Google and AMD are reportedly working together on a new Tensor Processing Unit (TPU) that could be built with on-package cores. This custom chip is expected to handle heavy artificial intelligence workloads in future hardware, though the specific devices remain unclear. The new silicon would join Google’s existing lineup, which includes the Tensor chips found in Pixel devices and the data center units used by cloud providers.
Why Google Needs a Partner
Google has previously designed and manufactured its own TPUs without outside help. The decision to partner with AMD raises questions about why the company would seek assistance for technology it has already mastered internally. According to reports, Google likely needs AMD’s expertise in areas where Google’s own capabilities may fall short, specifically regarding CPU intellectual property and programming logic.
It is suspected that Google wants to leverage AMD’s strengths to improve aspects of the TPU that are difficult to perfect alone. The collaboration might focus on on-package cores, which can improve performance metrics like power efficiency and general-purpose tensor compute. This approach could be a shift in strategy for Google, which has traditionally handled its own silicon design.
Related: ASUS Googlebook Prices Expected to Be High
For AMD, the potential partnership offers a new revenue stream from hyperscaler companies that are currently investing heavily in AI infrastructure. While neither company has confirmed the rumors, the report from Tom’s Hardware cites details shared on X by an individual named Sean. It notes that AMD could benefit by supplying specialized technology that complements Google’s existing roadmap, particularly as the company also develops the Tensor G7 for Pixel devices.
What This Means for the Market
Historically, Google has maintained tight control over its custom silicon to ensure optimization for its specific software stack and services. This rumored move suggests a pragmatic pivot toward shared manufacturing or design resources, a strategy often seen when scaling production demands exceed in-house capacity. By integrating AMD’s CPU IP or on-package cores, Google could potentially reduce the complexity of the design process while benefiting from AMD’s established manufacturing processes.
The integration of on-package cores represents a shift from traditional monolithic chip designs toward more heterogeneous architectures. This allows for specialized compute units to sit directly next to the main processing logic, which can reduce latency and improve energy efficiency. If the collaboration moves forward, it could signal a broader trend where major cloud providers rely more heavily on specialized partners for their most critical AI accelerators.
Leave a Reply