This article offers some potential dates for these ARM devices.
----------------------
ARMH: Pondering the 64-Bit Server Future
By Tiernan Ray
ARM Holdings (ARMH) director of marketing Ian Thornton was in town today and was kind enough to stop by the Barron’s offices. We talked a bit about one of the next substantial frontiers for the company and its partners, 64-bit memory addressing, which is most likely to show up in server and equipment products, although Thornton also had some thoughts on 64-bit in mobile devices, which you can read in an earlier post.
ARM’s 64-bit architecture is coming with the “Cortex A50” series, which now exists in microprocessor designs designated “A53? and “A57.” To give you a sense of the timeline, licenses started to be signed with partners two years ago, and ARM has only lately “taped out” its first microprocessor designs. “They’ve gotten to the Verilog stage, but they’re not quite at the Netlist stage at this point,” says Thornton.
That would imply first samples of chips from partners sometime toward the end of this year, with volume production in 2014, notes Thornton.
The leading partners signed so far are startup Calxeda, ST-Microelectronics NV (STMPR), Broadcom (BRCM), Samsung Electronics (005930KS), and the semiconductor design unit of China’s Huawei, which makes equipment in addition to mobile devices.
There will be a cost and a power savings from the up-front price of these devices as well as the performance they enable in servers and equipment, said Thornton. At this point, though, it’s difficult to quantify all that relative to Intel (INTC) Xeon server processors or Advanced Micro Devices (AMD) Opteron parts because none of the partners has produced a part yet. X86 rules the majority of volume of server shipments today. (Mind you, AMD is also in the ARM camp, see below.)
One can make an analogy to Intel‘s “Atom” chip for mobile devices, in comparison to ARM-based smartphone chips based on the “Cortex A9” style parts. Atom takes three times the number of transistors to implement functional blocks that perform similar workloads, a sign of being less efficient than ARM-based designs, for example, contends Thornton. And the Intel x86 architecture uses the “variable length instruction word” approach, making it harder to streamline aspects of what are “monolithic” server processors, in his words.
But “that’s just an analogy,” and not really a fair comparison until 64-bit silicon appears, says Thornton.
His larger point about the competition between Intel and AMD and ARM partners is that the Intel server world is monolithic in its overall approach.
“Today, in some cases, you’ve got just 10% of Xeon being used” in a server, contends Thornton, because the bottleneck is in other parts of the system. If, for example, a server is doing streaming video, and it is mostly passing through bits from an external memory system, then between the time waiting for those bits to come off of a wire, and sending them out to the Internet, the microprocessor itself may not have a lot of work to do.
That, says Thornton, suggests that it may be possible for the entire collection of partners — there are 17 licensees of 64-bit ARM, and parties such as Nvidia (NVDA) have talked about getting into servers in a big way — could provide specialized parts that fit the job for different kinds of servers.
“It’s almost like an application-specific integrated circuit [ASIC],” quipped Thornton, although in that sense it seems more like an “application specific standard product [ASSP],” I suggested, though in reality such a system-on-a-chip(SOC) might not really be either.
Of course, there remain those niggly little details: Software. That huge chunk of the server market running on x86 still uses a lot of legacy applications, such as back-office financial software, even though some workloads have shifted to newer Internet-style applications. Thornton says that when customers tell large application providers such as SAP AG (SAP) or Oracle (ORCL) that they want to run their apps on an ARM-based server, the natural response from the software giants might be “what’s in it for,” when it comes to all the expense involved in porting software to ARM. Still, he thinks that “eventually everyone comes together to do what’s right for the industry,” which, in ARM’s view, includes moving to ARM processors, obviously.
Notes Thornton, there is the open-source software project called Linaro, involving upward of 200 engineers from ARM and its partners, that is helping to standardized Linux for ARM in a variety of contexts, including networking and other equipment.
ARM shares today are up $1.35, or 3.4%, at $41.45.
Update: Note that I failed to mention above that AMD is also making a push into ARM-based server chips. The company last year bought micro-server designer SeaMicro and has talked extensively about its intentions for ARM-based systems.
AI
