Intel's PonteVecchio GPU and Future Plans

TapTechNews May 17th news, Intel recently issued a statement to the German media ComputerBase and Hardwareluxx, stating that the PonteVecchio GPU will still be provided in the form of cloud resources on the Intel Developer Cloud.

In the announcement, Intel said:

In order to meet the growing demand of enterprises for AI, we focus on leveraging the proven performance advantages and competitive price of the Intel Gaudi AI accelerator to achieve its rapid expansion.

The Intel Xeon is still the host CPU choice for HPC solutions, and our data center GPU Max can still be used on the Intel Developer Cloud.

This way will pave the way for developers and the ecosystem for FalconShores. FalconShores is our next-generation GPU for AI and HPC, which will utilize the Xe IP architecture that is the basis of the data center GPU Max.

Intel actually did not directly respond to what the foreign media ServeTheHome called 'The PonteVecchio GPU has entered the discontinued process', especially the core issue of whether the physical product will continue to be shipped.

As Intel's first data center GPU, PonteVecchio has strong performance.

But the chip is also criticized due to its overly complex multi-module structure: the extensive use of a large number of advanced packaging not only means insufficient production capacity and high cost, but also drags down the energy efficiency ratio of PonteVecchio.

Intel previously chose to skip the RialtoBridge, which is much simpler than PonteVecchio, in the data center GPU product line and go straight to FalconShores that integrates the Xe GPU architecture and the Gaudi AI accelerator architecture.

FalconShores is planned to be launched next year. ComputerBase said it learned that the overall power consumption of the GPU including the Ethernet connection will reach 1500 watts.

This value is twice the 750 watts maximum TBP of the AMD Instinct MI300X accelerator. Even without including the Ethernet connection part, the power consumption of FalconShores will be higher than that of any single GPU of Nvidia's Blackwell generation.

With such power consumption, liquid cooling is expected to become the mainstream choice for cooling the flagship model of FalconShores. TapTechNews previously reported that Intel plans to have the ability to cool devices with a TDP of more than 1300 watts in the medium term from 2024 to 2030.