dp4a
Here are 10 public repositories matching this topic...
DP4A FlashAttention-2 and GEMM for Volta GPUs. 46 TOP/s INT8 on CMP 100-210 where tensor cores are firmware-disabled. The .superl8 format loads weights at memory speed.
-
Updated
Sep 22, 2026 - Python
NVFP4 GEMM, GEMV, linear layers and quantization for NVIDIA GPUs without native FP4 (sm_50–sm_90): every NVIDIA NVFP4 variant (ModelOpt, Transformer Engine 1x16 / 16x16, 4over6, swizzled scales) bit for bit, exact block sums, tensor cores on sm_75–sm_89, C API and PyTorch bindings
-
Updated
Nov 16, 2026 - Cuda
llama.cpp CUDA DP4A/MMQ tuning for NVIDIA Pascal (SM 6.1), Tesla P40 and GTX 10. / Патчи llama.cpp для Pascal и Tesla P40.
-
Updated
Nov 18, 2026 - Shell
Everything I made run fast on a 2016 GPU with 8 GB: llama.cpp kernels, CUDA graphs, speculative decoding, ternary models, LoRA training, video diffusion on a GTX 1080 (sm_61). Measured, with sources.
-
Updated
Nov 3, 2026 - Python
Unlock NVIDIA CMP 50HX (TU102): ~32x FP32/INT8 compute + PCIe Gen2 retrain behind PLX switches. Patched 610.43.03 driver, verified on real hardware.
-
Updated
Nov 17, 2026 - C
A fused descriptor matcher for COLMAP: 1.28x faster SIFT matching with byte-identical output and 55% less matcher VRAM, plus an accounting of why that does not reach end-to-end runtime. Patch overlay, no vendored COLMAP source.
-
Updated
Aug 20, 2026 - TeX
Secure a proof-of-useful-work GPU blockchain where miners run auditable int8 matrix products for AI inference, turning hashing into useful compute.
-
Updated
Nov 10, 2026
Pascal (GTX 10xx / P104-100) dp4a GPU path for the official Vearl testnet miner, rc23 overlay, no dev fee
-
Updated
Oct 23, 2026 - Cuda
Add this topic to your repo
To associate your repository with the dp4a topic, visit your repo's landing page and select "manage topics."