Intel to Join Forces with NVIDIA

2023-09-18

As we all know, despite the progress made by competitors such as Intel and AMD, NVIDIA remains at the top of the market with its performance in generative AI. Now, the giant Intel is making a full-fledged entry into the AI hardware field, and they may have found a breakthrough.

Intel plans to release another AI accelerator superchip, Falcon Shores 2, in 2026. "We are in the process of integrating GPUs and accelerators into a single product, and our roadmap is quite simple," said CEO Pat Gelsinger.

The recently released Intel Xeon Max 9480 combines 56 cores and uses 64GB HBM2e memory, which is equivalent to the memory used by GPUs and AI accelerators, instead of the standard DDR5 memory. Interestingly, Intel believes that these GPUs are primarily used for inference-based tasks rather than actual AI model training.

More leaked information about the upcoming 14th generation Meteor Lake processor suggests that it may adopt DDR5 memory, similar to Apple's M2 chip design. In addition, AI is expected to play a significant role in the Meteor Lake processor. People are eagerly anticipating Intel's Innovation 2023 event, which will take place on September 19th.

Many Paths Lead to AI

While Intel adjusts its strategy, it is worth noting that NVIDIA is also expanding into the CPU market with its GH200 supercomputer, further expanding its business in the GPU and AI technology fields.

Furthermore, initially, Intel's Falcon Shores chip was envisioned as a fusion of CPU and GPU cores, representing the company's first attempt at the "XPU" architecture for high-performance computing. However, a few months ago, Intel unexpectedly decided to release the chip using only GPUs and postponed its release until 2025. The company has experienced a series of twists and turns in its exploration of AI and GPUs.

This comes after Intel already offered the Gaudi2 AI chip for model training. Interestingly, the Gaudi2 works 2.4 times faster than the NVIDIA A100 and performs almost on par with the H100 Hopper GPU.

However, Intel's decision to slow down the release of GPUs may put it at a disadvantage against advanced architectures such as NVIDIA Grace Superchips and AMD's Instinct MI300, both planned for release in 2023. This strategic choice could impact Intel's competitiveness in the high-performance computing market.

Intel Believes in Two AI Markets

One involves infrastructure, for which the company has Habana Labs Gaudi. The other is the inference market, where Intel claims that CPUs like Xeon are sufficient for handling inference tasks.

This seemed like a viable approach until NVIDIA also entered this field. In NVIDIA's recent earnings call, Jensen Huang announced the company's plans to release L40S, a GPU specifically designed for fine-tuning and inference. Considering that NVIDIA H100 has already been used for training, the shift towards Intel's Xeon processors could be a significant leap.

In the midst of these strategic shifts, Intel announced a strategic partnership with Boston Consulting Group (BCG) in May of this year to promote generative AI. This collaboration aims to leverage Intel's AI hardware and software to create customized generative AI solutions for enterprises while ensuring data privacy and security.

In August, Anthropic announced a partnership with BCG to provide responsible generative AI services to enterprise customers. However, NVIDIA and Microsoft subsequently invested in Anthropic, adding some complexity to the situation. This demonstrates that AI is open for everyone to develop. This may suggest that Google-backed startups could use Intel's supercomputers to build generative AI, which is rare in a landscape dominated by NVIDIA GPUs.

Intel's GPU, NVIDIA's CPU

You read that right. All these strategic shifts are interpreted as Intel no longer directly competing with AMD's Instinct MI300 and NVIDIA's Grace Hopper processors, both of which have CPU+GPU designs. NVIDIA entered the CPU market in March of this year, raising questions about what other aspects this GPU giant wants to challenge Intel in.

On the other hand, Intel has explained the reasons behind this strategic adjustment. While the initial plan for Falcon Shores allowed for flexible configurations of CPU/GPU, Intel emphasizes the importance of enabling customers to leverage a variety of CPUs, including those from competitors like AMD and NVIDIA. However, considering the announcements about Gaudi2, potential Gaudi3, and NVIDIA's entry into the CPU market, Intel may soon gain a larger share of the GPU market.

The competition between Intel and NVIDIA, covering both CPUs and GPUs, is destined to intensify and potentially reshape the landscape of AI and high-performance computing. Intel is considered the best choice for CPU purchases, while NVIDIA is considered the best choice for GPU purchases. However, as discussions about computing revolve around AI, this situation may soon change.

RECENT AI NEWS

RECENT AI TOOLS