Intel announces Rialto Bridge discrete GPU with up to 160 Xe cores and 800 watt TDP for data center deployment.

Intel has unveiled its next-generation Rialto Bridge data center GPU at the International Supercomputing Conference in Germany, positioning it for next-gen AI and HPC dominance. The Rialto Bridge will pack up to 160 Xe-Cores and enhanced tiles on the next-gen Intel process node, manufactured on Intel 4, with an expected arrival in 2023. Intel teased increased density, performance, and efficiency with the new architecture.

The Rialto Bridge will succeed the upcoming Ponte Vecchio GPU that launches later this year. While Ponte Vecchio features up to 128 Xe-Cores, the new Rialto Bridge GPU will deliver up to 160 Xe-Cores, representing a 25% increase in core count. This more powerful successor demonstrates Intel's commitment to scaling its data center GPU offerings for demanding workloads.

In terms of power consumption, the Rialto Bridge will reach 800W, a significant increase from Ponte Vecchio's 600W. Intel is expected to utilize super-fast HBM3 memory on the Rialto Bridge and has promised "increased I/O" bandwidth, though specific numbers were not provided. This contrasts with the Ponte Vecchio's 128GB of HBM2e memory on an 8192-bit memory bus.

Intel's Rialto Bridge GPU is designed to be better in every way, with the company aiming for up to 30% performance boosts in applications, though the company did not specify which applications these improvements would target. The improvements will come from higher GPU clocks, more memory bandwidth, and advantages delivered by the new Intel 4 process node used in its manufacture.