A handbook for your

TT-QuietBox® 2

Hover a path to see what you'll do. Click to dive in.

⚡ Ships with Qwen3-32B pre-loaded — your first token is minutes away, no download.

QB2 Guide Paths — choose your adventure Install Stack tt-smi Run Models First Model Explore Run & build Tinker Customize
Explore
Explore Tenstorrent hardware & software for the first time
7 chapters · 51 min
Enter →
Run & build
Run and build on top of models
6 chapters · 56 min
Enter →
Tinker
Tinker with models and the software & hardware stack
6 chapters · 59 min
Enter →
Customize
Customize your workstation and make it truly your own
5 chapters · 36 min
Enter →
TT-QuietBox 2 — a small-form-factor workstation with a glass side panel showing the liquid-cooled Blackhole cards and teal accent graphics

TT-QuietBox® 2

Blackhole® · 4 chips · liquid-cooled · $9,999

Product page ↗
AI Accelerator
Chips4× Blackhole® (2× p300c cards)
Tensix cores480 total (120 per chip)
AI clock1.35 GHz
SRAM720 MB total (180 MB/chip)
DRAM128 GB GDDR6 total (32 GB/chip)
Bandwidth2,048 GB/s total (1,024 GB/s per card)
Chip interconnect800 Gbps between any two chips (internal Samtec ARP6 cable between cards)
Card powerUp to 1,100 W (550 W per card)
Host System
CPUAMD Ryzen 7 9700X (8c, 3.8 GHz)
System RAM256 GB DDR5-5600
Storage4 TB NVMe (PCIe Gen4 x4)
MotherboardASRock B850M-C
OSUbuntu 24.04.3 LTS
PSU1,600 W
Maximum draw1,300 W
Idle power320 W
Connectivity
Display1× HDMI
USB8× (2× Gen2, 2× Gen1, 4× USB 2.0)
LANGigabit Ethernet
WiFiWi-Fi 6 (802.11ax) dual-band
Bluetooth5.3
Physical
Dimensions9.1 × 17.8 × 15.6 in
Weight20 kg / 44 lbs
CoolingLiquid + forced air (38 dBA at max load)
Operating temp10–35°C (50–95°F)

Source: docs.tenstorrent.com — TT-QuietBox 2 Specifications

What's Running on QB2

Live from the Tenstorrent compatibility matrix — models confirmed or in testing on Quietbox 2 / P150 / P300c hardware.

Updated 2026-11-01 05:00 UTC

Supported Text Generation
DeepSeek-R1-Distill-Llama-70B
70B
R1 reasoning traces distilled into a Llama 70B backbone for math and code. DeepSeek, 70B.
tt-inference-server
Supported
gemma 4 12b it
tt-inference-server
Supported
gemma 4 26b a4b it
tt-inference-server
Supported Text Generation
gemma-4-31b-it
31B
Instruction-tuned Gemma 4 for multilingual reasoning, coding, and long context. Google,...
tt-inference-server
Supported
granite 4 2 30b
tt-inference-server
Supported Text Generation
Llama-3.1-70B
70B
128K context, multilingual, tool-use ready, fully open weights. Meta, 70B.
tt-inference-server
Supported Text Generation
Llama-3.1-70B-Instruct
70B
Complex reasoning, coding, and agentic pipelines — no license restrictions. Meta, 70B.
tt-inference-server
Supported Text Generation
Llama-3.3-70B-Instruct
70B
Stronger structured tasks, tool use, and reasoning than prior Llama generations. Meta,...
tt-inference-server
Supported
qwen3 8 27b
tt-inference-server
Supported Text Generation
Qwen3-32B
32B
Toggleable chain-of-thought for on-demand deep reasoning. Alibaba, 32B.
tt-inference-server
Supported Text-to-Video
Wan2.2
14B
Causal video transformer for text-to-video with strong motion coherence. Alibaba, 14B.
tt-inference-server
Supported Text-to-Image
Z-Image-Turbo
6B
Few-step distilled text-to-image generation. Tongyi-MAI.
tt-inference-server
Supported Text-to-Image
FLUX.1 [dev]
12B
Flow-matching transformer for photorealistic text-to-image with strong prompt adherence....
tt-inference-server
Supported Text-to-Image
FLUX.1 [schnell]
12B
FLUX distilled to 4 steps — full quality, fraction of the compute. Black Forest Labs, 12B.
tt-inference-server
Supported Text Generation
Llama-3.1-8B
8B
128K context and multilingual base with a wide fine-tuning ecosystem. Meta, 8B.
tt-inference-server
Supported Text Generation
Llama-3.1-8B-Instruct
8B
Multilingual instruction following, tool use, and function calling. Meta, 8B.
tt-inference-server
Supported Speech-to-Text
distil-large-v3
756M
Distilled Whisper large-v3 — near-parity transcription at roughly six times the speed....
tt-inference-server
Supported Speech-to-Text
whisper-large-v3
1.5B
99 languages, 680K hours of training audio — built for robust speech recognition. OpenAI,...
tt-inference-server
Supported Object Detection
YOLOX-Nano
0.9M
Anchor-free detector tuned for real-time inference at minimal parameter cost. Megvii,...
tt-inference-server
Experimental Feature Extraction
BGE-M3
568M
Multi-lingual, multi-granularity retrieval across dense, sparse, and multi-vector modes....
tt-inference-server
Experimental Text Generation
diffusiongemma-26B-A4B-it
26B
Instruction-tuned text-diffusion Gemma 4 MoE that writes whole 256-token blocks per step,...
tt-inference-server
Experimental Text Generation
gpt-oss-120b
120B
Open GPT-style model for deep reasoning in self-hosted deployments. GPT-OSS, 120B.
tt-inference-server
Experimental
k2 horizon 7b
tt-inference-server
Experimental
kolibri 1 bf16
tt-inference-server
Experimental
muse glimmer 30b
tt-inference-server
Experimental Embedding
Qwen3-Embedding-0.6B
0.6B
Compact multilingual embeddings for retrieval on constrained hardware. Alibaba, 0.6B.
tt-inference-server
Experimental Embedding
Qwen3-Embedding-4B
4B
Multilingual embeddings for retrieval and semantic similarity. Alibaba, 4B.
tt-inference-server
Experimental Text Generation
Qwen3.6-27B
27B
Mid-scale Qwen3.6 for general reasoning, coding, and multilingual generation. Alibaba,...
tt-inference-server
Experimental Text-to-Speech
SpeechT5 (TTS task)
307M
Unified encoder-decoder for natural text-to-speech synthesis. Microsoft, 307M.
tt-inference-server