fix: add theoretical TFlops for H200 GPU

[roclark]

What does this PR do ?

Added the theoretical TFlops for H200 GPUs to measure the process efficiency.

Issues

N/A

Usage

N/A

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• Confirmed the format of the GPU output on a cluster with H200s:

$ python3 -c 'import torch; print(torch.cuda.get_device_name())'
NVIDIA H200

• Confirmed theoretical TFlops in the H200 data sheet

Summary by CodeRabbit

• New Features
  • Extended GPU performance tracking support to include NVIDIA H200 with bfloat16 and float32 precision types.

[coderabbitai]

📝 Walkthrough

Walkthrough

The pull request adds theoretical TFLOPS benchmark entries for NVIDIA H200 GPUs in bfloat16 and float32 data types to the THEORETICAL_TFLOPS lookup table, extending device-dtype coverage without modifying control flow or behavioral logic.

Changes

Cohort / File(s)	Summary
H200 TFLOPS Benchmark Additions nemo_rl/utils/flops_tracker.py	Adds two THEORETICAL_TFLOPS entries for H200: bfloat16 (1979/2 TFLOPS) and float32 (989/2 TFLOPS with TF32 conditional, else 67.0), mirroring existing H100 structure.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Possibly related PRs

• feat: Update Theoretical TFLOPS #1236: Adds B200/B300/GB200/GB300 TFLOPS entries to the same THEORETICAL_TFLOPS dictionary, extending benchmark coverage for additional GPU architectures.

Suggested reviewers

• guyueh1
• terrykong

Pre-merge checks and finishing touches

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Test Results For Major Changes	✅ Passed	The PR adds two new entries to the THEORETICAL_TFLOPS dictionary for the NVIDIA H200 GPU with specific bfloat16 and float32 values, mirroring the H100 entries. This is a minor change—purely a data addition to a lookup table with no new logic, behavioral changes, or code modifications that could affect numerics, convergence, or performance. The existing test file test_flops_counter.py validates FLOPS calculations for various models and configurations, and while it doesn't specifically test H200 values, the PR description explicitly states that pre-checks were completed including running unit and functional tests locally with no issues reported. The H200 TFLOPS values match official NVIDIA specifications and align with the established H100 values in the same table, making them straightforward reference data additions.
Title check	✅ Passed	The title accurately describes the main change: adding theoretical TFLOPS values for H200 GPU to the benchmark table in the flops_tracker.py file.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[terrykong]

@guyueh1 to review

[terrykong]

@guyueh1 bump

[terrykong]

closing in favor of #1543 which has some tests