As we noted earlier, AMD has taken the stance that this release is about bringing more cost and power-efficient graphics cards to market. Obviously this is a good thing but let's not sip too deeply on that kool-aide just yet. In plain, non-marketing speak, what we're really looking at here is a fairly aggressive redesign but targeted in the same manufacturing process technologies as the previous generation. The AMD Radeon Graphics team is coming out with this new GPU architecture right around the 14 month mark, which is about on par with their traditional cadence of design cycles. The cold hard facts of the matter is that, process technologies below 40nm for very large, complex designs like graphics processors, do not come easy and AMD had
to make due with the semiconductor manufacturing geometries that were going to be available in full production at their target release time frame.
So, in reality, doing more with less was not only the flag that marketing was going to wave but also it was very much what the Radeon graphics design team had to live by. Fitting actually, for this new post-recession (we're told we can call it that now...) economy we live in. As a result, focusing on not only performance-per-watt improvements but also performance per square millimeter of die real estate, where size matters from a cost standpoint, was also a key design goal for Radeon graphics chip architects. Through the culmination of these efforts, AMD is claiming their new 6800 series cards are going to be the fastest sub-150W cards on the market, at decidedly lower price points. We'll confirm this of course in our testing sections coming up.
But we digress, Barts, Cayman, Antilles; let's get back to warmer thoughts. There's frost on the pumpkin these days after all. The above slide from AMD's press deck shows where the new chips fall. Notice the relative size of each chip, as well as the number of each. Barts was designed to give near-Radeon HD 5870 performance at a lower price point, at its top end model but Cayman is the big Daddy island we're looking forward to sailing to sometime in Q4. Take two Caymans and you've got Antilles. Confused? Don't be. Follow along here...
As you can see on the above performance and release timeline, the Radeon HD 6870 and Radoen HD 6850 slide in just underneath their respective previous generation counterparts, if you consider model number conventions versus performance. However both of these devices have beefed up tessellation units, as we mentioned, as well as improved thread management and buffering versus the previous generation. Finally they offer these enhancements and performance characteristics in roughly 25% less silicon area. Which means, as noted before, these cards will cost less and consume less power, but also these optimizations afford room to roll in an even higher-end GPU later this year, which will be Cayman.
Cayman will be significantly faster than AMD's current flagship Radeon HD 5870 GPU is now, with the high-end Radeon HD 6970 approaching the performance that a dual-GPU Radeon HD 5970 offers today but again, with a better tessellation engine and certainly lower power than the previous generation dual-GPU card. Expect the Radeon HD 6970 to arrive first, then a lower power/cost Radeon HD 6950 and finally, perhaps some time in Q1, Antilles will emerge as the Radeon HD 6990. Antilles will be comprised of two Radeon HD 6950 GPUs on a single card.
So there you have it, the roadmap. Now let's look at what we have in hand today. Here are the reference cards AMD sent in for testing.
AMD Radeon HD 6850 - 775MHz Core, 960 Stream Processors, 1GB GDDR5 @ 4Gbps
AMD Radeon HD 6870 - 900MHz Core, 1120 Stream Processors, 1GB GDDR5 @ 4.2Gbps
Two miniDP, one HDMI, two DL-DVI and a fairly big slab of PCB; the Radeon HD 6850 and Radeon HD 6870 measure in at 9-inches and 9.75-inches long respectively. Both cards share the same display output configuration, which consists if dual DVI outputs (one single link, one dual link), and HDMI 1.4 output, and two mini-DP 1.2 otuputs. while the 6870 takes a pair of 6-pin PCI Express power connectors to the 6850's single connector. The primary differences between the higher-end Radeon HD 6870 and the Radeon HD 6850, are the number of stream processors enabled in the GPU, core clock speeds and memory interface speeds. The Radeon HD 6870 is comprised of 1120 stream processors clocked at a blistering 900MHz. The 6870's memory is running at 1050MHz or 4.2GHz quad pumped (GDDR5) which equates to 4.2Gbps for 134.4GB/s of memory bandwidth across its 256-bit interface. The Radeon HD 6850 is built with 960 stream processors running at 775Mhz and a slightly slower memory interface speed of 1GHz or 4GHz GDDR5, which results in a 4Gbps data rate or 128GB/s of memory bandwidth.
Let's look at some retail level product, shall we?