- Semiconductor BusinessHOME
- Products and Services of Macnica,Inc.
-
technical information
-
Events and Seminars
- Handling Manufacturer
- Support
- Inquiry
- Click here to purchase products
- Semiconductor business e-mail magazine registration
Overview
Tenstorrent's AI Server is a data center scale-out AI platform featuring high-density deployments of the latest generation Blackhole® Tensix processors. Its high-performance design, optimized for generative AI and LLM inference, combined with flexible node-to-node communication via 800G Ethernet, allows for seamless scalability from a single chassis to rack-scale. A fully open-source software stack delivers the performance and scalability required for full-scale operation.
Galaxy™ Blackhole
Tensorrent's Galaxy™ Blackhole is a native scale-out AI server with 32 Blackhole® ASICs densely packed into a 6U chassis. It delivers 23 PFLOPS (FP8)/1TB GDDR6 in a single chassis, enabling high-speed processing of cutting-edge workloads such as generated AI video and LLM inference. Its Ethernet-based integrated fabric allows for flexible scaling from a 4-node supercluster to thousands of chips.
spec
| Item | specification |
|---|---|
| Product form | Scale-out AI server |
| Casing size | 6U rack mount |
| Built-in AI processor | Blackhole® ASIC × 32 units |
| Total number of Tensix cores | 4,480 cores (140 cores x 32 ASICs) |
| AI performance (FP8) | Up to 23 PFLOPS |
| AI performance (BLOCKFP8) | Approximately 21 PFLOPS |
| Total memory capacity | 1 TB GDDR6（32GB × 32 ASIC） |
| Memory bandwidth (total) | 16 TB/s |
| Total SRAM capacity | 1,600 MB |
| Host Configuration | Dual AMD EPYC™ 4th Generation CPUs |
| system memory | 1 TB DDR5 |
| storage | NVMe SSD (capacity varies by SKU) |
| Network (external) | 800GbE x 32 ports (QSFP-DD) |
| Network (internal fabric) | Ethernet-based integrated fabric |
| cooling method | Air cooling/liquid cooling options available |
| power supply | Redundant power supply (details depend on SKU) |
| OS / SW | Ubuntu 22.04 LTS / TT-Metalium, TT-NN, TT-Forge (Open Source) |
| GA (general availability) | done |
| Intended use | AI-generated video / LLM inference (prefill & decode) / Large-scale AI inference platform |
Inquiry / Quotation
If you have any questions about this product or would like a quote, please contact us below.
Tenstorrent Manufacturer Information Top Page
To return to the Tenstorrent manufacturer information top page, please click below.