support for 5090 gpu

[manojdev-ai]

extend the aistudio-cli to support https://www.corespan.ai/products/corespan-5090-inference-system

[manojdev-ai]

@claude

[claude]

Claude finished @manojdev-ai's task in 3m 34s —— View job • claude/issue-6-20260524-1223 • Create PR ➔

---

Supporting CoreSpan 5090 Inference System

• Gather context: Read codebase and understand GPU provisioning structure
• Strengthen explicit RTX 5090 / Blackwell detection in driversInstallation.sh
• Add NVLink bandwidth test subcommand for multi-GPU inference system validation
• Update README to document CoreSpan 5090 support and NVLink testing
• Commit and push changes

---

What was done

ai-studio-cli/assets/driversInstallation.sh
Extended the product-name fallback regex in detect_gpu() to explicitly match RTX 5090, RTX 5080, RTX 5070, GB202, and B300 — ensuring the Blackwell code-path (CUDA 12.8 + driver 570+) is reliably selected whenever a CoreSpan 5090 node is provisioned. (The GB chip-prefix already worked, but product-name matching was too broad.)

ai-studio-cli/cmd/nvlink.go (new file)
Adds an ai-studio-cli nvbandwidth nvlink subcommand that runs six peer-to-peer nvbandwidth testcases covering unidirectional and bidirectional GPU-to-GPU NVLink transfers. This is the primary health-check for multi-GPU interconnects on the CoreSpan 5090 Inference System:

ai-studio-cli nvbandwidth nvlink
ai-studio-cli nvbandwidth nvlink --json --buffer-size 1024 --samples 10

README.md

• Added a sentence at the top calling out CoreSpan 5090 / Blackwell support.
• Added a new NVLink Benchmarking section with usage examples.

---

·