Unverified50% confidenceBenchmarkExact time
On dual Strix Halo, one developer measured that a USB4 40Gbps link only achieved about 9.2Gbps; running DeepSeek across two machines gives about 27 tokens/s written and 700-750 tokens/s prefill.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/31/2026
First Seen
Valid until: 1/29/2027
Sources
Related Entities
Related Claims
UnverifiedDeveloper 0x0's build: one RTX 3090 paired with 256GB of DDR4 memory and a 200GB SSD runs DeepSeek v4.1 Flash at 26 tokens/s for a single user and 41 tokens/s with four concurrent users, with prefill of 600-800 tokens/s, support for a 262K context, and an estimated total cost of about $6,000.74% similarUnverifiedIn tests with the 4-bit quantized version (fitting entirely into 32 GB of VRAM), Ollama reached 240 tokens/s while FreeToken reached 225 tokens/s, making Ollama actually faster.73% similarUnverifiedBased on public data for high power plus MLX or Ollama, the decoding speed of different models on M4 Max is roughly a ladder as follows: Nemotron 3 Nano about 160 tokens/s, Qwen 3.5 35B at 107 to 130 tokens/s, Gemma 4 26B about 75 tokens/s, DeepSeek V4 Flash about 27 tokens/s, and dense 27B about 22 tokens/s.72% similarUnverifiedThe bfloat16 baseline throughput is 18.8K tokens/s/GPU, while the fastest custom path reaches 37.9K tokens/s/GPU, nearly doubling the throughput.72% similarUnverifiedJetBrains self-reported data: Mellum2.1 reaches 7,969 tokens/s on a single H200, while Qwen3.5-9B reaches 4,347 tokens/s.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1121826API
curl https://kongchang.com/api/v1/knowledge/claims/1121826MCP
get_claim(id=1121826)