BioTriton

Core bioinformatics workloads (sequence alignment, k-mer indexing, quality control) are still bottlenecked on CPU tools like BioPython, samtools and Jellyfish, and the GPU tools that exist are CUDA-only, leaving AMD users on the CPU. BioTriton is a library of 20+ Triton kernels that delivers large speedups over CPU and GPU baselines while staying portable: one source compiles to both NVIDIA PTX and AMD ROCm without modification.

The key observation is that bioinformatics maps naturally onto Triton's block-structured model. Its data structures (2-bit DNA, 5-bit amino acids, scoring matrices resident in shared memory) and algorithms (wavefront-parallel dynamic programming) line up with how Triton expresses work, which enables domain-aware fusion and memory-hierarchy optimization without vendor lock-in.

1Write once, run on both vendors

BioTriton kernels compile through MLIR to PTX and AMDGPU with no source changes. Benchmarked on NVIDIA RTX PRO 6000 Blackwell, H100 and L4, and AMD MI300X, all 20+ kernels compile and pass correctness tests on both vendors. Existing GPU tools such as NVBIO and GASAL2 are CUDA-only.

2Benchmark results

Every kernel produces numerically identical outputs (alignment, scoring) or bitwise-equivalent results (quality control, translation). The comparisons against compiled C++ tools matter most: they show the gains hold against optimized native baselines, not just Python references.

3Why it matters

BioTriton shows Triton is ready for portable scientific GPU computing. Targeting MLIR rather than one vendor backend gives write-once, run-anywhere portability without giving up performance, which is what HPC centres with heterogeneous clusters need: DOE's Frontier and El Capitan run AMD, while most academic clusters run NVIDIA. Next steps include cross-architecture roofline analysis, multi-GPU whole-genome alignment, and BioKernelBench, a reproducible cross-vendor benchmark suite for bioinformatics kernels.

Bioinformatics data structures and algorithms map cleanly onto Triton's block model, so a single write-once kernel library can run identically across NVIDIA and AMD with no CUDA lock-in.

More research

Part of a broader line of work at Embedded LLM on portable, fused GPU kernels for biological and clinical foundation models, measured on both NVIDIA and AMD.

• NeurIPS 2026Change the Metric, Change the Winner: Auditing How We Decide Which Clinical Foundation Model to Deploy in Cancer Pathology
• MOML 2026From 13 Hours to 4.6: Profiling-Driven Kernel Fusion for TensorNet in Molecular Simulation
• COLM 2026Deploying Clinical Language and Vision-Language Models Where the Data Lives
• ICML 2026From 805 ms to 23 ms: Accelerating State-Space Models for Real-Time ICU Monitoring
• ICPP 2026Error-Bounded Fused Attention Compression for Long-Context Genomic Foundation Models Across Heterogeneous GPUs
• ISCA 2026When the LLM-Tuned Stack Misses: An Infrastructure View of Biological Foundation Model Inference Across NVIDIA and AMD
• RECOMB 2026Hardware-Portable Fused GPU Kernels for High-Throughput Biological Foundation Models
• ISC HPC 2026Portable GPU Kernel Acceleration for Biological Foundation Models & Algorithms using OpenAI Triton

All research →