IBM will halt development of its next-generation PowerXCell 32iv chip. According to the company, the design of the processor did not provide expected performance and Cell would reappear in another form.
The next-generation Cell processor – the PowerXCell 32i – for high-performance servers or supercomputers was projected to feature four PowerPC processor elements (PPE) as well as 32 synergistic processing elements (SPEs). However, the chip will never see the light of the day. According to David Turek, vice president of deep computing at IBM, parts of Cell design would reappear in another form, but the high-ranking executive remains tight lipped about this form, reports Heise.de.
Current PowerXCell 8i chip aimed at servers and supercomputers features radically improved double-precision floating-point performance (compared to typical Cell processor found inside Sony PlayStation 3) on the SPEs from a peak of about 12.8Glops to 102.4Gflops total for eight SPEs. Potentially, the PowerXCell 8iv could achieve about 500Gflops double-precision floating-point performance. However, it might make little sense for IBM to release such a chip since, for example, modern graphics processing units (GPUs) are already achieving such speed levels at lower cost: IBM has said numerous times that Cell chip had insufficient yield.
It will take some time before GPUs will find themselves into supercomputers created by IBM. However, already now IBM claims that the current form of Cell has to be redesigned, which means we can well see other Cell concepts in future.
It remains to be seen how co-developers of the original Cell concept will react on IBM's decision to halt development of next-gen server-specific chip.