Intel Begins Deprecation of Ponte Vecchio and Arctic Sound Data Center GPUs
Intel has initiated a gradual phase-out of its "Ponte Vecchio" Data Center GPU Max and "Arctic Sound" Data Center GPU Flex series. This transition was highlighted in the changelog for Intel XPU Manager, an open-source tool designed for monitoring and managing Intel data center GPUs. With the release of version 1.3.3, support for the Data Center GPU Flex and Max Series has been deprecated. Users of these GPUs are now encouraged to continue using version 1.2.42 to maintain full feature compatibility, as the latest update may not support all functionalities.
The move to deprecate these GPUs comes just two years after their initial launch, raising speculation that Intel is preparing for a shift toward its next-generation "Jaguar Shores" AI accelerators. This transition signals a significant change in Intel’s data center GPU strategy.
Ponte Vecchio: Ambitious Design Meets Market Challenges
Ponte Vecchio, built on the Xe-HPC microarchitecture and manufactured using Intel’s 10 process, is a highly complex chip. It measures approximately 1,280 mm² and contains around 100 billion transistors. The GPU features 16,384 shading units, 1,024 tensor cores, and a peak TDP of up to 600 W per card. The Max 1100, 1350, and 1550 models, offering between 48 GB and 128 GB of memory, were designed for high-performance computing (HPC) and AI workloads.
Despite its impressive specifications, Ponte Vecchio’s large scale led to high production costs, significant power requirements, and integration complexities. These factors limited its adoption in a market already dominated by established accelerators with mature software ecosystems.
Arctic Sound: Compact Design for Midrange Data Centers
The "Arctic Sound" Data Center GPU Flex series, based on the Generation 12.5 GPU microarchitecture and also produced on the Intel 10 node, is a more compact solution. With a die size of about 190 mm² and roughly 8 billion transistors, it features 8,192 shading units, 128 ROPs, and a TDP close to 500 W. The Arctic Sound 1T, 2T, and M models, each equipped with 16 GB of memory, targeted midrange data center applications and first appeared in 2021.
However, Arctic Sound struggled to gain significant traction in enterprise environments. The primary challenges included competition from established GPU vendors, the need for robust driver support, and the simultaneous development of both hardware and software ecosystems. These hurdles made it difficult for integrators and cloud providers to justify migration and support investments.
Deployment Successes and Strategic Shifts
Despite limited market adoption, Intel achieved a notable milestone by integrating Ponte Vecchio GPUs into the Aurora supercomputer. This deployment demonstrated the hardware’s capability to operate at scale and meet demanding performance requirements. However, frequent changes in Intel’s accelerator roadmap have made vendors hesitant to commit exclusively to Intel’s solutions.
Looking ahead, Intel is focusing on the upcoming Jaguar Shores platform, which promises advanced silicon, HBM4 memory, and a mature OneAPI software stack. This next-generation architecture is expected to address many of the challenges faced by previous GPU offerings and position Intel more competitively in the data center accelerator market.
