Unverified50% confidenceFactExact time
CrofAI声称用Vast租用的RTX Pro 6000运行Kimi K3，但即便采用Q2_K量化该模型也需约802GiB显存，而Vast上最大8卡RTX PRO 6000机器仅有765GiB显存
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/15/2026
First Seen
Valid until: 12/14/2026
Sources
Related Entities
Related Claims
Unverified6GB显存的RTX 3060可以流畅运行基于Flux的Krea2 Turbo大模型72% similarUnverified实测在RTX A6000（48GB显存）机器上，将10亿参数草稿模型挂载到Qwen3的3.4B目标模型进行测试70% similarUnverified将8个请求合并批处理时，单张RTX PRO 6000能够达到约220 output tok/s的聚合吞吐量，而两张RTX 3090的旧配置只有约19 tok/s70% similarUnverifiedUP主的测试环境为RTX 5070显卡（12GB显存）、Qwen3系列9B参数模型、64K上下文长度69% similarUnverified使用QLoRA在RTX 4090上微调Llama 7B模型，显存占用约5-8GB69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/922187API
curl https://kongchang.com/api/v1/knowledge/claims/922187MCP
get_claim(id=922187)