Capacity-aware single-GPU SGLang benchmarks with MTP A/B, core/extended context matrices, raw telemetry, model/KV memory capture, and generated Markdown/PDF reports.
-
Updated
Aug 1, 2026 - Python
#
sm80
Here are 6 public repositories matching this topic...
Evidence-led vLLM tuning and qualification notes for NVIDIA CMP 170HX (sm80)
-
Updated
Aug 10, 2026
GLM-5.3-Flash (320B MoE) served with vLLM on 8x CMP 170HX (SM80, 64 GB, PCIe Gen2 x4): pipeline-parallel 8, NVFP4 weights, DFlash2 speculative decoding, 1M context. Patches, launch config and benchmarks.
moe pipeline-parallelism vllm llm-inference speculative-decoding nvfp4 glm-5 dflash cmp-170hx sm80 glm-5-3-flash
-
Updated
Aug 31, 2026 - Python
Strict-FP32 CUDA SGEMM with shape-dispatched SM80 cp.async kernels and reproducible A100 evidence
performance-engineering cpp cuda matrix-multiplication high-performance-computing blas cuda-kernels gpu-computing sgemm a100 fp32 sm80
-
Updated
Aug 21, 2026 - Python
Native dual-GPU long-context inference runtime for Qwen3.8-Flash-Next
-
Updated
Aug 31, 2026 - Cuda
Add this topic to your repo
To associate your repository with the sm80 topic, visit your repo's landing page and select "manage topics."