Intel introduces MAX Series CPUs and GPU featuring Sapphire Rapids-HBM and Ponte Vecchio - VideoCardz.com

« property merchandise »

Intel Max Series Brings Breakthrough Memory Bandwidth and Performance to HPC and AI

New household of products delivers CPU with precocious bandwidth memory, 4.8x faster than the competition, and Intel’s highest density GPU, to lick the world’s biggest challenges.

What’s New: In beforehand of Supercomputing ’22 successful Dallas, Intel Corporation has introduced the Intel Max Series merchandise household with 2 leading-edge products for precocious show computing (HPC) and artificial quality (AI): Intel® Xeon® CPU Max Series (code-named Sapphire Rapids HBM) and Intel® Data Center GPU Max Series (code-named Ponte Vecchio). The caller products volition powerfulness the upcoming Aurora supercomputer astatine Argonne National Laboratory, with updates connected its deployment shared today.

The Xeon Max CPU is the archetypal and lone x86-based processor with precocious bandwidth memory, accelerating galore HPC workloads without the request for codification changes. The Max Series GPU is Intel’s highest density processor, packing implicit 100 cardinal transistors into a 47-tile bundle with up to 128 gigabytes (GB) of precocious bandwidth memory. The oneAPI unfastened bundle ecosystem provides a azygous programming situation for some caller processors. Intel’s 2023 oneAPI and AI tools volition present capabilities to alteration the Intel Max Series products’ precocious features.

“To guarantee nary HPC workload is near behind, we request a solution that maximizes bandwidth, maximizes compute, maximizes developer productivity and yet maximizes impact. The Intel Max Series merchandise household brings precocious bandwidth representation to the broader market, on with oneAPI, making it casual to stock codification betwixt CPUs and GPUs and lick the world’s biggest challenges faster.”

–Jeff McVeigh, Corporate Vice President and General Manager, Super Compute Group astatine Intel

Why It Matters: High show computing (HPC) represents the vanguard of technology, employing the astir precocious innovations astatine standard to lick subject and society’s biggest challenges, from mitigating the impacts of clime alteration to curing the world’s deadliest diseases.

The Max Series products conscionable the needs of this assemblage with scalable, balanced CPUs and GPUs, incorporating representation bandwidth breakthroughs and agreed by oneAPI, an open, standards-based, cross-architecture programming framework. Researchers and businesses volition lick problems faster and much sustainably utilizing Max Series products.

When It’s Arriving: The Max Series products are slated to motorboat successful January 2023. Executing connected its commitments to customers, Intel is shipping blades with Max Series GPUs to Argonne National Laboratory to powerfulness the Aurora supercomputer and volition present Xeon Max CPUs to Los Alamos National Laboratory, Kyoto University and different supercomputing sites.

What the Intel Xeon Max CPU Delivers: The Xeon Max CPU offers up to 56 show cores constructed of 4 tiles and connected utilizing Intel’s embedded multi-die interconnect span (EMIB) technology, successful a 350-watt envelope. Xeon Max CPUs incorporate 64GB of precocious bandwidth in-package memory, arsenic good arsenic PCI Express 5.0 and CXL1.1 I/O. Xeon Max CPUs volition supply much than 1GB of precocious bandwidth representation (HBM) capableness per core, capable to acceptable astir communal HPC workloads. The Max Series CPU provides up to 4.8x amended show compared to contention connected real-world HPC workloads.¹

• 68% little powerfulness usage than an AMD Milan-X clump for the aforesaid HPCG performance.
• AMX extensions boost AI show and present 8x highest throughput implicit AVX-512 for INT8 with INT32 accumulation operations.²
• Provides flexibility to tally successful antithetic HBM and DDR representation configurations.
• Workload benchmarks:
  • Climate modeling: 2.4x faster than AMD Milan-X connected MPAS-A utilizing lone HBM.
  • Molecular dynamics: On DeePMD, 2.8x show betterment against competing products with DDR memory.

What the Intel Max Series GPU Delivers: Max Series GPUs present up to 128 X^e-HPC cores, the caller foundational architecture targeted astatine the astir demanding computing workloads. Additionally, the Max Series GPU features:

• 408MB of L2 cache – the highest successful the manufacture – and 64MB of L1 cache to increase throughput and performance.
• The lone HPC/AI GPU with autochthonal ray tracing acceleration, designed to velocity technological visualization and animation.
• Workload benchmarks:
  • Finance: 2.4x show summation implicit NVIDIA’s A100 connected Riskfuel recognition enactment pricing.
  • Physics: 1.5x betterment implicit A100 for NekRS virtual reactor simulations.

Max Series GPUs volition beryllium disposable successful respective signifier factors to code antithetic lawsuit needs:

• Max Series 1100 GPU: A 300-watt double-wide PCIe paper with 56 X^e cores and 48GB of HBM2e memory. Multiple cards tin beryllium connected via Intel Xe Link bridges.
• Max Series 1350 GPU: A 450-watt OAM module with 112 X^e cores and 96GB of HBM.
• Max Series 1550 GPU: Intel’s maximum show 600-watt OAM module with 128 X^e cores and 128GB of HBM.

Beyond idiosyncratic cards and modules, Intel volition connection the Intel Data Center GPU Max Series subsystem with x4 GPU OAM bearer committee and Intel Xe Link to alteration precocious show multi-GPU connection wrong the subsystem.

Intel Max GPU Series, Source: Phoronix

What Max Series Products Enable: In 2023, the Aurora supercomputer, currently nether operation astatine Argonne National Laboratory, is expected to go the archetypal supercomputer to transcend 2 exaflops of highest double-precision compute performance³. Aurora volition besides beryllium the archetypal to showcase the powerfulness of pairing Max Series GPUs and CPUs successful a azygous system, with much than 10,000 blades, each containing six Max Series GPUs and 2 Xeon Max CPUs.

In beforehand of SC22, Argonne and Intel unveiled Sunspot, Aurora’s Test Development System consisting of 128 accumulation blades. Researchers from the Aurora Early Science Program volition person entree to the strategy opening successful precocious 2022.

The Max Series products volition powerfulness respective different HPC systems captious for nationalist information and basal research, including Crossroads astatine Los Alamos National Laboratory, CTS-2 systems astatine Lawrence Livermore National Laboratory and Sandia National Laboratory, and Camphor3 astatine Kyoto University.

What’s Next: At Supercomputing ’22, Intel and its customers volition showcase much than 40 upcoming strategy designs from 12 archetypal instrumentality manufacturers utilizing Max Series products. Attendees tin research demos showcasing the show and capableness of Max Series products for a scope of AI and HPC applications, arsenic good arsenic perceive from Intel architects, customers, and end-users astir the powerfulness of Intel’s level solutions astatine the Intel booth, #2428. More accusation connected Intel’s activities astatine SC22 is available.

The Intel Data Center Max Series GPU, code-named Rialto Bridge, is the successor to the Max Series GPU and is intended to get successful 2024 with improved show and a seamless way to upgrade. Intel is past readying to merchandise the adjacent large architecture innovation to alteration the aboriginal of HPC. The company’s upcoming XPU, code-named Falcon Shores, volition harvester X^e and x86 cores connected a azygous package. This groundbreaking caller architecture volition besides person the flexibility to integrate caller IPs from Intel and customers, manufactured utilizing our IDM 2.0 model.

« extremity of the property merchandise »