Like the Arc A series in the consumer gaming market, it usesXe architecture, DG2 Alchemist core,integratedWith up to 32 Xe cores, 32 optical tracking units, 4 Xe media engines, XMX AI acceleration units, and AV1 hardware codec, it is also Intel’s first independent GPU in the data center field that supports AV1 codec.
For video transcoding,It can process more than 30 1080p video streams simultaneously, and in terms of cloud gaming,Can satisfy more than 40 gamers at the same timeIn addition, it supports 62 virtualization functions and AI inference computing power of 150 trillion operations per second (150 TOPS).
ATS-M adopts a fully open concept, developers can use the open software stack supported by oneAPI to carry out related design work, and the system and software adaptation work is also progressing rapidly.
It will be paired with the fourth-generation Xeon Scalable processor code-named Sapphire Rapids, and each node supports two ATS-Ms and two Sapphire Rapids.
ATS-M has two different forms of design, one is a single-chip, for workloads requiring peak performance, with a power consumption of 150W; the other is a dual-chip, for high-density, multi-purpose workloads, with a power consumption of only 75W.
Intel ATS-M graphics cards will be released in the third quarter, has now acquired more than 15 designs fromDell, AMD, Cisco, HPE, Inspur, H3Cand other partners.
At the same time, Intel further introduced the Xe HPC architecture, the first high-performance computing accelerator card code-named Ponte Vecchio, and the supercomputer “Aurora” based on it and Sapphire Rapids Xeon Scalable processors.
The supercomputer is jointly built by Intel, Hewlett Packard Enterprise, the U.S. Department of Energy, and Aragon National Laboratory.The latest disclosed computing power has exceeded 2EFlops, which is 20 billion billion times per second.
When it comes to Sapphire Rapids Xeon processors, Intel recently revealed,It will integrate accelerators for different purposes such as AI, encryption, networking, data centers, etc., optimized for rapidly growing workload types to meet massive data processing demands.
it will alsoThe first integrated package of HBM2E high-bandwidth memory, up to four, with a total capacity of up to 64GB, but this is optional.
Sapphire Rapids has already started shipping the first models.
Four HBM2E memory for Sapphire Rapids and the edge