Hello,
I am working on benchmarking a mixed-precision Conjugate Gradient (CG) solver with a custom preconditioner on Habana Gaudi / Gaudi2 hardware.
I am currently using AWS, where the Habana dl1 instance type is listed as the primary way to access Gaudi accelerators. However, the AWS platform now directs users to Habana/Intel for availability and support, and I am unable to start or request dl1 instances from my AWS environment.
Could you please clarify:
• Whether Intel Developer Cloud (IDC) or another Intel-managed platform currently provides external access to Gaudi or Gaudi2 instances for benchmarking and development.
• If so, what the process is to request access for running custom non-DL workloads (numerical solvers written in Python and Rust, without a deep-learning framework dependency).
My goal is to evaluate numerical stability and performance scaling of a mixed-precision iterative solver compared with CPU and GPU baselines.
Any updated information, eligibility requirements, or partner channels for accessing Habana hardware would be greatly appreciated.
Thank you for your time and assistance.
