AMD Radeon HD 6870 Final Thoughts

Every now and then, tech companies manage to pull a rabbit out of a hat. I don't know exactly how they do it, or else I'd be rich and famous, but I believe its equal parts chance and effort. I know that runs counter to the words of the great Thomas Edison, who certainly pulled out more than a few rabbits in his day. As a product reviewer, nothing makes us happier than having said rabbit delivered to our door via FedEx. It's great when you can be a material witness to the impossible dream. It's exciting, and it's contagious; everyone wants in on the act. Every time I get a new product to evaluate, I ask myself if this thing looks like a rabbit or a tortoise. Sorry to mix my metaphors, but a great deal of progress is also made by products that move the ball down the playing field ten yards at a time. I have a lot of respect for those products because sometimes it's tougher to make the next first down than it is to make the touchdown.

By now, the story is perfectly crystallized: Barts, as in St. Barts, is an incremental product. Taiwan Semiconductor Manufacturing Company (TSMC) failed to deliver the 32nm technology node to its customers on time. At some point they just gave up and decided to skip it altogether, leaving some major customers with a real problem on their hands. The holiday buying season waits for no man, so AMD punted and took what they knew about the 40nm fabrication process back to the drawing board. They had used very conservative design rules for their Radeon HD 5xxx series of GPUs, and with one year of production under their belt they knew where they could stretch the rules a bit and get away with it. They also knew that they had left a gaping hole in their product line at the upper end of the mid-market. So, they bit the bullet and renumbered the product line to slot some new cards in the gap that the HD 5830 had failed to fill. Just like that, the HD 6870 was created, or at least the concept for it.

As I said in my reviews of the GTX460, that chip is really comparable to an HD 5850 from a technology standpoint, and NVIDIA chose to sell it at a price point that was occupied by a lesser model, the HD 5830. Sounds like a good marketing plan, especially since every Cypress-based card and every GF104-based card share the same cost structure. That cost is based strictly on the technology node, the manufacturing platform used to produce it, and the size of the die. Sure, you can add or subtract features, but the fundamental production costs are comparable for both chips, even if the performance is not. At the time I thought, "All ATI needs to do is lower prices on their midrange cards, and the compelling case of the GTX460 just goes away." Well, AMD has gone one better. Not only have they reduced the selling price for 5850-class performance, they've also reduced their production costs by achieving the same level of performance with a GPU die size that's about 30% smaller than a Cypress or GF104.

When NVIDIA designed the GTX460 reference card, they went for the simplest design and lowest part count in order to keep the cost down. I didn't see the same thinking with the Radeon HD 6870 reference design. To me, it looked more like a "halo" product; one that is meant to impress. I am definitely NOT the marketing guy, but I thought that role was reserved for the high-end product line. Now that the non-reference cards are out, we're seeing some simplification of the design, and some price drops. The GPU itself has a 30% cost advantage over the Cypress part, and now the rest of the video card's design has been pared down to match it. The other good thing that has happened is software control of the GPU core voltage. More than one vendor is offering support for voltage control, and so the HD 6870 can finally compete on equal terms with the GTX460.

<< Previous Page - Next Page >>