Image 1 of 2
Jeff McVeigh, vice president and general manager of Intel’s super computing group, announced today via a blog post that Intel is canceling its upcoming Rialto Bridge series of GPU Max for data centers and moving to a two-year cadence for releases. of GPUs for data centers. So the company’s next data center GPU offerings will come in the form of hybrid chips based on the Falcon Shores chiplet, but the blog notes they’ll be in production by 2025, a year after Intel’s previous projections. of 2024.
HPC-focused Falcon Shores XPUs are designed for supercomputing applications and combine CPU and GPU technology in a combined chipset, but now they will first arrive as a GPU-only architecture in 2025. They were supposed to arrive as a CPU architecture +GPU in 2024, which means Intel’s positioning against competing products AMD and Nvidia, due for release this year, will be severely affected: Intel will now be several years late for a key architectural inflection point for higher end chips (more on that later).
Image 1 of 2
Intel’s Falcon Shores XPU is key to competing with Nvidia’s Grace Hopper superchips and AMD Instinct MI300 data center APU. Nvidia’s Grace and AMD’s MI300 will be released this year with CPU and GPU cores in the same package with HBM memory. These designs represent a new type of architecture that provides tremendous benefits for HPC workloads and will be difficult or impossible to combine with hardware based on existing designs.
Intel tells us that the delayed Falcon Shores will come with only GPU cores in 2025, but hasn’t indicated when it will integrate the CPU cores into the design. As such, Intel’s HPC-centric designs will lag its competitors by several years. Furthermore, Intel will be forced to compete with the HPC-centric designs of AMD and Nvidia with their Xeon CPUs and Ponte Vecchio GPUs for several years, a significant disadvantage.
Falcon Shores represents the continuation of Intel’s heterogeneous architecture design arc with the ultimate goal of delivering 5x the performance per watt, 5x the compute density on an x86 socket, and 5x the memory capacity and bandwidth of existing server chips. Intel’s High Performance Computing (HPC) CPU and GPU roadmap converges with Falcon Shores, indicating that these chips will serve both roles in the future.
This disaggregated chip design will eventually have separate tiles of x86 compute and GPU cores, but Intel can use those tiles to create any combination of the two additives, such as a full CPU model, full GPU model, or a mixed ratio of the two. two. Intel notes that these tiles will be made on an unspecified Angstrom-era process node, though Intel’s 20A seems to fit the bill for tiles it could make itself.
Image 1 of 9
Intel’s current-generation Ponte Vecchio GPUs were meant to be followed by Rialto Bridge, the next-generation data center GPU expected in 2023. That won’t happen now, so Intel will be forced to use Ponte Vecchio to compete with Nvidia’s Hopper H100. , a key shortcoming at a time when LLMs like ChatGPT are rising in importance and generating billions in Capex investments.
However, the Rialto Bridge wouldn’t have helped Intel much in its competition against Nvidia’s incredibly powerful H100. Intel originally disclosed that the Rialto would feature up to 160 Xe cores, a substantial increase over the 128 cores present in Ponte Vecchio. In addition, it was said that the chip would come with unspecified architectural improvements, similar to a “tick”, which would confer up to 30% improvement in application performance over Ponte Vecchio. Intel listed the Rialto Bridge’s maximum power consumption at 800W, an increase over the Ponte Vecchio’s 600W peak. The Rialto Bridge was compatible with the Ponte Vecchio’s packaging, so it was designed to be a direct upgrade.
|AMD vs. Intel Roadmap||2022||2023||2024|
|Intel P Cores||Sapphire Rapids – Intel 7 – 56 cores||Emerald Rapids – Intel 7||Granite Rapids – Intel 3|
|AMD P-cores||Milan-X – 7nm | Genoa – 5nm – 96 Cores||?||?|
|Intel E Cores||—||—||Sierra Forest – Intel 3|
|AMD electronic cores||—||Bergamo – 5nm – 128 Cores||?|
On the data center side of the house, Intel says its Xeon products are still on track, and more importantly, its compute node roadmap is on schedule as well.
However, Intel doesn’t plan to release its Sierra Forest processors, a special hyperscale-optimized chip, until 2024. Meanwhile, AMD’s Bergamo will arrive this year, meaning Intel will fall behind its competition in another crucial architectural breakthrough for a while. prolonged period. time frame.
Similarly, Intel’s decision to slow its GPU release cadence is not ideal, as it will have to leverage older products to compete with much more advanced architectures for HPC, such as Nvidia’s Grace Superchips and upcoming exascale APUs. from AMD, Instinct MI300, which will be released in 2023.
Additionally, Intel will cancel its upcoming Lancaster Sound GPU for its Flex series of data center GPUs. These GPUs are designed for less intensive work like media encoding. Instead of moving forward with Lancaster Sound, Intel will focus on next-generation Melville Sound products for the Flex series.
Intel says the new release cadence is based on customer expectations for data center GPU products and generally matches the incremental release rate we see from other GPU companies, like Nvidia. The moves come in the wake of Intel’s recent restructuring of its AXG graphics group to address the gaming and data center markets separately by placing it under two other business units. The restructuring was designed to increase the focus on the end markets served by GPU products, and these new developments represent a further narrowing of the focus. Additionally, Intel says it will now enhance its focus on its software ecosystem and provide ongoing updates to the Max and Flex series GPUs that include more performance, features, and expanded operating system support.