Open Call for Large Models’ Training Lab to Accelerate AI Research (Israel)

It should be noted that the goal of the R&D lab is not to replace existing infrastructures and cloud services in Israel, but rather add value and new capabilities that cannot be implemented today in existing R&D infrastructures, at a relevant financial cost-effective scale to users while still offering performance that meets the needs, as detailed below.

They invite applications to establish such lab as well as to provide supporting services to laboratory users. The proposed lab will focus on training large models, and the supporting services will be aimed at enabling optimal use of the infrastructure as part of research and development processes in the field of artificial intelligence. These services may include, among other things:

• Assistance in the preparation, optimization, and accessibility of data,
• Planning of model training processes, and methods for analyzing results.

Funding Information

• The total grant for an approved application under this process will be up to NIS 160 million for the approved program period, with the grant rate being 55% or 66% of the approved budget, as detailed in the benefit program.

Criteria

• The establishment and/or access of an R&D lab in the State of Israel used to train large-volume models should feature the following: A volume of at least 1,000 accelerators (GPUs), with a minimum of a single accelerator performance being 60 teraFLOPS at a precision level of 32FP, 80GB internal memory, 2 Tbit/s bandwidth, and an effective clock speed of 1,500 MHz (or the total processing capacity of the computer with an equivalent accelerator composition in terms of processing power, processing speed, bandwidth and memory) in performing large model training throughout the entire program period.
• Examples of possible accelerators currently available on the market that meet the above requirements (the following examples do not constitute a closed or exhaustive list of possible accelerators that meet the requirements detailed above, rather are intended for the convenience of readers only):
  • AMD – MI300X OAM
  • INTEL – Intel Gaudi 3 AI
  • NVIDIA – SXMH100
• The lab’s infrastructure must include:
  • Required hardware,
  • Required software,
  • Required service, including communication solutions, storage infrastructure, cooling infrastructure, scheduler software and management and control tools for users who will contact the R&D laboratory to use it to train a new large model or adjust an existing large model, as a single service or as part of a package of services.

For more information, visit, Israel Innovation Authority.