Developer Tools●●●Banger
mtp-rs – pure-Rust MTP library, up to 4x faster than libmtp
Pure-Rust MTP library beats libmtp 4x faster with no C dependencies.
Solve My ProblemBig Brain
vdavid
312mo ago
Back to browse
GitHub Repository
49 starsRust
Nabla – Pure Rust GPU math engine, 7.5× faster matmul than PyTorch
by fumishiki·Mar 1, 2026·1 point·1 comment
AI Analysis
●●SolidWizardryBig Brain
Pure Rust autodiff + GPU math avoids C++ FFI hell, but matmul claim needs apples-to-apples benchmarks.
Strengths
- •Pure Rust implementation eliminates C++ FFI complexity and dependency bloat, enabling true cross-platform CUDA/Vulkan/AMD kernels from one codebase
- •Loss-backward() API and kernel fusion (fuse!()) with einsum! macros reduce boilerplate vs hand-rolled CUDA
- •Benchmark on GH200 is credible hardware, though TF32 vs FP32 comparison requires scrutiny
Weaknesses
- •7.5× speedup is precision-qualified (TF32 vs PyTorch default); apples-to-apples FP32 shows only 1.6× advantage
- •Minimal ecosystem maturity: only 7 GitHub stars, unclear adoption, and no real-world training examples beyond micro benchmarks
Category
Target Audience
Rust developers building ML systems, inference pipelines, and numerical computing
Similar To
PyTorch · tch-rs · Candle
Similar Projects
Developer Tools●●Solid
NeuralScript – A pure-Rust AOT compiler
Compile-time tensor shape checking beats PyTorch's runtime dimension errors.
Bold BetBig Brain
AkaiNa
501mo ago
Data●●●●Gem
Cuckoo-GPU – A 350x faster Bloom filter alternative for GPUs
350x faster GPU Bloom filter with academic paper backing the performance claims.
WizardryBig BrainDark Horse
tdortman
112mo ago
AI/ML●●●Banger
TRELLIS.2 image-to-3D running on Mac Silicon – no Nvidia GPU needed
Runs 4B-parameter image-to-3D on Mac without CUDA—Microsoft's original required NVIDIA only.
WizardryNiche GemZero to One
shivampkumar
202401mo ago
AI/ML●●●Banger
I built a 2nd-order PyTorch optimizer for LLMs that runs on 16GB GPUs
Runs Shampoo-quality second-order optimization on a 16GB T4 where others OOM immediately.
WizardryBig Brain
dnosoz
241mo ago
Developer Tools●●●Banger
Profine – Profile and rewrite your PyTorch training loop on real GPUs
Automates the painful torch.compile and mixed-precision tuning loop with measured 3x speedups.
Big BrainSolve My Problem
aisinghal
401mo ago