Intel’s Arctic Sound M is being announced today as the Intel Data Center GPU Flex Series.
Going back months Intel has been talking up Arctic Sound M (ATS-M) as a “media supercomputer” that would be shipping in mid-2022 and be able to handle video transcoding of eight simultaneous 4K streams, 30+ 1080p streams, 30+ cloud gaming streams, and 150 TOPS for media AI analytics. Today Intel is announcing Arctic Sound M under the Data Center GPU Flex Series branding.
An earlier slide on Arctic Sound M… Now known as the Intel Data Center GPU Flex Series.
Arctic Sound M as reported many times in the past is built off their DG2 (Alchemist) Xe HPG architecture. The GPU Flex Series aims to offer better TCO in data centers than offerings from NVIDIA or AMD. Intel also is heavily promoting their open-source software stack with oneAPI for the Flex Series as being more robust than the likes of NVIDIA’s proprietary CUDA.
Unfortunately I don’t have any ATS-M hardware here for testing today, but Intel’s announcement promotes the Flex Series GPU as having 5x media transcode throughput and 2x decode throughput at half the power consumption of NVIDIA’s A10 compared to the new Intel Flex Series 140 GPU. Intel’s announcement later goes on to note that the Flex Series 140 consists of two Xe HPG GPUs on one card. Intel says up to 10 Flex Series cards can fit within a 4U server configuration so in turn being able to support up to 360 1080p streams per server for HEVC content.
In addition to talking up their open-source software stack, Intel’s Data Center GPU Flex Series announcement also heavily talks up their hardware-based AV1 encode capabilities. The GPU Flex Series is initially targeting the spaces of video encoding/transcoding as well as powering Android cloud gaming. Later on they aim to embrace Windows cloud gaming, AI, and VDI workloads. Much of the ATS-M information being announced today has previously been shared in various forms over the prior months.
Intel didn’t announce any GPU Flex Series pricing today but says over the coming months there will be systems featuring these data center graphics cards from the likes of Supermicro, Lenovo, Inspur, HPE, Dell, and others. Today’s announcement overall was a bit light on technical details but hopefully more will be revealed soon and the ability to get hands-on with the hardware for Linux testing.
That’s the short story for today’s Data Center GPU Flex Series announcement. Unfortunately no hardware today for benchmarking to evaluate the video performance or overall Linux support. Separately, for those wondering more broadly about the Intel DG2/Alchemist Linux support in its current shape, beginning tomorrow I’ll have up information on the Arc Graphics A380 under Linux around driver support and moving onto the Linux driver performance. I am also waiting to hear back from Intel about their GPU Flex Series support plans for enterprise Linux distributions. While they have been upstreaming the DG2/ATS-M bits, they are all on the bleeding-edge Linux kernel and other components so will be interesting to see their back-porting or packaged driver approach for supporting these discrete graphics cards on the enterprise Linux distributions — since after all that will be the primary target for these new data center wares, not bleeding edge distributions like Arch Linux. Stay tuned for more about the Intel discrete graphics support and performance on Linux.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal tips are also graciously accepted. Thanks for your support.