DeepSeek-V4-Flash tested: hits 2,100 tok/s aggregate on 4x RTX PRO 6000
BanghuaZ · x · 2026-08-01
Shared benchmark results for the rumored DeepSeek-V4-Flash-0731 model running on 4x RTX PRO 6000 (SM120).
- Stack: Deployed via SGLang and DSpark.
- Throughput: 200–225 tok/s single-stream; 2,100 tok/s aggregate at 32 concurrent requests.
- Accuracy & Fix: Achieves 0.99 on GSM8K. Required a 3-line FlashInfer patch (topk=192) to run on the new architecture.
The author notes that all figures are measured and fully reproducible.
More from Infra
- 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
- Google Raises 2026 Capital Spending to Up to $205 Billion — emmanuelvivier · 2026-10-16