Applied’s X-Gene (below) will initially pack 8 custom 64-bit ARM cores on a single 40nm chip running at 2.5 GHz, each with its own L1 cache and shared L2 and L3 pools as well as I/O, networking and an interconnect fabric. A follow up 28nm part will pack 16 cores and run at 3 GHz.
The chip’s interconnect initially can link up to 128 cores in a system and eventually 512 cores. The company showed a mockup of a server card but does not have first silicon back from TSMC.
“They have been very quiet since they announced X-Gene almost a year ago, so the fact they stood up and described the same SoC and it still will be available by the end of the year means they are executing on their plan,” said Nathan Brookwood, principal of market watcher Insight64 (Saratoga, Calif.).
The ARM chips are showing levels of integration not in their x86 competitors, including clustering interconnects AMD and Intel may not have for another two years, he added.
Looks like a study in contrasts - the Power7 designers continuing to go after the highest speed - 5+ GHz, 80MB caches, the AppliedMicro designers going after integration, efficiency and right sizing and Intel doing what it does best - leading edge fabs that give it the advantage. It appears that all three are targeting different segments of the server market.
Oh and Pat comes across as a petty partisan. Maybe he should run for political office.
Good to see that Oracle is continuing the Sparc line though there were fears of it's imminent demise. Sparc lost the battle for supremacy a decade back which made it easy for X86 to proliferate but these new chips could bring about some refreshing change in server market. How low can they make a server to own while still providing the reliability and throughput is the question.
The Core is a tiny portion of any Server, so "which core ?" is asking the wrong question.
More important are GB/s/watt numbers, and longevity and process.
Servers is a simple numbers game, there is no marketing sizzle, or packaging design froth.
That means the big players, with strong FAB backup, will continue to be the main forces here.
It is also why highly focused solutions, like IBMs, can find a place.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.