Intel Begins Phasing Out Ponte Vecchio and Arctic Sound Data Center GPUs
Table of Contents
Intel is initiating a slow deprecation of its Ponte Vecchio Data Center GPU Max and Arctic Sound Data Center GPU Flex series, signaling a strategic shift in its approach to the competitive high-performance computing and artificial intelligence accelerator market. Owners of these GPUs are being advised to remain on older software versions for continued functionality.
According to a changelog in the Intel XPU Manager-a free and open-source tool for monitoring and managing Intel data center GPUs-version 1.3.3 officially deprecates support for both the Data Center GPU Flex Series and the Data Center GPU Max Series. Users are now recommended to stick with version 1.2.42 to maintain maximum feature support and avoid compatibility issues with the latest release.
The move, coming just two years after the launch of the Data Center GPU Flex/Max series, has fueled speculation that Intel is preparing for a transition to its next-generation “Jaguar Shores” AI accelerator platform.This suggests a recalibration of Intel’s strategy as it navigates a market dominated by established players like NVIDIA and AMD.
Ponte Vecchio: A Bold, But Challenging, design
The Ponte Vecchio GPU, built on the Xe-HPC microarchitecture using Intel’s 10nm process, represents a significant engineering feat. This massive chip measures approximately 1,280 mm and contains roughly 100 billion transistors.It boasts impressive specifications, including 16,384 shading units, 1,024 tensor cores, and a peak thermal design power (TDP) of up to 600W.
The Max 1100, 1350, and 1550 SKUs, offering between 48 and 128 GB of memory, initially appealed to specialized HPC and AI applications. However,its sheer scale presented considerable challenges. “Its large scale results in high costs, significant power consumption, and integration challenges,” one analyst noted.
Arctic Sound: A Compact Alternative with Limited Impact
In contrast, Intel’s Arctic Sound GPU, based on the Generation 12.5 GPU microarchitecture and also manufactured on the Intel 10nm node, is a more compact design. Measuring around 190 mm with approximately 8 billion transistors, 8,192 shading units, and 128 ROPs, it has a TDP of around 500W. The Arctic Sound 1T, 2T, and M SKUs, featuring 16 GB of memory, were targeted at midrange data center tasks, debuting in 2021.
Despite its smaller footprint, Arctic sound failed to gain significant market share.The GPUs saw limited adoption in enterprise settings, largely due to the competitive landscape.
Ecosystem Challenges and a Shifting Strategy
Both Ponte Vecchio and Arctic Sound entered a market already well-served by established accelerators and mature software ecosystems. For many potential customers, factors like driver stability, comprehensive tooling, and a robust ecosystem were as crucial as raw processing power.Intel faced the difficult task of developing both the hardware and software ecosystems concurrently, creating hurdles for integrators and cloud providers evaluating migration and support costs.
Despite these challenges, Intel successfully integrated these gpus into the Aurora supercomputer, a notable achievement demonstrating the chips’ ability to operate at scale and meet the demands of a complex workload. However, Intel’s frequent adjustments to its accelerator strategy ultimately discouraged vendors from making long-term commitments to an Intel-exclusive approach.
Looking ahead,intel is focusing on future designs like the Jaguar Shores rack-scale platform,which promises more advanced silicon,HBM4 memory,and a more mature OneAPI software stack. This transition signals a renewed effort to establish a stronger foothold in the rapidly evolving AI and HPC accelerator market.
