Ampere's next-generation processors will abandon Arm and use customized cores-EEWORLD

Collect

Last week, Microsoft announced the general availability of Azure virtual machines powered by Ampere Altra processors in 10 Azure regions around the world. This marks another major win for Ampere, which is winning over many vendors in the cloud space, including Google, Cloudflare, Tencent, and Oracle, among others.

These new virtual machines are designed and configured to efficiently run emerging scale-out, cloud-native workloads such as application servers, massive databases, cloud gaming and media servers, and more.

“Ampere’s cloud-native processors are uniquely designed to meet both the high-performance and energy-efficiency needs of the cloud. Through our close collaboration with Microsoft, the Ampere Altra processors are now generally available as Azure virtual machines, helping new cloud-centric processor users so they can deploy the next generation of innovative cloud applications at scale and in a sustainable manner,” said Jeff Wittich, Chief Product Officer at Ampere.

Ampere's current strategy is a bit different from many of its main competitors. At this point, the company is not focusing on discrete accelerators for specialized workloads, such as artificial intelligence. Instead, Ampere is designing multi-core processors specifically optimized for cloud services and workloads, and if the wide range of partnerships Ampere has established is any indicator, then it means that this strategy seems to be paying dividends.

Cloud-native processors designed for the modern data center

Ampere’s cloud-native processors are well suited for cloud workloads for a variety of reasons. First, Ampere has built its current Arm-based chips and platforms for optimal power efficiency and performance per watt. This is particularly important for today’s (and tomorrow’s) data center needs and in regions where more data center centers are being built, as the demand for processing resources continues to expand and tight power constraints align with the desire to build more sustainable data centers.

Ampere's processors also don't suffer from the problems that many of today's x86 processors do, which take advantage of multithreading, or SMT. SMT lets a single processor core handle two threads at the same time. This is a good thing for many workloads, but in the cloud, some threads may require the full resources of a single processor core, which can lead to severe resource contention as two threads compete for processing power. This can result in erratic, unpredictable performance, which is a problem for cloud service providers (and their customers). Ampere's processor cores are single-threaded, which means there is no resource contention per core and more predictable performance. With Ampere, each thread runs individually on its own core.

Ampere’s processor designs also offer near-linear scaling as cores are added. For example, when Ampere released a 128-core Altra processor to replace its previous 80-core product — a 60% increase in core count — performance improved by about 57% on some benchmarks.

Next-generation AmpereOne processor will use custom cores

Despite having already established key partnerships with leading cloud service providers and growing rapidly since its founding in 2018, Ampere is still accelerating. Earlier this year, the company announced that its next-generation AmpereOne processor will eschew existing Arm Neoverse cores in favor of fully custom cores, designed in-house, and built on the most advanced next-generation process nodes. Going forward, the company also plans to release each generation of products at an annual frequency, all of which will use custom cores.

Although details are still scarce, some information about AmpereOne has been revealed. The first wave of chips will be available in 2023 and will be manufactured using a 5nm process with cores specifically designed to further improve performance, scalability, and efficiency. Ampere learned where optimizations needed to be made by analyzing workloads running on its existing parts, and plans to configure AmpereOne optimally for the needs of evolving cloud workloads.

The move to its own custom cores should help Ampere further differentiate itself in the competitive data center market and distance its cloud-native processors from future competition.

Keywords：Arm Reference address：Ampere's next-generation processors will abandon Arm and use customized cores

Previous article：Achronix Acquires Key IP and Expertise from Accolade Technology, a Leader in FPGA Networking Solutions
Next article：How did the Arm architecture become the cornerstone of global computing step by step?

Recommended ReadingLatest update time:2024-11-16 11:43

Arm architecture Arm core analysis

　The Arm architecture dominates the embedded processing and computing market today, but it has come a long way over the past few decades. It started as a processor for home computers in the 1980s, then became the basis for mobile phone chips in the 1990s. Today, Arm is a strong competitor in almost every technology ma

[Microcontroller]

ARM Study Notes 002 gcc-4.3.2 compiler does not support hardware division operation solution

Generally, if gcc cannot use division, the error after make is: Division is used at each error location. My makefile code is as follows: CC = arm-linux-gcc LD = arm-linux-ld AR = arm-linux-ar OBJCOPY = arm-linux-objcopy OBJDUMP = arm-linux-objdump INCLUDEDIR := $(shell pwd)/include CFLAGS := -Wall -O2 CPPFLAG

[Microcontroller]

ARM Study Notes 002 gcc-4.3.2 compiler does not support hardware division operation solution

S3C2440 ARM chip clock system

The clock control logic in S3C2440A can generate the necessary clock signals, including FCLK of CPU, HCLK of AHB bus peripherals and PCLK of APB bus peripherals. S3C2440A contains two phase-locked loops (PLL): one for FCLK, HCLK and PCLK, and the other for USB module (48MHz). Figure 7-1 shows the block diagram of

[Microcontroller]

ARM Basics: Why does C language (function call) require a stack, but assembly language does not need a stack

I have read a lot of analysis about uboot before, which mentioned that the stack should be prepared for the operation of C language. In the start.S assembly code of Uboot, I also saw the stack pointer initialization action for system initialization. However, I have only seen people say that system initial

[Microcontroller]

[ARM Application] Solve the problem of LCD screen automatically closing and closing without changing the kernel code

It can be implemented in the application. Do not change the kernel driver. Write a small program: #include stdio.h #include fcntl.h #include sys/ioctl.h void keep_LCD_screen(void) { int fd; fd = open("/dev/tty0", O_RDWR); write(fd, "\033 ", 8); close(fd); } int main(void

[Microcontroller]

How to learn embedded development ARM

1. What is embedded? 　 2. What knowledge is needed for embedded systems? 　　　Knowing these two points, it's easy! I started planning my study route. The basic theory of computers cannot be lost, so I focused on the composition principle, data structure, operating system, and C++. At the same time, I learned from the

[Microcontroller]

ARM processor architecture--processor working status

Before, I have been looking at ARM driver development, and recently I have looked at UCOS, and I have also begun to have a deeper understanding of ARM architecture. When I studied microcomputer principles in college, I learned the x86 architecture, which is quite different from ARM. Below I will sort out some of the c

[Microcontroller]

ARM9 study notes - MMU

I remember one time I applied for an ARM-Linux software engineer position. I was asked how virtual memory is managed in ARM. Since I only knew about MMU in X86 platform, I was stumped. It turned out that what I learned was just the tip of the iceberg. There are still many things worth learning in depth. To develop dri

[Microcontroller]

Popular Resources
Popular amplifiers