Tracked topic
…It supports FP8 and NVFP4 quantization , efficient video sampling, and NVIDIA‑optimized kernels to deliver predictable, low‑latency inference. Combined with convolutional 3D‑based temporal‑spatial processing, these optimizations enable sustained multimodal…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
Here's a brand new Blackwell-native, prefill-optimized 4-bit quant that runs 50% faster on compatible hardware than a Q4 quant of the same memory footprint. And it runs 4-7% faster than other NVFP4 quants as benchmarked …
These are single stream numbers Following on from my previous post about v100s (here) and inspired by this comment (here) I decided to work on kernels that allow for an extremely fast path for Nvfp4 weights on sm70 and a…
Four Tesla V100s from 2017 matched my RTX 5090 on single-request Qwen 3.8 decode. Repo: https://github.com/dnv2003/v100-skinny https://i.redd.it/5ws2ak3uqckh1.gif The 5090 was not being held back. It ran NInfer, a specia…
Safetensors, llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved: https://huggingface.co/llmfan46/Qwen3.5-35B-A3B-uncensored-heretic-v2-Native-MTP-Preserved GGUFs, llmfan46/Qwen3.5-35B-A3B-uncensored-here…
Poolside have updated the FP8 and NVFP4 checkpoints for Laguna S 2.1, increasing the default context size to 1 million, and updating the configs. Here's hoping they fixed the looping issue, this model has been great in m…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
…a full-stack architecture built for speed, efficiency and scale: The Blackwell architecture features include: NVFP4 low-precision format for efficiency without loss of accuracy Fifth-generation NVIDIA NVLink that connects 72…
…NVIDIA NVFP4 further improves efficiency with hardware-friendly 4-bit computation. Together, these approaches enable faster, cost-effective inference for large models and resource-constrained systems. Support varies across Jetson platforms—refer…
…Dynamo, NVFP4, optimized CUDA kernels, advanced parallelization techniques, and beyond. Build with NVIDIA GPU-accelerated endpoints Developers can start building with DeepSeek V4 through NVIDIA GPU-accelerated endpoints on build.nvidia.com…