QCOM: 1st data center CPU, 10nm Centriq 2400, shipping.....................................
Putting aside its legal battles and lawsuits for a few hours, Qualcomm today said is it shipping the Centriq 2400 – its ARM-based server-grade processor, and the world's first 10nm data-center CPU.
"This chip is now shipping for revenue," said Anand Chandrasekher, the boss of Qualcomm Datacenter Technologies, adding that the silicon is likely to be deployed in production by customers over the next 12 months.
Previous ARM-compatible server CPUs have failed, notably the Calxeda parts, because, basically, they were 32-bit. Qualcomm's Centriq is, crucially, 64-bit as well as ARMv8-A compatible, multicore, draws up to just 120W, has suitably fat caches, and server-friendly IO and memory interfaces, and is aimed at data-center workloads. It took more than four years to produce, we're told.
It's aimed at single-socket cloud servers and cloud builders: think Azure, Google, Tencent, Baidu, Alibaba, and so on. Microsoft and Google have, off the top of our head, publicly expressed a desire to power its cloud platforms using Qualcomm-designed silicon to break free from Intel and its price gouging.
Certainly Google, which buys chips by the boatload, is seemingly eager to deploy anything-but-Intel in its warehouses of computers, which should at least leave Chipzilla a little worried.
“Google is excited to see Qualcomm Datacenter Technologies launch the Qualcomm Centriq 2400 processor,” Bart Sano, veep of platforms at the ads giant, said. “We welcome choice in the processor design space for data centers. Choice leads to innovation which ultimately benefits our users. The 64-bit ARMv8-A architecture and ecosystem is now a viable alternative for scale-out data center designs.”
Microsoft is also enthusiastic... in sending a coded message to Intel.
“Qualcomm Centriq 2400 is well suited for highly parallelized workloads like those found in hyper-scale clouds," said Mike Neil, corporate veep of Azure Infrastructure and Management at Microsoft. "We believe that Qualcomm Datacenter Technologies has the ability to deliver extremely compelling performance per watt and total cost of ownership.”
Azure, we're told, could end up using Centriq in cloud storage and Bing search systems among other infrastructure things.
We've covered the Centriq in depth, here, where we also revealed that Qualcomm had moved engineers from its Snapdragon mobile side to the data-center chip team to complete the designs, signaling a seismic shift in focus for the business. Our colleagues at our sister site, The Next Platform, have also detailed the Centriq architecture here and here. It goal is to be high performance and power efficient.
Here's a quick summary, though, of the specifications, according to Qualcomm:
- Fabricated using Samsung's 10nm FinFET process with 18 billion transistors on a 398mm2 die in a system-on-chip package.
- Up to 48 single-thread cores running at 2.2GHz, peaking to 2.6GHz. It is 64-bit-only: there is no 32-bit mode.
- The cores are connected internally on a bidirectional ring bus with an aggregate bandwidth of 250GBps.
- 512KB of shared L2 cache for every core pair on the bus.
- 60MB of unified L3 cache spread out along the ring interconnect.
- Six DDR4 RAM channels that can interface up to 768GB of memory per processor.
- 32 PCIe gen-3 lanes and six controllers.
- ARM TrustZone for secure boot.
- A power draw of up to 120 watts.
- A $1,995 price tag for the 48-core part.