斯推出开源TT-QuietBoxTM 2人工智能工作站,可在2654 TFLOPS的速度下进行本地120B参数模型推断. Tenstorrent launches open-source TT-QuietBox™ 2 AI workstation, enabling local 120B-parameter model inference at 2,654 TFLOPS.

出租人已经启动了TT-QuietBoxTM 2号,这是一家基于RISC-V的AI工作站,价格为9 999美元,提供高达1 200亿参数的模型(包括Llama 3.1 70B和Qwen3-32B)的梯形级推理,支持当地图像和视频生成。 Tenstorrent has launched the TT-QuietBox™ 2, a RISC-V-based AI workstation priced at $9,999, offering teraflop-class inference for models up to 120 billion parameters, including Llama 3.1 70B and Qwen3-32B, with support for local image and video generation. 它具有四个黑洞ASIC,480个Tensix核心,128 GB GDDR6和256 GB DDR5内存,提供2,654 TFLOPS的BlockFP8精度,同时通过芯片上的SRAM避免DRAM瓶. It features four Blackhole ASICs, 480 Tensix cores, 128 GB GDDR6, and 256 GB DDR5 memory, delivering 2,654 TFLOPS at BlockFP8 precision while avoiding DRAM bottlenecks via on-chip SRAM. 该系统在从硬件到软件的完全开放源码堆叠上运行,支持PyTorch和TensorFlow等主要框架,并以标准120V电源静悄悄地运行。 The system runs on a fully open-source stack from hardware to software, supports major frameworks like PyTorch and TensorFlow, and operates quietly on standard 120V power. 预计在2026年第二季度全球推出,并将在GDC 2026上进行展示. Available globally in Q2 2026, it will be demonstrated at GDC 2026.