Intel on Wednesday formally launched the Intel Max Collection product household for high-performance computing (HPC) and synthetic intelligence (AI). Comprising the Intel Xeon CPU Max Collection (code-named Sapphire Rapids HBM) and the Information Middle GPU Max Collection (code-named Ponte Vecchio), the brand new merchandise will energy the upcoming Aurora supercomputer at Argonne Nationwide Laboratory.
When the Aurora supercomputer comes on-line in 2023, it is anticipated to be the primary supercomputer to exceed two exaflops of peak double-precision compute efficiency. It’ll even be the primary system to showcase the ability of pairing Max Collection GPUs and CPUs collectively – the supercomputer will home greater than 10,000 blades, every containing six Max Collection GPUs and two Xeon Max CPUs.
Whereas the Max Collection will make a splashy debut at Argonne, the product household is de facto designed to convey HPC and AI to a broader market, Intel stated.
“We’re making an attempt to assist resolve the world’s largest challenges quicker and try this in a sustainable approach,” Jeff McVeigh, Intel’s company VP and GM of the Tremendous Compute Group, stated to reporters forward of the announcement.
Additionally: Google makes large dedication to assist extra languages utilizing AI
Issues like local weather change and world well being crises will be addressed extra shortly with the assistance of HPC and AI, he famous. There is a “vast variety” of workloads within the HPC and AI area, he added. Some are closely compute-bound, whereas others have excessive constraints on reminiscence bandwidth.
“Our view is, all these are necessary, and we have to have the correct options to handle them,” he stated.
The CPUs and GPUs are united by oneAPI, Intel’s open, standards-based, cross-architecture programming framework. Researchers and companies will resolve issues quicker and extra sustainably utilizing Max Collection merchandise.
This is extra on the Xeon Max CPU, the primary and solely x86-based processor with excessive bandwidth reminiscence:
- Provides as much as 56 efficiency cores constructed of 4 tiles
- Comprises 64GB of excessive bandwidth in-package reminiscence
- Supplies greater than 1GB of excessive bandwidth reminiscence (HBM) capability per core, sufficient to suit commonest HPC workloads
- Supplies as much as 4.8x higher efficiency in comparison with competitors on real-world HPC workloads
- 68% much less energy utilization than an AMD Milan-X cluster for a similar HCPG efficiency
The Max Collection GPU, in the meantime, is Intel’s highest density processor, with greater than 100 billion transistors in a 47-tile package deal:
- Provides as much as 128 gigabytes (GB) of excessive bandwidth reminiscence
- Delivers as much as 128 Xe HPC cores
- Options 408MB of L2 cache and 64MB of L1 cache for greater throughput and efficiency
- Options native ray tracing acceleration, for scientific visualization and animation
Along with powering the upcoming Aurora supercomputer, Intel will ship Xeon Max CPUs to Los Alamos Nationwide Laboratory, Kyoto College and different supercomputing websites.
In the meantime, Argonne and Intel unveiled Sunspot, Aurora’s Check Improvement System consisting of 128 manufacturing blades. Researchers from the Aurora Early Science Program may have entry to the system starting in late 2022.