Intel Introduces Xeon Max & Data Center GPU Max Series
With SC2022 kicking off next week and AMD set to unveil their next-generation server processors tomorrow, Intel is using today to announce the Xeon Max Series and the Data Center GPU Max Series.
The Intel Xeon CPU Max Series is their product branding for the CPUs formerly known as the Sapphire Rapids HBM parts for featuring onboard HBM2 memory.
Intel is reaffirming today as part of the Xeon CPU Max Series that the initial parts will feature up to 56 P-cores, up to a 350 Watt TDP, CXL 1.1 support, and 20 accelerator engines built-in.
Very exciting is the Xeon Max Series will feature up to 64GB of HBM2e memory using four stacks of 16GB HBMe2. The Xeon Max Series should yield around 1TB/s of effective memory bandwidth and provide more than 1GB of HBM memory per CPU core. It's also interesting Intel confirmed that these Xeon Max Series processors will be able to boot simply by themselves and do not need any DDR5 system memory attached. The Xeon Max Series can run in HBM-only memory mode without any additional memory, an HBM Flat mode where there are two memory regions shared with the DDR5 DIMMs, and then lastly an HBM caching mode where HBM acts as a cache for the main DDR5 system memory.
While AMD's Milan-X with a large L3 at 768MB per socket yielded impressive results, with 64GB of onboard HBM2e memory this will be quite fascinating to see how well it performs once we have our hands on it for independent benchmarking.
Sapphire Rapids also has the Advanced Matrix Extensions (AMX) and various accelerator blocks to also crank up the performance, beyond simply the HBM2e memory.