ARM server CPUs will further widen the performance gap with x86 CPUs
ARM server CPUs will further widen the performance gap with x86 CPUs.
After nearly ten years of ecological construction, ARM’s high-performance CPU has finally emerged in the server market.
Amazon and Marvell have both applied the ARM architecture to the cloud.
Ampere Computing launched its first cloud-native server CPU last year, which has already demonstrated performance and power consumption advantages.
Recently, Ampere revealed in its 2022 annual strategy and product roadmap update sharing that Ampere will release its self-developed core 5nm CPU AmpereOne this year.
Ampere Computing Chairman Renee James
Jeff Wittich, chief product officer of Ampere Computing, said: ” AmpereOne is also based on the ARM ISA. Every year customers expect higher performance, better energy efficiency, and better scalability. Developing and launching self-developed core CPUs can help us focus all our attention and focus on the needs of our customers and meet their corresponding requirements.”
That is to say , Ampere will further improve the performance and power consumption of its CPU after being recognized by the world’s seven ultra-large data center customers including Microsoft Azure, Alibaba Cloud, and Tencent Cloud.
As a representative of ARM’s high-performance server CPU providers, Ampere’s further improved products also mean that the gap between ARM server CPUs and x86 server CPUs will further widen.
Rely on high performance, low power consumption, cloud native to grab the x86 server market
A notable feature of ARM-based server CPUs compared to x86 server CPUs is higher performance and lower power consumption.
According to the data given by Ampere, the performance of its CPU exceeds that of traditional x86 processors by 3 times, and the performance-to-power ratio is nearly 4 times ahead.
Compared to x86 server CPUs, the Ampere Altra series can use 50% less power and deliver 200% more performance.
The key to the high performance of the Altra CPU is the high core count.
Jeff Wittich explained, “ Our 128-core product is now the industry leader, twice as fast as other CPUs. Only a single thread runs on each core and keeps all cores running at a consistently high frequency . The core provides a large-capacity low-latency dedicated cache, and with the help of an intelligent high-bandwidth Mesh interconnect structure, all high-performance cores are connected together, breaking the traditional CPU usage bottleneck that will produce diminishing returns when demand increases. Advanced DDR is also used. And PCIe technology design, can achieve maximum capacity, expand memory and I/O bandwidth. It can effectively avoid mutual interference between users, while providing excellent scalable performance, and ultimately achieve maximum utilization.”
There is also a very critical point. The Ampere Altra series is born for cloud computing. A
s a cloud-native CPU, Ampere removes the characteristics of traditional architecture that are not popular with the cloud in order to optimize power consumption and area, and achieve a single-core power consumption ratio lower than traditional CPUs. 67% or more.
“One of the biggest disadvantages of traditional x86 is that they are not specially designed for data centers, nor for cloud and cloud services. They were used in data centers in the past because there was no better choice than x86 CPUs at that time.” Jeff Wittich.
The advantages of cloud native processors can be embodied in that, due to the advantage of high core count, each core only runs a single thread, which can show better performance and security, without worrying about the increase in workload and pressure.
The performance degradation can also resist a lot of user interference.
Altra CPUs also have better scalability than x86 CPUs and are also useful for cloud computing customers.
Further widening the performance gap with x86 CPUs
Renee James, chairman of Ampere Computing, said, “We have applied for hundreds of patents on the performance and functions of our products. The latest product, AmpereOne, uses Ampere Computing’s self-developed core, is based on the 5nm process, and supports PCIgen5 and DDR5. The new product has begun to send samples. , very much looking forward to customer feedback.”
“Our upcoming self-developed core AmpereOne is also based on ARM ISA. The microarchitecture is very different from the Neoverse provided by ARM, but we can’t provide too much information yet.” Jeff Wittich revealed, “The self-developed core AmpereOne can be perfect It is compatible with our current products Altra and Altra Max cloud native processors. From the user’s point of view, the optimizations they have implemented for Ampere Altra/ Altra Max can also be perfectly applicable on our AmpereOne. Besides, We have also worked accordingly with compilers such as GCC and LLVM to further ensure compatibility across all of our products.“
Although no more specific information is available, Ampere’s CPU will further improve performance and power consumption, which will undoubtedly further widen the gap with x86 CPUs.
However, ARM’s high-performance CPU still has an obvious shortcoming, that is, the software ecosystem.
Ecology is still the short board of ARM server CPU
To improve the ARM high-performance computing ecosystem, both software and hardware need to be continuously improved.
In addition to cooperating with cloud platforms and hardware ODMs and OEMs, Ampere also cooperates with partners such as Gigabyte and ADLINK to develop development kits for artificial intelligence and autonomous driving. and other scenarios to expand application scenarios.
Ampere highlights that its high-performance, low-power edge cloud-native processors can also reduce power to an ultra-low 40W (32C) or less than 70W (64C).
That means Ampere Computing’s 32-core processor consumes 4.6 times less power than an equivalent x86 CPU, and can open up new territories other competitors can’t, such as Cruise’s self-driving cars.
Of course, Ampere is also exploring the direction of AI and hybrid quantum computing.
In addition, Ampere hopes to attract more developers by setting up the Ampere Developer Program, providing more than 130 applications in the developer community, and returning and updating on the website every day, ranging from databases to various solutions for AI inference .
“Through regression testing, we can guarantee a wide range of adaptability, while ensuring high-performance results based on adaptability.” Jeff Wittich said.
But in any case, the ARM high-performance computing ecosystem chosen by Ampere needs to be jointly built by more enterprises and developers to truly compete with x86 CPUs in the server market.
Statista, a global comprehensive data database, predicts that by 2028, the market size of ARM-based processors in the data center and cloud will reach $58 billion, more than 14 times that of 2019, and the market share will also grow from 5% in 2019 to 5% in 2019. 25%.
ARM still has some way to go in the data center cloud market.