AMD officially lifted the curtain today on its much-anticipated Radeon HD 5970, a dual-GPU solution which as of today has become the world's fastest graphics card on the market. Packed with some pretty impressive specs, the new card outclasses the competition in every benchmark thrown at it with ease, and puts AMD at the head of the high-end graphics segment for the first time in a long while. Not one to just sit on the sidelines and let its rival get all the glory, Nvidia is hoping to build some anticipation for its upcoming DirectX 11 lineup by showing a picture of what appears to be the first working sample of a 40nm, GF100 desktop card based on the Fermi architecture. Posted on the company's Facebook account, the picture shows a black GeForce-branded card installed on an ASUS Rampage II Extreme LGA 1366 motherboard, running Uningine's Heaven DirectX 11 benchmark on a Dell 24-inch monitor.
Details are scarce, as even the corner showing framerates has been cut off, but from what little we can see the card covers two slots and packs an 8-pin plus 6-pin PCI Express connector, meaning it can suck up to 300W and will likely provide healthy overclocking headroom. Expectations are running quite high for Nvidia's answer to the Radeon HD 5800 series, although clock and overall performance numbers remain unknown.
ATI revealed its last generation Radeon 5000 graphics family last September, Radeon HD 5970.
In a few tests the single-GPU Radeon HD 5870 was able to outgun the mighty GeForce GTX 295, while in most it managed to match or improve upon the Radeon HD 4870 X2. As you are likely aware, both of these products carry dual GPUs, which bring a number of implications, not to mention bigger price tags.
Looking forward we knew Nvidia would have no immediate response to the new Radeons, while on the other hand ATI was not done unleashing its full series of products. Subsequently we looked at two more products that were meant to underperform the flagship HD 5870: the slightly cut down Radeon HD 5850 and the mainstream aimed HD 5770.
From left to right: Radeon HD 5770, HD 5850, HD 5870, and HD 5970.
But as we discussed in our preliminary Radeon 5870 review, on the horizon was also a follow-up to the Radeon HD 4870 X2, code-named "Hemlock XT", which in essence would put together a pair of Radeon HD 5870 GPUs on a single PCB.
Today AMD is officially unveiling the Hemlock as the new ATI Radeon HD 5970, hoping to expand its current dominance in single-GPU performance with the fastest single-slot graphics card on the market -- a title that until today was reserved for Nvidia's dual-GPU GeForce GTX 295.
The Radeon HD 5970 looks to be well suited for the job. The GPUs used in this card use exactly the same configuration found on the Radeon HD 5870, while core and memory frequencies match those of the Radeon HD 5850. This provides the HD 5970 with an unmatched memory bandwidth of 256GB/s.
All this horsepower will come at a hefty price, as AMD expects to charge as much as $600 for it. This is a bit hard to swallow, but all things considered it's about where we expected it to stand among the rest of high-end graphics offerings.
Full exposure: A "naked" Radeon HD 5970 graphics card.
The $400 Radeon HD 5870 is still sold out virtually everywhere due to extreme shortages. The Radeon HD 5850 version still costs $300, meaning that a pair of them would cost the same amount as a single Radeon HD 5970, and we do not expect the performance to be nearly as good. On the other side of the equation, Nvidia's top performer, the GeForce GTX 295 is selling for around $500.
As we see it, only extreme users looking for the best possible gaming experience - or the bragging rights attached to the label - will find the Radeon HD 5970 to make complete sense.
Knowing this is the 5970's intended market, AMD has stamped a big "unlocked" label over the card in our press kit. Perhaps more marketing buzz than actual functionality, the Radeon HD 5970 allows you to jack core and memory frequencies as high as you like -- or at least as high as they will go without compromising stability. This in itself is far from exciting news as there's been software available to do just that for ages.
Radeon HD 5970 is a truly massive graphics card, even for a high-end dual-GPU product. Its dimensions whale anything previously seen, measuring in at 30cm long and weighing a little over 1kg.
Cooling the Hemlock XT GPUs is a large aluminum heatsink made up of 36 fins. Measuring 16cm long, 7cm wide, and 2.5cm tall, it is slightly larger than that used in the Radeon HD 5870. Connected to the base of this heatsink is a massive copper plate which helps improve efficiency, while a 75x20mm blower fan draws air in from within the case and pushes it out through the rear of the graphics card.
For the most part this fan operates very quietly, helped by the impressively low 42 watt idle consumption of the Radeon HD 5970. When gaming, the fan will inevitably spin up as the card can consume up to 294 watts under load, but even with the increased thermal stress the 5970 does not scream at intolerable levels.
The heatsink and fan have been enclosed within a custom built housing that conceals the entire graphics card, which is the same design used with the Radeon HD 5870. This setup helps protect the card very well; Nvidia has been using similar enclosures for some time with their most prized graphics cards, such as the GTX 295.
Removing the heatsink exposes the two GPUs, GDDR5 memory chips, and a few other critical components.
With the heatsink off the Radeon HD 5970 looks almost identical to the Radeon HD 4870 X2, though it is noticeably longer.
Key changes involve the new GDDR5 memory, improved power circuitry, and an updated 2.0 PEX bridge from PLX Technologies, used to link the GPUs together. All this hardware is cooled via the large heatsink that covers the majority of the card’s 30cm long PCB.
GDDR5 memory works at an ever more impressive frequency (1.0GHz x 4) on this particular model, which features a total capacity of 2GB. The core configuration of the Radeon HD 5970 doubles everything in the 4870 X2 with 1600 (320x5) SPUs, 80 TAUs (Texture Address Units), and 32 ROPs (Rasterization Operator Units).
AMD’s reference card came equipped with Hynix ICs (H5GQ1H24AFR-T2C parts). These GDDR5 modules are rated for 1.25GHz operation, so we expect to push them beyond the stock 1.0GHz configuration without hassle.
In order to feed the graphics card enough power, a pair of PCI Express power connectors are used (6-pin and 8-pin sockets). This is the same configuration used by previous dual-GPU graphics cards such as the GeForce 9800 GX2 and Radeon HD 4870 X2, so current high-end power supplies should cater for this new graphics card.
The Radeon HD 5970 naturally supports CrossfireX technology, and therefore in the standard position we find a single connector for bridging two cards together.
The only other connectors can be found on the I/O panel. Our sample featured two dual-DVI connectors and a Mini Display Port connection. Just like the Radeon HD 5870, the 5970 can support a maximum resolution of 2560x1600 on not one but rather three monitors.
Not one to just sit on the sidelines and let its rival get all the glory, Nvidia is hoping to build some anticipation for its upcoming DirectX 11 lineup by showing a picture of what appears to be the first working sample of a 40nm, GF100 desktop card based on the Fermi architecture. Posted on the company's Facebook account, the picture shows a black GeForce-branded card installed on an ASUS Rampage II Extreme LGA 1366 motherboard, running Uningine's Heaven DirectX 11 benchmark on a Dell 24-inch monitor.
