Broadcom (NASDAQ: AVGO) and Marvell Technology (NASDAQ: MRVL) are eerily similar companies. Both of them produce networking equipment for data centers, but they have also launched custom AI chip businesses with a few critical customers. Custom AI chips are likely to grow in popularity, as they can offer superior cost performance to traditional GPU-based training.
This primes both stocks to benefit as these units become more popular. But which one is the better buy? I think there's one metric that settles the debate.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
To date, most training and inference has been performed on graphics processing units (GPUs), mostly made by Nvidia (NASDAQ: NVDA). GPUs are incredible computing units, and are great for a wide variety of workloads. The amount of flexibility that Nvidia has designed into its computing unit is amazing, but if a GPU only sees one workload type during its whole service life, then this flexibility is wasted. GPUs are still necessary in the computing landscape, as there are several reasons why GPUs should still be utilized; however, custom AI chips could provide better performance in some areas.
Custom AI chips are known as ASICs, application-specific integrated circuits. ASICs are nothing new, but their application to AI is. By building a computing chip around a workload rather than the other way around, designers can optimize and make these products efficient and cost-effective. This leads to more computing power at a lower cost, often outperforming GPU-based computing.
#custom
This primes both stocks to benefit as these units become more popular. But which one is the better buy? I think there's one metric that settles the debate.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
To date, most training and inference has been performed on graphics processing units (GPUs), mostly made by Nvidia (NASDAQ: NVDA). GPUs are incredible computing units, and are great for a wide variety of workloads. The amount of flexibility that Nvidia has designed into its computing unit is amazing, but if a GPU only sees one workload type during its whole service life, then this flexibility is wasted. GPUs are still necessary in the computing landscape, as there are several reasons why GPUs should still be utilized; however, custom AI chips could provide better performance in some areas.
Custom AI chips are known as ASICs, application-specific integrated circuits. ASICs are nothing new, but their application to AI is. By building a computing chip around a workload rather than the other way around, designers can optimize and make these products efficient and cost-effective. This leads to more computing power at a lower cost, often outperforming GPU-based computing.
#custom
1 day ago
