Repository navigation
RunPod one-command GPU benchmark script - #17
Merged
Merged
Conversation
Colab tiers gate the 24 GB GPUs (L4/A100) behind Pro+; this script is the deterministic ~$0.50 alternative: RunPod (or any Ubuntu CUDA box, root) -> paste one curl|bash command. Self-contained: toolchain, repo, cuda prover build, libcudart workaround, VRAM-logged benchmarks of both formats, and a copy-paste summary block.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Colab gates the 24 GB GPUs (L4/A100) behind Pro+; this is the deterministic ~$0.50 alternative.
RunPod path: account → add $5 → create a Pod with RTX 4090 (24 GB) on a PyTorch 2.x / CUDA 12.x template → open the web terminal → paste:
(after merge: the
mainURL). Self-contained: toolchain pin, repo clone, cuda prover build, the documented libcudart workaround, VRAM-logged benchmarks of both formats (V2 native ~1.8M cycles; legacy meta-emulated ~169M cycles that OOM-killed a 32 GB CPU host), and a copy-paste summary block at the end. ~30–45 min total, terminate the pod after.Skips CI (docs/scripts-only change per #16's paths filter).