Thus, after a lot of speculation, NVIDIA has officially announced the acquisition of ARM chip design company from SoftBank. The $ 40 billion deal will see ARM belong to a chip giant for the first time. threatens the balance of power in the mobile world.
It seems that NVIDIA already anticipated this. In an official announcement, the GPU giant affirmed that ARM will “continue to operate as an independent and maintain neutrality towards global customers”. Although not directly mentioning major customers (of ARM) such as Apple, Qualcomm, Samsung, MediaTek …, NVIDIA’s CEO is billionaire Jen Hsun-Huang said “NVIDIA spent a lot of money acquiring ARM and there is no reason to do anything that could keep customers away.“.
Invade Intel and AMD
Immediately after announcing the acquisition of ARM, NVIDIA hastily asserted that its ambition was not to return to compete with Apple, Qualcomm and Samsung.
In other words, ARM is not looking to return to the mobile chip market in the near future. But of course, no tech giants spend $ 40 billion without big ambitions. For NVIDIA, that ambition is the server market: “We know the cloud and data centers are waiting for ARM processors. Energy consumption efficiency directly leads to limited computation, processing speed and service delivery cost“.
Right now, at least two big guys see a threat from NVIDIA: Intel and AMD. Apparently, NVIDIA’s sum of $ 40 billion has turned ARM into a weapon to attack the server market, which has been owned by Intel and AMD together. According to the latest IDC data, 90% of server revenue in Q1 / 2020 was spent on x86, ARM processors made up only the remaining 10%.
This is the most stable and potential source of revenue for Intel and AMD. In the second quarter of the year, social disparity caused revenue from online and cloud services to soar. As a result, AMD’s Epyc sales doubled over the same period in 2019. Intel, despite going through a period of crisis, still recorded a 43% increase in revenue from the server business last quarter.
Threats from NVIDIA
The failure of Qualcomm and AMD shows that not everyone can bring ARM to the cloud.
But, unlike its predecessors, NVIDIA has proven its superiority in the cloud sector.
But the threat from ARM has grown stronger over the years. Compared to x86, ARM is known to have better power efficiency (less heat when processing the same computation). As a reference design, ARM also allows companies to create customized chips for their own needs. For example, Amazon has designed its Graviton chip for deployment to the AWS cloud. Most notably, some companies have customized ARM to create better parallel processing power than x86 – which is completely understandable given the superior performance of ARM chips.
The ARM server’s biggest problem is the capacity of the chipmakers who turn ARM’s reference design into practical power for business. Not everyone can use ARM to make finished chips that outperform x86. Not everyone has the capacity to create a software ecosystem and support services like Intel. This is the biggest worry for server makers, data centers, and cloud providers, leaving them to stay with x86.
This mentality has caused the majority of ARM server chips to fail. AMD and Qualcomm are on the list of those who dreamed of bringing ARM to the cloud: the ARM chips for servers of these two companies, Opteron and Centriq, did not live more than 2 years after launch. AMD recently returned to the server market in an impressive way with Epyc, a line of x86 chips that are rated to outperform Xeon at a cheaper price point.
Luckily for ARM, the new “boss” of the company has the power to dispel these doubts. Its success in AI has allowed NVIDIA to penetrate the server market – server makers and cloud providers have the right to believe that ARM CPUs will be as well supported by NVIDIA as the Tesla GPU. Last year, NVIDIA even launched a reference design for NVIDIA’s ARM supercomputers and GPUs. CUDA-X, a suite of libraries and tools popular in the data and AI communities, will be fully supported on the aforementioned platform.
The last piece of the puzzle from Apple
NVIDIA / ARM’s ultimate concern: ARM cloud alone is not enough, professional PC makers have to switch to ARM as well.
However, NVIDIA and ARM still have one final concern: software developers. When it comes to ARM and x86, the father of the Linux operating system – which is the dominant operating system on servers and the cloud, once said that ARM will never succeed:
“If you develop software on x86 you will want to deploy on x86 environment (server) … You will pay more to buy cloud x86 simply because this environment coincides with what you can test on your computer, and the errors you encounter will be easier to spot“According to him, x86 became the dominant platform for servers because in the past, developers chose this architecture for their personal devices. From personal devices, x86 has spread to the market.” host and kill other architectures.
This statement comes from 2019. In 2020, a move for Apple could help ARM reverse the battle with Intel: Macs are migrating from x86 to ARM. If Linus thinks that PCs and servers need design consistency, future Macs will be the choice for those who want to bring software to the cloud running ARM chips (and GPUs). NVIDIA).
Luckily for NVIDIA, both Apple, Google and Microsoft are looking to the future of PCs on ARM.
Apple is not alone. Microsoft in 2019 released the first Surface to use a Qualcomm chip instead of Intel. Google’s low-cost ARM Chromebooks have also flooded the market. Most notably, with even ARM cloud Smartphones can also be transformed into professional PCs – compatibility will be pushed to the highest level.
The only question remaining is, how will Apple, Samsung, Qualcomm, Microsoft, Google … react to the deal of NVIDIA? Currently, this acquisition has only been announced but has not been approved by law enforcement. There is also a risk that Apple and other chip design companies are so strongly opposed that it is possible to completely replace ARM and switch to RISC-V. Incidentally, Intel’s future fate is heavily dependent on those who didn’t choose x86.