返回
原文
The Decoder：AI News（RSS）
52AI 编辑部评分，满分 100

AMD 收购 Taalas，将 AI 模型直接烧录进芯片

2026-08-08 02:01· 24分钟前· Matthias Bastian
AI 导读

AMD 收购加拿大 AI 芯片初创公司 Taalas，后者将模型架构与训练参数直接嵌入芯片，推理速度极快但每颗芯片锁定单一模型。其演示芯片运行 Llama 3.1-8B 时，每用户每秒处理超 16,000 tokens，远超竞品。AMD 计划将该技术纳入加速器路线图，与 Instinct GPU 作为系统级方案共同提供，交易尚待监管批准。

原文 · 未翻译

AMD is buying Canadian AI startup Taalas, which builds specialized inference chips. Founded in Toronto in 2023, Taalas came out of stealth in February with an unusual approach: the company embeds a model's architecture and trained parameters directly into the chip. That makes inference extremely fast but locks each chip to a single model. A demo chip hit over 16,000 tokens per second per user running Llama 3.1-8B, many times faster than competing hardware. Google is reportedly working on a similar chip for Gemini.

The Taalas demo chip with Llama hard-coded into it blows past every other inference chip in raw speed, but it's locked to that one model. | Image: Taalas

AMD plans to fold the technology into its accelerator roadmap and offer it alongside Instinct GPUs as a system-level solution. Vamsi Boppana, SVP of AMD's AI division, said the deal strengthens the company's AI portfolio. Taalas co-founder Ljubisa Bajic said AMD provides the scale and reach the startup needs. The acquisition is subject to standard regulatory approvals.

AMD

Taalas

端侧行业动态

来源：The Decoder：AI News（RSS） · the-decoder.com

导出 Markdown
返回
52打开原文
导出 Markdown

AMD 收购 Taalas，将 AI 模型直接烧录进芯片

The Decoder：AI News（RSS）·2026-08-08 02:01·24分钟前·Matthias Bastian
AI 导读

AMD 收购加拿大 AI 芯片初创公司 Taalas，后者将模型架构与训练参数直接嵌入芯片，推理速度极快但每颗芯片锁定单一模型。其演示芯片运行 Llama 3.1-8B 时，每用户每秒处理超 16,000 tokens，远超竞品。AMD 计划将该技术纳入加速器路线图，与 Instinct GPU 作为系统级方案共同提供，交易尚待监管批准。

原文 · 保持原样，未翻译

AMD is buying Canadian AI startup Taalas, which builds specialized inference chips. Founded in Toronto in 2023, Taalas came out of stealth in February with an unusual approach: the company embeds a model's architecture and trained parameters directly into the chip. That makes inference extremely fast but locks each chip to a single model. A demo chip hit over 16,000 tokens per second per user running Llama 3.1-8B, many times faster than competing hardware. Google is reportedly working on a similar chip for Gemini.

The Taalas demo chip with Llama hard-coded into it blows past every other inference chip in raw speed, but it's locked to that one model. | Image: Taalas

AMD plans to fold the technology into its accelerator roadmap and offer it alongside Instinct GPUs as a system-level solution. Vamsi Boppana, SVP of AMD's AI division, said the deal strengthens the company's AI portfolio. Taalas co-founder Ljubisa Bajic said AMD provides the scale and reach the startup needs. The acquisition is subject to standard regulatory approvals.

AMD

Taalas

来源：The Decoder：AI News（RSS）· the-decoder.com

端侧行业动态