Intel delays its Falcon Shores data centre GPUs to 2025

In a blog post, Jeff McVeigh, VP and General Manager of Intel's super compute group, announced that the company's next data centre GPU offerings will be Falcon Shores chiplet-based hybrid…

By Nivedita Bangari

March 4, 2023

3 min read

Comments

In a blog post, Jeff McVeigh, VP and General Manager of Intel’s super compute group, announced that the company’s next data centre GPU offerings will be Falcon Shores chiplet-based hybrid chips, but the blog notes that these will be available in 2025 — a year later than Intel’s previous projections of 2024. Intel is also cancelling its upcoming Rialto Bridge series of data centre Max GPUs and shifting to a two-year cadence for data centre GPU releases.

The HPC-focused Falcon Shores XPUs are designed for supercomputing applications and combine CPU and GPU technology into a single chip package, but they will now be available as a GPU-only architecture in 2025. These were supposed to arrive in 2024 as a CPU+GPU architecture, which severely impacts Intel’s positioning against competing AMD and Nvidia products, both of which will launch this year — Intel will now be multiple years late to a key architectural inflection point for the highest-end chips.

The Falcon Shores XPU from Intel is critical for competing with Nvidia’s Grace Hopper Superchips and AMD’s Instinct MI300 Data Center APU.

Both Nvidia’s Grace and AMD’s MI300 will be available this year, with both CPU and GPU cores on the same package and HBM memory. These designs represent a new type of architecture that provides significant benefits for HPC workloads and will be difficult to match with hardware based on existing designs.

Intel says the Falcon Shores will first ship with only GPU cores in 2025, but it hasn’t said when it will incorporate CPU cores into the design. As a result, Intel’s HPC-centric designs will trail competitors for several years. Furthermore, Intel’s Xeon CPUs and Ponte Vecchio GPUs will be forced to compete with AMD and Nvidia’s HPC-centric designs for several years, putting it at a significant disadvantage.

Furthermore, Intel will scrap its upcoming Lancaster Sound GPU for its Flex Series of data centre GPUs. These GPUs are intended for lower-intensity tasks such as media encoding. Instead of pursuing Lancaster Sound, Intel will concentrate on next-generation Melville Sound products for the Flex series.

According to Intel, the new release cadence is based on customer expectations for data centre GPU products, and it roughly matches the incremental launch rate seen from other GPU companies such as Nvidia. The changes follow Intel’s recent restructuring of its AXG graphics group to address the gaming and data centre markets separately by relocating it to two other business units. The restructuring was intended to narrow the focus on the end markets served by GPU products, and these new developments represent a further narrowing of that focus.

Also Read:

Source