Skip to main content

ATI Radeon HD 5870: DirectX 11, Eyefinity, And Serious Speed

Cypress Measures Up

ATI is really proud of its “Sweet Spot” strategy. First implemented last year, Sweet Spot mandates that ATI design a GPU capable of addressing the performance segment. The company is then able to leverage that GPU into a dual-processor solution, used to snag the performance crown. Then, it applies the same general design principles and spins off mainstream variants able to tackle sub-$100 price points.

Sweet Spot has worked out very well for ATI, and the company knew it as far back as its RV770 launch. Of course, history tells the complete tale: Radeon HD 4870 proved to be a great value, Radeon HD 4870 X2 was indeed top dog through the holiday season of '08, and the Radeon HD 4600- and 4500-series chips caught ATI up to Nvidia’s more mainstream offerings, which had previously been more attractive than the older Radeon HD 3800-series stuff.

This approach, of course, runs counter to Nvidia’s as-of-yet unnamed approach, which might as well be called the “Go Big Or Go Home” strategy. As we already know, GT200 was a 1.4 billion transistor, 576 square millimeter monster at 65nm. Nvidia didn’t seem particularly keen to talk dimensions when it made the jump to 55nm with GT200b. However, we’ve unofficially seen 490 square millimeters tossed around. Even at 40nm, the company’s next-generation part will likely be larger than today’s 55nm flagship, which means the thing will still be mighty substantial.

Ain’t It Sweet?

Gotta love the sweet spot...

The slide illustrating ATI’s approach with its Evergreen GPUs looks almost identical to the one used last year.

Code-named Cypress, the first (and most complex) Evergreen chip hugs the $310-$410 range. ATI plans to use a pair of the GPUs in its Hemlock design later this year. And while we've heard that power limitations will force lower clocks on the Hemlock part, ATI says it'll be defining everything from bill of materials to the BIOS on that card right up until launch. Juniper will also emerge in Q4, breaking under the $199 price point, while Redwood and Cedar follow up next year, introducing us to DirectX 11 support for less than $100.

ATI first tested TSMC’s manufacturing capabilities back in April with the Radeon HD 4770. At the time, we paired two of the cards up in a CrossFire configuration and found them unbeatable for $220. Unfortunately, the company cheesed off a number of prospective customers when it launched the card without enough supply in the channel. The problem? Reportedly, high-leakage parts from TSMC resulting in poor yields. Not much ATI could do, but we can’t imagine the etailers who slashed the prices on Radeon HD 4850s to stay competitive were very happy.

With five months passed, those teething issues have purportedly been worked out, as Cypress centers on the same manufacturing node. It’s a good thing too, as 40nm photolithography is a virtual necessity given a massive 2.15 billion transistor count. Equally amazing is that Cypress’ complexity is packed on a die measuring 334 square millimeters. Though quite a bit larger than RV770’s 263 square millimeter die, ATI’s most modern creation is still significantly smaller than GT200b at 55nm. Therein lays the advantage of more advanced manufacturing. ATI now has the most complex GPU in the world at 2.1 billion transistors, and it’s smaller than its principal competition at 1.4 billion.

Cypress sports more than two times the number of transistors found in its predecessor, which boasts 956 million. What on earth did ATI add to make its latest generation so much more complex?

Chris Angelini
Chris Angelini is an Editor Emeritus at Tom's Hardware US. He edits hardware reviews and covers high-profile CPU and GPU launches.