A monster with more than 1000+ billion transistors! Intel's GPU Max will gradual

2024-08-03

For the Future, Intel Makes Trade-offs.

Intel has officially begun the process of phasing out its Ponte Vecchio HPC GPU. According to ServeTheHome, Intel has started retiring these GPUs two years after their initial release, as the company looks forward to launching the successor to Ponte Vecchio, Falcon Shores, as well as the Gaudi 2 and Gaudi 3 AI accelerators.

Intel has not completely phased out Ponte Vecchio, but it is no longer prioritizing the sales of Ponte Vecchio. More specifically, Intel will continue to produce Ponte Vecchio GPUs, but only for existing customers. This means that if new customers want to purchase Intel's latest GPU hardware, they will be out of luck and will be forced to look for alternatives from AMD and Nvidia.

Intel's decision to start phasing out Ponte Vecchio makes sense, as the HPC and AI markets are currently extremely competitive with the current artificial intelligence boom. Intel's Ponte Vecchio is, in fact, a last-generation architecture, thanks to AMD's newer MI300 and Nvidia's B200 Blackwell. Therefore, customers do not have much incentive to purchase Intel's current inferior GPU hardware when they can get newer, better hardware from competitors.

Ponte Vecchio was initially launched in 2022, and it is the largest GPU that Intel has produced to date. The architecture itself has over 100 billion transistors and is manufactured on five different process nodes. Depending on the configuration, a single Ponte Vecchio GPU may have up to eight GPU tiles, as well as four HBMe2 stacks based on Intel's Xe HPC architecture—more powerful compared to the regular Xe architecture that Intel uses in its consumer-grade Arc Alchemist GPUs.

Advertisement

It is also worth noting that the biggest showcase of Ponte Vecchio did not surprise anyone: the performance of the Aurora supercomputer is inferior to AMD's Frontier supercomputer, and it has a higher power consumption. According to the latest Top500 data, Aurora ranks second with 1,012 petaflops and 38,698 kW of power, while the older Frontier has 1,206 petaflops and 22,786 kW of power. Aurora performs better in some workloads (mixed precision), but it should reach 2 exaflops (2,000 petaflops)—so it falls short by half.

Despite its impressive size, compared to AMD's updated MI300 and Nvidia's B200 APU/GPU, Ponte Vecchio is now outdated. As usual, Intel is very late in delivering its MI300/B200 competitor in the form of Falcon Shores. This new architecture, originally scheduled to debut in 2024, combines CPU and GPU cores to compete with AMD's MI300 APU (which also has CPU and GPU cores) and Nvidia's Grace Hopper superchip. However, Intel has been forced to postpone Falcon Shores to 2025 and limit it to GPU-style chips only.Initiating the retirement process of Ponte Vecchio before the arrival of Falcon Shores will allow Intel to focus more resources on Falcon Shores and accelerate its production after its debut. This should help Intel catch up with AMD and Nvidia, but since AMD and Nvidia are also building next-generation HPC hardware, it will still lag behind the development curve.

Nvidia Continues to Dominate the 2024 AI GPU Market

Over the past year, the market demand for artificial intelligence (AI) and high-performance computing (HPC) has been growing rapidly. AI tools led by ChatGPT have sparked a global craze, significantly increasing the market demand for high-performance data center GPUs. This has also led to rapid revenue growth for Nvidia, with its stock price soaring, making it the most dazzling star in the industry.

According to relevant media reports, analysts have forecasted sales for the 2024 artificial intelligence (AI) GPU market, stating that Nvidia will continue to dominate the market with sales reaching $40 billion (approximately 289.876 billion yuan). In comparison, competitors AMD and Intel's sales are $3.5 billion (approximately 25.364 billion yuan) and $0.5 billion (approximately 3.623 billion yuan), respectively.

From this data, we can see Nvidia's influence in the industry, far exceeding its competitors in the sales of AI GPUs. The recently launched Blackwell architecture GPU has once again solidified its dominant position in the AI market. Such a huge gap may have exceeded many people's imaginations, and to some extent, the distance between companies participating in the AI competition may be widening.

Although AMD's similar products have excellent performance, the recent Instinct MI300 series has also been very strong in sales, and even $3.5 billion exceeds the initial internal expectation of $2 billion. However, it is still far behind Nvidia. AMD has recently done a lot of work, including in software, such as frequently updating ROCm, but it still seems far from enough. As for Intel, the gap is even greater. The recently launched Gaudi 3 accelerator has also fallen short of market expectations in sales, which is also one of the reasons for the stock price drop after the release of its financial report for the first quarter of 2024.

The future prospects of the AI market are broad. Although Nvidia occupies a dominant position, other chip companies are still very optimistic and hope to get a share of the pie. However, if AMD and Intel want to catch up with Nvidia in the field of AI GPUs, there is still a lot of work to be done.

Leave a comment

Submit