The future of Fusion and Llano

Posted by Aharon Etengoff

Silicon Valley chip expert David Kanter says AMD has "clearly articulated" both a software and hardware roadmap for unified CPU-GPU computing that contains "all the right pieces."  



"Llano and Zacate are the first integration steps and focus on reducing overhead by sharing – rather than copying – data between the CPU and GPU," the RealWorld Tech editor opined in a recent analysis. 



The future of Fusion and Llano"[Yes], Llano [did make] a number of compromises in favor of time to market, such as using a very mediocre CPU core and relying on memory rather than on-die caches for CPU/GPU communication. However, it still packs the best performance integrated graphics - period, [along with] good media decoding."

According to Kanter, Llano is a "serious improvement" for AMD's notebooks, mostly due to power management - which should be well received by consumers eager for longer battery life.



"The trick for AMD going forward is to execute on their vision and consistently deliver compelling improvements in a timely fashion. [As such], AMD [can be expected to] continue enhancing their integration - with unified power management across the CPU and GPU and more aggressive DVFS that accounts for temperature," he said.

"[Now, for the next-gen 32nm Trinity APU], AMD will likely begin to move more CPU/GPU communication on-die and take advantage of caching for lower power and better performance. Moving to a unified memory model is also possible, but seems more likely for the 3rd generation of Fusion. [In addition, the company] will upgrade to PCI-E gen 3, which runs at 8GT/s, for external I/O, possibly with extensions for coherency with discrete GPUs."



Kanter also noted that if AMD wanted to be "particularly aggressive," it could use 3D packaging to attach high bandwidth memory to Trinity, thereby significantly improving graphics performance. 



"One of the last real advantages of a discrete GPU is high bandwidth and dedicated memory. Even as little as 256MB of attached DRAM using WideIO or LP-DDR3 could bring GPU performance to a new level – at a time when programmable graphics will begin hitting its stride. 



"[Yet], there are still a number of thermal challenges to 3D integration, so 2013 or 2014 may be [a] more realistic estimate," he added.