AMD Radeon HD 7770 review

Southern Islands architecture with a friendly price tag

TODO alt text

HD 7770

The Graphics Core Next architecture is a bold move from AMD in that it represents a move away from the VLIW instruction used in 6-series cards towards a GPU-processing-friendly SIMD vector processor.

The previous processor type was great for graphics processing, but not suited to general purpose GPU computing - AMD left that side of things up to Nvidia and its CUDA cores.

Graphics Core Next is a u-turn on that philosophy though. GCN allows up to 16 data elements to be processed in a single clock cycle.

Grouping data before it runs through the vector processor is really efficient when dealing with general processing tasks - but the bad news for games is that you won't notice that difference in Battlefield 3 - the strengths of this architecture are wider-reaching than that, even as far as the professional market.

GCN also understands advanced languages like C++, meaning that in the long run, it'll be easier for developers to make use of the 7-series cards for complex programs.

The performance improvement from this architecture comes from passing data through a ton of compute units, which all work on the same operation until it's completed, and the resulting compute performance of this HD 7770 card is impressive at 1.28 TFLOPS.

It's built with ten compute units rather than the HD 7970's sixteen, but that's still enough to demonstrate a marked performance increase on last generation's equivalent model. And with AMD and Nvidia now adopting similar stances in their design, it's becoming an increasingly straight battle between the two - no hiding behind the blurred lines of CUDA cores and stream processors.

So the number of compute units and the simplified SIMD instructions they perform give AMD's 7-series cards the brains, but the clock speed is still the brawn of the operation. And at a world-first 1,000 MHz, it's fair to say the HD 7770 has brawn in check.