Ampere Amping Up ARM-Based Server Processor Lineup With Higher Core Count Chips
Ampere recently revealed plans to expand its ARM-based cloud native processor lineup to include Altra Max processors with up to 128 cores that are socket compatible with the existing Ampere Altra chips. The new Altra Max processor is slated to begin sampling in the fourth quarter of 2020 with volume production in 2021. While Ampere’s Altra processors top out at 80 cores clocked at 3.3 GHz with a 250W TDP, Ampere is targeting up to 128 cores with Altra Max along with 250W TDPs which suggests that these processors will trade clockspeeds for core counts to stay within that 250W power budget. The Altra Max processors share the same 1P and 2P platform and socket compatibility with the existing Altra chips along with having the same 8 channel DDR4-3200 memory support (2 DIMMS per channel, max of 4TB per socket), L1 and L2 cache, 128 lanes of PCI-E 4.0 I/O (192 lanes in 2P systems, 64 of which can be cache-coherent CCIX, and supporting bifurcation down to x2), 2 x 128-bit SIMD, AI inference acceleration (INT8/FP16), ARMv8.2+ architecture and “server-class” RAS error checking and reliability features. The L3 cache, however, is still being evaluated and may change depending on production and yields of the higher core count parts according to Serve The Home.
Currently Altra spans from the Q32-17, a 32 core chip clocked at 1.7 GHz with 48-58W TDP to the Q80-33 a 250W chip with 80 cores clocked at 3.3 GHz. The lineup also includes 48,64, and 72 core parts. Altra Max will extend the core counts to 128 cores while still using the same 7nm process. Ampere has also completed tape-out of a 5nm test chip that represents the successor to the Altra and Altra Max lineup and is expected to launch in 2022.
Current customers for Altra include Packet and Cloudflare among others. Ampere is working to build up it’s distribution and OEM/ODM relationships.
These chips are aimed at data centers running web hosting, database, cloud storage, and other cloud computing workloads as well as with edge computing/AI workloads.
According to Ampere Founder and CEO Renee James: