X58 mainboard with military grade componentry, hosting the Xeon 980X 3.33GHz six core Intel processor with twelve gigabytes of A-Dat DDR3-1600 memory and Windows 7 64-bit Ultimate running off an Intel X25-M SSD 160GB drive.
The new Xigmatek Aegir direct heat-pipe touch HSF cooled the CPU, quite well at that with 37C idle temps with Turbo enabled. The Aegir is a bit more compact than the previous models, and direct touch heat pipe models like this are now prevailing in the high end heatsink market. A Xigmatek 700W PSU feeds the whole setup. This PSU actually handled two HD5870s in Crossfire very well too, together with the 6-core setup, not bad for a 700W unit.
The display used was the brand new Gigabyte Envision P2271wL LED monitor, a 1920x1080 unit with a low power high contract display, 10 million to 1 contrast ratio at that, and full HD resolution with 5ms response. The very compact, light monitor has none of the usual hardware buttons, but just on-screen display touch controls for on-off and settings. At just 3.5kg net weight and less than 30W power consumption, the monitor is perfect for mobile benchmark testbeds, too. It's just a pity that there's no 1920x1200 resolution monitor unit available with 16:10 aspect ratio proportions.

So, the cards we will look at here will be, in order of reference: the original reference ATI Radeon HD5870 1GB; the Asus Matrix HD5870 Platinum 2GB at 900MHz GPU clock factory pre-set; and the AMD HD6950 and HD6970, both reference 2GB designs.
Here are the results. Let's look at the now already old 3Dmark Vantage DX10 test in both Performance and Extreme modes:
First the HD5870 reference...

Then the Asus Matrix HD5870 Platinum...

Followed by HD6950...

And finally, the HD6970 top of the crop...

Well, you can see that there are differences, but not that much. The HD6950 is about the same as the old HD5870, and a little slower than the Asus HD5870 pre-overclocked Platinum version.
Now we move to the Unigine Heaven 2.1 benchmark, focused on DX11 of course:

Here, the difference between the old and new becomes far more obvious, due to the HD6900 series optimisations for DX11 and tesselation as well.
Finally, here's our first run on the new 3Dmark 11 benchmark, just in the basic mode this time. We only ran it on the three cards that had the full two gigabytes of memory here, the Asus Matrix HD5870 as well as the HD6950 and HD6970:

We can see that, despite the new DX11 enhancements in the HD6900 series, the best of the HD5870 does come really close here in 3Dmark11. The differences between the generations are not so pronounced as in the Heaven test.
In Short
The AMD Radeon HD6970 and HD6950 bring along a bunch of new features and power management capabilities, plus the required high-end speedup to fend off the Nvidia GTX580 assault as well as its GTX570 lower-end spin off. With 8-pin plus 6-pin power connectors on the HD6970, there's enough extra juice flowing for some decent overclocking headroom even on this reference design. And, if the GPU vendors put in some efficient but a bit slimmer - like, say, 1-3/4 slot thick coolers instead of full two slot depth - we could have Crossfire pairs or even quads without totally blocking the airflow between the cards. After all, Crossfire now scales quite a bit better than in the previous generation, with near double performance using two cards in quite a number of games, and triple the single card performance in Quadfire setups. So, let's make sure the system airflow is not an issue when putting these cards together.
And, if you have problems finding them at the beginning, keep in mind that the good HD5870 upper-end cards like the Asus Matrix HD5870 are still near the top of the performance bracket. I wonder, how would the Matrix HD6970 look then - over 950MHz GPU clock for a start?
In the future, we'll look at the GPU compute performance as well as a Crossfire test, likely together with a Nvidia Geforce GTX580 setup. µ
The Good
Top notch performance without a more complex card design, overclocking potential.
The Bad
The card is still somewhat long for a single-GPU setup - how long will the HD6990 dual GPU one be?
The Ugly
The price - the 6970 is around $380. Don't expect the price to fall on this anytime soon, the card will be in hot demand.
Bartender's Score
9/10
This report sadly reflects the continued demise of the PC based gaming industry due to the lack of tech developments in the video card industry for the last few years. Now why do I make this wild claim?
Why, it's elementary my dear Watson. For years, all we've seen from Nvidia and AMD (even with AMD allegedly incorporating ATI technology) are incremental speed increases as a result of them flogging increasingly obsolete technology designs. Sure there's more and faster memory, more shaders and multi-monitor support, but the heart of these Paleontology period beasts is still a single core sub-1Ghz GPU. Wow! Deja Vu!!! It's like we're back in the late 1990s talking about Pentium II & AMD K6 pre-Athlon processors!
It's absolutely amazing (really a tragedy) that even these newest high end video cards are still massive glacier melting global warming dinosaurs requiring what...300 watts or more power? All I can think of is WTF???!!!
How is it that both Intel & AMD, among others, can make super-efficient low power +6 core CPUs operating in the +3Ghz range AND while only consuming around 1/3 the power of these latest perplexing Rube Goldberg inspired GPU designs? Why is it that nobody is able to harness that excellent technology to bring GPU performance and efficiency anywhere near on par? What die-process shrink or complete new GPU design philosophy will it finally take to tame these anachronistic beasts?
And when you see the GPU manufacturers incorporating proprietary architecture such as Nvidia and their CUDA, it's no wonder that the future of PC based games is almost nil at this point. I've already run into the CUDA roadblock several times. Backwards proprietary philosophies allow the game console industry to accelerate its assault assimilating the entire gaming community. There's just no longer a compelling reason for the PC gaming industry to continue, much less thrive, as there are no longer any GPU based advantages over game consoles. And with new developments like the Kinect, the PC will soon return to its humble commercial/industrial use only beginnings.
How very sad...
In the UK buying two 570s would be approx £110 more expensive too.
So the 6950x2 setup looks better value.
The 6950s scale better in Xfire than the 570s in SLI so it does work out a little better but at that end of the scale the difference either way is minimal.
YMMV.
My Xfire 5770 setup scores 4789 in 3DMark11 with a lowly OC'd X3 720 CPU so I'd take any results from that with a pinch of salt.
Or you can get 2 GTX570's and be a bit faster still, right matt?
And a bit quieter I understand.
You can get 2 6950s for the price of one GTX580, and considering two 6870s are faster than a GTX580, that's pretty good value