As many of you may be aware, Cayman is the codename given to AMD’s newest flagship GPU, which will power the Radeon HD 6900 series (Antilles, aka the Radeon HD 6990, will uses a pair of Cayman GPUs when it arrives sometime next year). Cayman is the successor to the Cypress XT that was employed on the Radeon HD 5800 series, and while it does leverage many of the technologies from that generation, Cayman is a new chip that embraces a fresh core architecture designed to increase efficiency, performance and feature-set.
Cayman does not, however, employ a new manufacturing process. AMD’s typical M.O. is to release a chip using a given process, revise that chip at some point during its life-cycle to enhance performance and/or power, and then release a totally new chip built using a more advanced process 12 – 16 months later. This time around though, TSMC, AMD’s GPU foundry partner (for now) simply isn’t ready with their next-gen 28nm process, so Cayman, like Cypress, is built at 40nm. Because of this, Cayman is actually the largest graphics core that AMD has built to date.
Initially, the Cayman GPU will power two graphics cards, the Radeon HD 6970 and the Radeon HD 6950. And wouldn’t you know it; we have a pair of each in house and have been testing them with a completely revamped test-bed for a few days now. Take a moment to peruse the specs and soak up the first couple of pics of the new Radeon HD 6900 series cards below. Then we’ll dive into the particulars and check out their performance.
Feast Your Eyes On The Radeon HD 6970
And Here's The Radeon HD 6950
As you can see in the two images above, the new Radeon HD 6970 and 6950 look virtually identical from the front, save for their model number decals. We suspect that those specifications may leave some of you scratching your heads, though. If the Radeon HD 6970 is supposed to be an upgrade over the 5870, why does it have a lower total number of stream processors? The Radeon HD 6970 has 1536 ALUs versus 1600 in the Radeon HD 5870, and the 6950 has “only” 1408 ALUs. Well, the answer to that question lies in AMD's new "VLIW4" architecture, which we’ll explain a little later.
Disregarding the shader counts for a moment, you can see the Radeon HD 6970 offers compute performance in the neighborhood of 2.7TFLOPs (single precisions) and 675GFLOPs (double-precision) and the Radeon HD 6950 offers 2.25TFLOPs / 564GFLOPs. The 6970's GPU is clocked at 880MHz and the 6950's at 800MHz. The Radon HD 6970 sports 96 texture units while the 6950 offers 88 (up from 80 in the 5870) and each has 32 ROPs. Both cards are also outfitted with 2GB of GDDR5 memory that utilize a 256-bit interface. The memory speed on the 6970 tops out at 5.5Gbps (1375MHz), while the 6950 hums along at 5.0Gbps (1250MHz). That puts memory bandwidth in the 175GB/s and 160GB/s ranges for the Radeon HD 6970 and 6850, respectively, which is a big step up from the 153GB/s of the 5870. Max board power is 250W for the 6970 and 200W for the 6950, and both have the same digital video output configuration.