Nvidia's competitors are making efforts again

2024-07-06

Ampere Teams Up with Qualcomm to Launch AI Servers.

According to multiple media reports, chip startup Ampere Computing LLC, which is backed by Oracle, has partnered with Qualcomm to enter the field of artificial intelligence, jointly launching AI servers. It is understood that the two companies have collaborated to develop devices based on Ampere processors and Qualcomm's AI 100 Ultra accelerator chips, the latter of which are products that help AI models process large amounts of data. The new devices will be applied to servers produced by Super Micro in the United States.

Chip manufacturers have been developing hardware and software to ensure that new products work smoothly and efficiently in so-called AI inference (image or voice recognition). Like Ampere and Qualcomm, many chip manufacturers are racing to capture a share of the multi-billion dollar AI infrastructure market. Currently, AI devices are dominated by NVIDIA, and most companies spend a significant portion of their funds on NVIDIA's products.

Ampere's product chief, Jeff Wititch, stated that for those who do not want to spend a lot of money on NVIDIA equipment, the company's newly launched servers can provide "five times the performance for every 1 US dollar." Wititch believes that the company's new products can address the increasingly serious problem of data centers: AI-related infrastructure leading to uncontrollable growth in power consumption.

Advertisement

At the same time, Ampere also announced on the same day plans to launch a leading 256-core processor in the industry next year, higher than the current chip's 192 cores, using TSMC's 3nm process.

Although Ampere and Qualcomm's new products will not compete directly with NVIDIA, there is an indirect competitive relationship because AI chips are usually sold in systems that are paired with a variety of chips.Tirias Research founder Jim McGregor stated that the collaboration between the two companies can prevent potential competitors from acquiring customers; for them, the key is to keep competitors out of the data center.

Competitors Target NVIDIA

NVIDIA has become the king of the adjacent field to the core processor domain dominated by Intel, with its GPUs for accelerating artificial intelligence applications reigniting the data center market.

Intel has long dominated the entire server market with its Xeon CPU series. Five years ago, Intel's PC chip competitor AMD re-entered the lucrative server market after a long absence. According to Mercury Research, AMD has already captured a 23% share of the server market, while Intel still holds a dominant position with a 76.7% share.

Nowadays, the data center story is all about GPUs, with NVIDIA's GPU sales growing much faster than the core server CPU.

Intel is fighting back, seeking to revive growth in the data center and personal computers, both of which are declining after a significant increase in information technology and personal computer spending during the pandemic. This month, Intel launched a new series of chips for servers and personal computers, aiming to accelerate artificial intelligence locally on the devices themselves, which can also alleviate some of the artificial intelligence computing load in data centers.

At the same time, AMD is directly targeting the hot GPU market and the PC market. The company also launched major products this month, introducing a new GPU series that has been widely praised on Wall Street, as well as new processors for data centers and PCs. The company predicts that AI GPU sales will reach at least $2 billion in the first year of launch, posing a huge challenge to NVIDIA.

Another development that will further shape the computing hardware landscape is the rise of the x86 competitive architecture, namely Reduced Instruction Set Computing (RISC). In the past, RISC mainly entered the computing field of mobile phones, tablets, and single-task dedicated embedded systems through chip designs from Arm and Qualcomm.

NVIDIA once tried to acquire Arm for $40 billion, but the deal was not approved by regulatory authorities. Instead, Arm went public earlier this year and has been promoting its architecture as a low-power option for artificial intelligence applications. NVIDIA has been working with Arm for many years. Its Arm-based CPU is named Grace, which is used in conjunction with the Hopper GPU in the "Grace-Hopper" artificial intelligence accelerator for high-performance servers and supercomputers. However, Tirias Research analyst Kevin Krewell pointed out that these chips are still often used in conjunction with Intel or AMD's x86 CPUs in the system.Krewell stated in an email: "Due to more modern instruction sets, simpler CPU core designs, and less legacy overhead, the Arm architecture has an energy efficiency advantage over x86. x86 processors can close the gap between Arm in terms of power consumption and core count. That is to say, there are no restrictions on running applications on the Arm architecture, except for traditional x86 software."

Until recently, systems based on Arm only accounted for a small share of the server market. However, the open-source version of RISC-V is now attracting the attention of large internet and social media companies as well as startups. Power consumption has become a major issue in data centers, and the amount of energy used by AI accelerators is incredible, so companies are looking for alternatives to save on electricity usage.

"The share of Arm CPUs is rapidly increasing, but most of them are internal CPUs (such as Amazon's Graviton), rather than products sold in the open market," said McCarron. Amazon's Graviton processor series was first launched in 2018 and has been optimized to run cloud workloads in Amazon's web services business. A report from The Information earlier this year stated that Alphabet is also developing its own custom Arm-based CPUs, codenamed Maple and Cypress, for its Google Cloud business.

In addition, some AI chip and system startups are designing around RISC-V, such as Tenstorrent, a startup co-founded by renowned chip designer Jim Keller.

Leave a comment

Submit