Code-named Ponte Vecchio, Intel’s highest-performing Data Center GPU Max compute GPUs will be available as OAM modules for blade supercomputers. Additionally, the business is preparing cheap Data Center GPU Max add-on cards that can be installed in standard servers and workstations. Those cards will be up to date with the latest style and have a 12VHPWR power connector. A scaled-down single-stack version of Intel’s Ponte Vecchio CPU is used in the Data Center GPU Max 1100. It has 56 Xe-HPC cores, 48GB of HBM2E memory, and 448 512-bit engines, 448 4096-bit matrix engines.
The device uses up to 300W of energy. According to an Intel video, the graphics card depends entirely on a secondary PCIe 5.0 power port. The GeForce RTX 4090 from Nvidia was the first consumer graphics card to use the 12VHPWR power connection. Sadly, 12VHPWR connections now have a terrible reputation among fans. There have been three instances of the native 16-pin power connector experiencing the same breakdowns, and there have been 20 customer reports of the 16-pin power adapter melting.
Let’s hope the data center professionals know better how to insert that 12VHPWR adaptor correctly, HardwareLuxx.de’s Andreas Schilling playfully tweeted. There have been a few hypotheses from professionals in the area, but nothing definitive since we are still awaiting Nvidia’s investigation findings. Why Intel used a 12VHPWR power connector for its 300W card is unknown. However, the business likely chose to create a PCIe Gen5 solution mainly employed with Intel Xeon Sapphire Rapids servers created following the new standard.
The Data Center GPU Max 1100 board from Intel has been designed for both general-purpose servers and HPC machines. The card can operate as a single board, in pairs, or 4-way setups using Xe-Link interconnects. Four of these accelerators in a PC will require 1,200W of power, which is a significant amount. The Arc Alchemist graphics cards from Intel and AMD use eight-pin PCIe auxiliary power connectors. The 12VHPWR power connectors are currently only used by Nvidia’s GeForce RTX 2080 and 2090 graphics cards but will soon be available for other manufacturers’ cards.