Extend the JBenchmark system to support running model performance benchmarks on multiple GPU architectures, including:

• AMD (e.g., MI250, MI300 – new)

• IBM Spyre (new)

The goal is to ensure that the benchmarking tool can:

• Detect and initialize GPU-specific configurations

• Launch benchmark workloads on AMD and IBM hardware with the same coverage and reliability as NVIDIA

• Capture hardware-specific metadata (e.g., GPU type, memory, driver version)

• Normalize results to allow comparison across vendors

Acceptance Criteria:

• Benchmark workloads successfully run on AMD and IBM GPU nodes

• GPU-specific init logic integrated into infra scripts (e.g., Docker/K8s setup)

• Results include vendor and architecture metadata

• Dashboard and GuideLLM display support for non-NVIDIA benchmarks

• Documentation updated for enabling benchmarking on AMD and IBM clusters