AMD Southern Islands: Released later this year

The manufacturer's official statement contains several relevant information.

AMD senior vice president Rick Bergman has personally announced that the next generation of video cards, codenamed Southern Islands, will hit store shelves later this year. In addition, the first specific technical information was published. So far, it was known that representatives of the generation would be made on the TSMC 28 sqm production lines, however, it turned out that the applied architecture would be completely different from the previous ones. Surely many still remember the end of last year when the Cayman tile was released. It also underwent an architectural modification at the time, where the VLIW5 design was replaced by VLIW4. Now a much larger change has taken place.

AMD engineers have thrown the VLIW (Very Large Instruction Word) structure itself, which has been present since the Radeon HD 2000 series, in the trash and use a new, more efficient solution instead. They primarily use scalar units, all of which will be able to count independently. The Front-End and Cache structure will also be completely redesigned to increase efficiency. Geometric calculations will run in parallel, which will have a beneficial effect on tessellation. Another interesting fact is that the SIMD units used so far are also a thing of the past, they migrate to a so-called Compute Unit (CU). Such a CU will hide four SIMDs and one scalar unit. A CU will be able to perform as much as a single Cayman SIMD unit, and will also be able to count more efficiently. As far as the memory area is concerned, the ECC error correction is also included in the formula here. Incidentally, this was introduced by NVIDIA in the Fermi generation, so it is not an unknown feature. The queue is closed by so-called ACE (Asynchronous Compute Engine) units. These will be responsible for various calculations, but the point is that in previous generations they were also responsible for 3D rendering, but now the latter task has been taken off their shoulders.

Finally, AMD said it would be a completely redesigned architecture, but reassured that it would not significantly increase manufacturing costs. More information is expected in the next few months. It is not yet known whether the company will remain with the name used so far (Radeon HD xxxx) or whether a major change may be planned here as well. Based on the above, we can say that AMD is planning a thorough reform of the video card field, as the method introduced by Radeon HD 2000 and used since then (VLIW) is one-in-one to create a system with significantly better efficiency. Even taking into account the fact that engineers have been working on the new architecture for several years, it’s no exaggeration to say that AMD is preparing for a very big step, a big bang.