After the AI boom in 2022, NVIDIA has been the leading manufacturer in the market, but facing AMD's continuous pursuit and aggressive expansion of its product lines, NVIDIA will become increasingly difficult to face the situation. Foreign media...
After the AI boom in 2022, NVIDIA has been the leading manufacturer in the market, but facing AMD's continuous pursuit and aggressive expansion of its product lines, NVIDIA will become increasingly difficult to face the situation.
Foreign media WCCFtech analyzes AMD's recent strategies, saying that NVIDIA has achieved advantages in the AI market because it first launched AI chip solutions. At that time, AMD and Intel were still in the traditional CPU and GPU stages. It was not until the chatbot ChatGPT that AMD realized that AI development was no longer unstoppable, so in 2023 it announced that it would fully convert to AI.
But during this period, NVIDIA had occupied the market with Ampere and Hopper architectures and built a strong software ecosystem, AMD was like a single-smashing market.
Under the AI craze in 2023, it is absolutely not easy to destroy NVIDIA's control, because many technology giants have signed contracts and can only rely on their AI chips and cannot choose other alternatives. At the same time, under the "city protection" ecosystem created by CUDA and other services, users can hardly turn to AMD's embrace.
Huang Rensha successfully established a market that can only be operated by their products, which led to the fact that AMD launched AI accelerators such as the Instinct MI300X, it was difficult to gain large-scale customers' adoption.
AMD not only competes in chip performance, but also the entire NVIDIA ecosystemAMD's first major AI solution is the Instinct MI300 series, which is comparable to the NVIDIA H100. It not only has twice the memory capacity and is comparable in memory bandwidth, but also has a CDNA 3 architecture with significantly improved performance. According to internal tests, the MI300X has indeed threatened the H100 on AI-recommended applications.
However, AMD is not facing a single opponent, but a complete ecosystem established by the entire NVIDIA.
Due to its control over development tools and AI frameworks, NVIDIA has become the first choice platform in the industry, and AMD is considered a "challenger brand". To build an AI-based construction not only purely to purchase AI chips, but also to consider the entire soft hardware system. Even if AMD provides stronger hardware capabilities, the overall cost of turning to AMD is still higher than that of continuing to use NVIDIA chips.
For NVIDIA's city river ecological environment, AMD chooses to open the route with hug. Currently, AMD has launched the open-source AI software platform "ROCm", which will free users from the CUDA ecosystem and provide support for high-performance computing (HPC) and AI Instinct GPU. In addition to existing Linux operating systems, ROCm now fully supports Windows operating systems and can be run on client devices such as AI PCs and workstations. In addition, in the AI network technology sector, AMD also meets the "Scale-up+" and "Scale-out" expansion standards through UALink and UEC.
AMD recommends Instinct MI400, Helios solutions to fightIn addition, the media's vocabulary rights will also shape different industrial atmospheres. According to the report, many technology giants will mention in the press release that the use of NVIDIA equipment will not only boost investors' confidence, but also make NVIDIA occupy the main vocabulary rights in the market. If OpenAI and Microsoft change their words and use AMD's equipment, they may be interpreted by the outside world as negative ideas such as "declining technical strength", which will lead to AMD's faced image failure.
However, after AMD launched competitive products, it actively promoted large technology companies with more attractive prices, higher supply capabilities, and an increasingly sound ecosystem. It is reported that its MI300 series is 20%-30% cheaper than NVIDIA's similar products and has better performance ratios. It has gradually been adopted by companies such as Microsoft and OpenAI, but its investment scale is still not as good as NVIDIA.
As for AMD's next-generation product, the Instinct MI400 series, expected to be launched in 2026, it is designed for large-scale training and distributed reasoning design. The MI400 is equipped with a high-level 432GB HBM4 memory, and the memory capacity is 50% higher than the existing solution, and the memory bandwidth reaches 19.6TB/s. At the same time, the peak computing power with FP4 accuracy is doubled to 40 PFLOPS, the peak performance of FP8 reaches 20PFLOPS, and the vertical expansion bandwidth of each GPU can achieve high-frequency width interconnection across the rack and the cluster.
In addition, AMD also expects to launch the AI rack-mounted server architecture "Helios" next year (2026), which is also the company's first AI rack-level solution, and the number is to compete with Rubin NVL72 solution..
Helios integrates AMD EPYC Venice CPU, MI400 series GPU, Pensando Vulcano NIC network card and ROCm software to support tightly coupled DIRECT expansion domains, can accommodate up to 72 MI400 series GPUs, DIRECT expansion bandwidth up to 260TB/s, and supports Ultra Accelerator Link.
Reported that these next-generation products all prove that AMD is preparing to be on par with NVIDIA in computing power, but to break the market, the single-corresponding performance is still not enough, and a complete soft and hardware system that can match NVIDIA is also needed. However, the market is gradually accepting another view, that is, AMD does not necessarily replace NVIDIA completely, but coexist with its opponents.
Is AMD the Next Major Threat to NVIDIA’s Long-Standing AI Dominance? A Deep Dive into How the Firm’s Recent Strategies Might Put It in a Much More Competitive Position
Extended reading:
NVIDIA H20 Loading card connection! US export review and storm pressure creation in 30 years
Super low price to win Tesla! Samsung re-used the same trick, cut HBM3E price NVIDIA
