Announced details
MoE expert matrix multiplication accelerated 3.68-fold
TPOT reduced by 31.3%, from 144.85 ms to 99.49 ms
Per-user decoding speed increased by 45.7%, from 6.9 tok/s to 10.05 tok/s
Average end-to-end latency (E2EL) reduced by 23.0%, from 25.0 seconds to 19.3 seconds
Fixstars and Macnica jointly support hardware evaluation, performance optimization, AI infrastructure development, deployment and operation for Tenstorrent AI products. Macnica provides TT-QuietBox, a remote evaluation environment, an AI model execution environment and technical support, while Fixstars is responsible for analyzing and optimizing AI models. Fixstars and Tenstorrent are engaged in technical collaboration. No capital or personnel relationship is disclosed.
Macnica, Inc. is a services and solutions company headquartered in Yokohama, Kanagawa Prefecture, with operations focused on semiconductors and cybersecurity. It operates from 92 locations across 26 countries and regions and has more than 50 years of history. The source does not provide Tenstorrent's location, representative, business activities or other details.
Macnica, Inc.: Yokohama, Kanagawa Prefecture. 92 locations across 26 countries and regions. TT-QuietBox installation location: not disclosed.
No information was provided on the impact on business performance.
The company does not address Contract Value / Offering price.
Fixstars Performance Engineering Series 06, “Practical Acceleration of gpt-oss-120b Inference on the Tenstorrent TT-QuietBox”
| Item | Details |
|---|---|
| Names of Products / Services Covered | Tenstorrent AI products, TT-QuietBox, gpt-oss-120b |
| Counterparty | Macnica, Inc., Tenstorrent |
| Counterparty overview | Macnica, Inc. is a services and solutions company headquartered in Yokohama, Kanagawa Prefecture, with operations focused on semiconductors and cybersecurity. It operates from 92 locations across 26 countries and regions and has more than 50 years of history. The source does not provide Tenstorrent's location, representative, business activities or other details. |
| Relationship with the Company | Fixstars and Macnica jointly support hardware evaluation, performance optimization, AI infrastructure development, deployment and operation for Tenstorrent AI products. Macnica provides TT-QuietBox, a remote evaluation environment, an AI model execution environment and technical support, while Fixstars is responsible for analyzing and optimizing AI models. Fixstars and Tenstorrent are engaged in technical collaboration. No capital or personnel relationship is disclosed. |
| Launch date | 2 October 2026 (announced as the start of joint support; it is not specified whether the activities began on that date) |
| Applicable area | Corporate use of generative AI, Healthcare, Manufacturing, Finance, Mobility |
| Location | Macnica, Inc.: Yokohama, Kanagawa Prefecture. 92 locations across 26 countries and regions. TT-QuietBox installation location: not disclosed. |
| Location scale | TT-QuietBox: 8 Wormhole chips and 96 GB of total GDDR6 memory. Macnica: 92 locations across 26 countries and regions. |
| Quantitative Effects | MoE expert matrix multiplication accelerated 3.68-fold, TPOT reduced by 31.3%, from 144.85 ms to 99.49 ms, Per-user decoding speed increased by 45.7%, from 6.9 tok/s to 10.05 tok/s, Average end-to-end latency (E2EL) reduced by 23.0%, from 25.0 seconds to 19.3 seconds |
| Impact on Business Performance | No information was provided on the impact on business performance. |
| Source | Fixstars Performance Engineering Series 06, “Practical Acceleration of gpt-oss-120b Inference on the Tenstorrent TT-QuietBox” |
Items this document does not state (7)
These are items the company does not state in this document. We do not leave them blank, nor fill them in from other sources or by inference.
The reasons the company gives
The company's own explanation, not something inferred from the numbers.To combine Macnica's support for AI infrastructure development and product evaluation with Fixstars' performance engineering capabilities, enabling customers' AI models to be evaluated and analyzed on AI hardware and their software implementations to be improved.