Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizens Wu Yanzu in South China for the delivery of clues! CTOnews.com, November 10, Intel today officially launched the world's first x86 CPU--Intel Xeon Max series with HBM memory, a flagship product line we previously called Sapphire Rapids (HBM).

The 56 cores in Xinzhiqiang are all P cores, providing 112 threads and 350 W TDP. It adopts EMIB-based design and is divided into four clusters. But the most interesting thing is that it also has 64 GB of HBM2e memory, divided into four 16 GB clusters, with a total memory bandwidth of 1 TB / s, and each kernel has more than 1 GB of HBM.

Intel also said that the integration of HBM memory does not require code changes and should be seamlessly transparent to users.

"to ensure that there is no HPC workload lagging behind, we need a solution that maximizes bandwidth, computing, productivity, and ultimately impact. The Intel Max family of products brings high-bandwidth memory to the broader market, as well as an API, making code sharing between CPU and GPU easier and more effective in solving the world's biggest challenges.

-- jeff McVeigh, vice president and general manager of supercomputing group of Intel

According to reports, the 56 cores are composed of four Die and are connected by Intel's Multi-Chip Interconnection Bridge (EMIB), which encapsulates 64GB HBM memory. The platform will use PCIe 5.0and CXL 1.1I / O interfaces.

When the HCPG performance is the same, the power consumption is 68% lower than that of the AMD Milan-X cluster.

AMX expansion improves AI performance and provides peak throughput up to 8 times higher than AVX-512 for INT8 and INT32 cumulative operations.

Provides flexibility to run in different HBM and DDR memory configurations.

Workload benchmark:

Climate modeling: using only HBM is 2.4 times faster than AMD Milan-X on MPAS-A.

Molecular dynamics: on DeePMD, the performance is 2.8 times higher than that of competing products with DDR5 memory.

In terms of performance. It provides five times the performance in some workloads compared to the older Intel Xeon 8380 series processors or the AMD EPYC 7773X.

It is worth noting that AMD will release Genoa-based data center CPU tomorrow, and both series of CPU have been delivered to OEM, so these numerical comparisons are of little use.

The new CPU also includes 20 acceleration engines, mainly for AVX-512, AMX, DSA and Intel DL Boost workloads. Intel's performance in MLPerf DeepCAM training is said to be 3.6 times higher than that of AMD 7763 and 1.2 times higher than NVIDIA's A100.

According to reports, Intel Xeon Max CPU will make its debut in the Aurora supercomputer and is currently under construction at Argonne National Laboratory. CTOnews.com learned that the Aurora is expected to be the first supercomputer to exceed the peak double-precision computing performance of 2 exaflops.

In addition, Aurora will be the first to demonstrate the power of pairing Max series GPU and CPU in a single system, with more than 10000 "blades", each containing six Max series GPU and two extremely strong Max CPU.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.