Goodbye to Rialto Bridge and Falcon Shores only in 2025 (with changes): Intel disrupts its HPC roadmap

intel he announced important changes as for the product roadmap for the HPC industry. The first is that after the current generation Ponte Vecchio, commercially known as Intel Data Center GPU Max, no successor will come known by the code name of Rialto Bridge (here more information).

The other novelty is that, by virtue of this decision, the Falcon Shores project scheduled for 2024 will be postponed to 2025. Presented as the fusion of x86 cores, GPUs and other IP on a single package, in intentions very similar to the Instinct MI300 project announced by AMD, Falcon Shores however, it will not immediately be a CPU+GPU product as anticipated, but initially only GPU. To tell Patrick Kennedy of Servethehomeciting Intel sources.

Intel’s new roadmap

Evolutions that seem to undermine the company’s competitiveness in the HPC world and that means only one thing: AMD (Instinct MI300) and NVIDIA (Grace CPU Superchip and Grace Hopper Superchip) – debuting this year – will not have a serious competitor to Intel in the xPU or “exascale APU” world for at least 3 years . Furthermore, given that the server / HPC market does not like change but stability, it is plausible that the industry’s trust in Intel will crack.

The announcement of these upheavals, in an already very difficult period for the company, came from Jeff McVeigh, interim general manager of the Accelerated Computing Systems and Graphics Group division. In a postthe executive announced that “with the goal of maximizing return on investment for our customers, we will move to a biennial cadence for datacenter GPUs“.

As summarized above, this implies the cancellation of Rialto Bridge to transition directly to Falcon Shores in 2025. “The flexible chiplet-based architecture of Falcon Shores will address the exponential growth of computing needs for HPC and AI. We are working on variants of this architecture that support AI, HPC and the convergence of these markets. This architecture will have the flexibility to integrate new IPs over time (including CPU cores and other chiplets) from Intel and customers, built using our IDM 2.0 model. The development of Rialto Bridgewhich was meant to provide incremental improvements over our current architecture, will be interrupted“.

The Flex series will also move to a biennial basis. Intel will then discontinue development of Lancaster Soundsuccessor to Arctic Sound M, focusing efforts on Melville Sounda project that will represent a “significant architectural leap over the current generation in terms of performance, features and the workloads it will support.”

“In addition to streamlining our roadmap, we are increasing our focus on the software ecosystem. We will provide continuous updates for our products Max and Flex Series, with performance enhancements, new features, expanded operating system support, and new use cases to extend the benefits of these products,” adds Jeff McVeigh.

Finally, an update on the Aurora supercomputers for the Argonne National Laboratory, a system predicted to offer computing performance superior to 2 exaflops thanks to over 60,000 Max Data Center GPUs and over 20,000 Max Xeon CPUs.” The installation is going well. […] Argonne predicts that the system will be accessible to the first researchers by the third quarter of 2023“.