Intel Launches 4th Gen Xeon Scalable "Sapphire Rapids", Xeon CPU Max Series

In addition to the accelerator prospects with Sapphire Rapids, equally exciting is the Intel Xeon CPU Max Series for Sapphire Rapids with 64GB of HBM2e memory onboard. These Sapphire Rapids HBM SKUs can operate purely off the high bandwidth memory, in a flat mode in conjunction with conventional DDR5 system memory, or as an HBM caching mode that mixes the HBM-only and flat modes of operation.

Unfortunately I don't currently have my hands on any Xeon CPU Max Series CPU for testing the performance but should be interesting especially for the wide range of workloads I benchmark as well as from the prior Milan-X testing with the plethora of benefits observed in those benchmarks, with 64GB of HBM2e memory should be even more dramatic gains.

Unlike needing to make software changes to take advantage of the accelerators, with the Xeon CPU Max Series there isn't that requirement and thus a lot of performance out-of-the-gate for memory bandwidth optimized workloads.

Out of the gate, there is likely to be better availability of 4th Gen Xeon Scalable processors and servers/platforms than what we have seen from 4th Gen EPYC. Two months after the 4th Gen EPYC "Genoa" was introduced it's still difficult finding the new EPYC 9004 processors being readily available from major Internet retailers -- it's likely to be a better story within large organizations but for individuals and small organizations it can be difficult getting your hands on Genoa right now. There have been a few more Genoa CPU listings in recent weeks but still seems to be limited. Additionally, EPYC 9004 series motherboards and server platforms are still rare to see listed for easily purchasing online in single quantities.

Intel server motherboard partners have been ready for some time with their Sapphire Rapids offerings but setback by the CPU launch delays, so it's looking like Intel may have the advantage of more robust availability while we'll see how that plays out over the coming weeks and next few months. It will also be interesting to see how motherboard and server pricing compare between Genoa and Sapphire Rapids as well as how close the CPU pricing plays to the suggested customer prices.

That's the summary of today's announcements around Intel 4th Gen Xeon Scalable "Sapphire Rapids" and the Xeon CPU Max Series. Sapphire Rapids is very interesting from the hardware perspective but how well it can compete in the short-term may be more challenging as with Sapphire Rapids it's all about accelerators, accelerators, and accelerators. While the upstream Linux support for these new accelerator blocks is in place, it will take time for more upstream software to begin making use of the new accelerator potential, initial deployment/setup headaches, etc, so the full potential may not be realized for some time -- just as it took a while for initial adoption around AVX-512 and overcoming those early hurdles. It will also be interesting to see how Intel's On Demand strategy plays out, again, due to accelerator use being so important for Sapphire Rapids to compete and really live up to the performance claims focused on by Intel while many of the lower-tier SKUs are running at reduced accelerator engine counts or even none by default.

Due to the early adoption issues around software making use of the accelerators, the Intel Xeon CPU Max Series is in a very interesting position. With the 64GB of HBM2e memory and the ability to even run servers just off that HBM2e memory, there is very interesting prospects for exceptional performance in memory intensive bound applications without having to worry about any software requirements or other code changes. The only accelerators with the Xeon CPU Max series is the Data Streaming Accelerator. The Xeon CPU Max Series 9480 model also only retails for $12,890 in comparison to the flagship Xeon Platinum 8490H at $17,000. Given the great uplift I've reported on from AMD Milan-X from a wide variety of real-world software, the Xeon CPU Max Series should be a real treat for the HPC space and other memory bandwidth intensive workloads especially if your RAM requirements allow you to run the entire server in HBM2e-only memory mode.

Continue reading with the next article looking at the Intel Xeon Platinum 8490H review and benchmarks.