TensorSharp Beats llama.cpp in DeepSeek V4 Flash Inference Benchmark
fuzhongkai · reddit · 2026-08-01
Open-source inference engine TensorSharp has added support for the DeepSeek-V4-Flash-0731 model, outperforming llama.cpp in multi-GPU benchmarks.
Tested on 4x Nvidia A40 GPUs (CUDA 12.8) using unsloth's Q8KXL quantized version:
- Prefill speed: TensorSharp's CUDA backend hit 836 tok/s, beating llama.cpp's 558 tok/s.
- Decode speed: TensorSharp also slightly outperformed llama.cpp on both short and 16K context sequences.
The engine supports CUDA, Vulkan, Metal, OpenAI-compatible APIs, continuous batching, and multi-GPU/node inference.
Related event: TensorSharp Engine Doubles DeepSeek Inference Speed(4 posts)→
More from Infra
- NVIDIA quietly delists the entry-level RTX 5050 from its marketplace, following the RTX 5090 — chemist_slime · 2026-10-16
- Grok claims 816B MoE model runs locally on dual M5 Ultra Mac Studios via MLX — grok · 2026-10-16
- Same 1,000 Cloudflare Durable Objects: $4,050/month cut to $22 with one trick — threepointone · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16
- DuoDiscover agents auto-build fast eval proxies, scoring 2.68x TPU kernel speedup — berkeley_ai · 2026-10-16