Decision Workspace
sgemm-bi vs burn-cuda vs torsh-backend
Side-by-side comparison of Rust crates
49
sgemm-bi
experimentalv0.1.1
Deterministic, batch-invariant CUDA GEMM engine with a full training triad (forward, dW, dX) in f32 / bf16 / f16, plus an opt-in tensor-core tier that is faster than cuBLAS PEDANTIC. Bit-identical results across runs; fixed reduction order; no atomics; no cuBLAS dependency.
65
burn-cuda
growingv0.21.0
CUDA backend for the Burn framework
62
torsh-backend
experimentalv0.1.3
Backend abstraction layer for ToRSh
Core Metrics
| sgemm-bi | burn-cuda | torsh-backend | |
|---|---|---|---|
| Health Score | 49 | 65 | 62 |
| Total Downloads | 34 | 747.1K | 1.8K |
| 30d Downloads | 0 | 60.2K | 0 |
| Dependents | 0 | 274 | 32 |
| Releases | 2 | 23 | 8 |
| Last Updated | 31d ago | 67d ago | 13d ago |
| Age | 1m | 1y 10m | 9m |
Health Breakdown
sgemm-bi
Maintenance
12
Quality
14
Community
6
Popularity
2
Documentation
15
burn-cuda
Maintenance
16
Quality
17
Community
12
Popularity
7
Documentation
13
torsh-backend
Maintenance
21
Quality
12
Community
10
Popularity
4
Documentation
15
Technical Details
| sgemm-bi | burn-cuda | torsh-backend | |
|---|---|---|---|
| Version | 0.1.1 | 0.21.0 | 0.1.3 |
| Stable (≥1.0) | ✗ No | ✗ No | ✗ No |
| License | MIT OR Apache-2.0 | MIT OR Apache-2.0 | Apache-2.0 |
| Dependencies | 2 | 4 | 39 |
| Crate Size | 101KB | 30KB | 1.2MB |
| Features | 1 | 8 | 17 |
| Yanked % | 0.0% | 0.0% | 0.0% |
| Edition | 2024 | 2024 | 2021 |
| MSRV | 1.94 | — | 1.77 |
| Owners | 1 | 1 | 1 |
Links
Quick Verdict
- •burn-cuda leads with a health score of 65/100, but none of the options score above 80.
- •burn-cuda is depended on by 274 crates — strongest ecosystem trust.