Research map / Large language models / AI and society
The Grand Illusion: The Myth of Software Portability and Implications for ML Progress.
Fraser Mince, Dzung Dinh, Jonas Kgomo, Neil Thompson, Sara Hooker
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| move PyTorch functions from GPU to TPU | PyTorch · stratified function sample | pass rate: 92% -> 57% | Hurt |
| move TensorFlow functions from GPU to TPU | TensorFlow · stratified function sample | pass rate: 78% -> 71% | Hurt |
| move JAX functions from GPU to TPU | JAX · stratified function sample | pass rate: 98% -> 97% | No clear effect |
| move PyTorch functions from GPU to TPU | PyTorch · stratified function sample | partial failure rate: 3% -> 27% | Hurt |
| move functions from GPU to TPU | PyTorch · latency | TPU/GPU ratio: 6039.554 | Hurt |
| move tf.linalg.svd from GPU to TPU | TensorFlow · latency | TPU/GPU ratio: 0.931 -> 112.843 | Hurt |
| move TensorFlow functions to TPU | TensorFlow · latency | share of slower functions: 96% | Hurt |
| move PyTorch functions to TPU | PyTorch · latency | share of slower functions: 100% | Hurt |
| move JAX functions to TPU | JAX · latency | share of slower functions: 8% | Mixed |
| restrict to top-20 most used functions | PyTorch · GPU failure rate | failure rate: 10% -> 0% | Improved |
| restrict to top-20 most used functions | TensorFlow · GPU failure rate | failure rate: 22% -> 33% | Hurt |
| restrict to top-20 most used functions | PyTorch · TPU failure rate | failure rate: 44% -> 46% | No clear effect |
Closest papers by meaning
- Position: Application-Driven Innovation in Machine Learning (ICML 2024 · similarity 0.86)
- On The Fairness Impacts of Hardware Selection in Machine Learning (ICML 2024 · similarity 0.85)
- What Fits (Into Few Tokens) Doesn't Overfit: Compression and Generalization in ML Research Agents (NeurIPS 2026 · similarity 0.83)
- MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering (ICLR 2025 · similarity 0.83)
- MLR-Bench: Evaluating AI Agents on Open-Ended Machine Learning Research (NeurIPS 2025 · similarity 0.83)
- A hierarchical decomposition for explaining ML performance discrepancies (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
