start and win? Demand continues to grow

A source

We wrote about server processors with ARM architecture a couple of years ago. Even then they were not exotic, although they were just beginning to penetrate server development. The expansion, so to speak, of ARM chips to the server market became more active after their capabilities were appreciated by industry players. As it turned out, they perform very well.

Now, judging by the results of a study by a number of analytical companies, their popularity continues to grow, and this trend will continue in the foreseeable future. In particular, thanks to cloud platforms and data centers, which are gradually ordering more and more equipment based on ARM processors. More details – under the cut.

What’s up with the rise in popularity?

Having studied the market, TrendForce
predicted the development of the global market server processors with ARM architecture. The forecast is positive, according to experts, the growth in the supply of servers based on such chips can be called a record.

Chips and hardware based on them are in demand due to another trend – the active growth of high-performance computing platforms, as well as AI platforms. ARM chips consume less power than their traditional x86 counterparts. Well, if so, then they are quite willingly bought up by manufacturers of server equipment that is supplied to data centers.

According to analysts from TrendForce, by 2025, which, for a moment, is only three years, the level of penetration of ARM chips into the server market will increase to 22%. Of course, x86 will remain on horseback for a long time, but ARM servers will gradually become more and more serious competition.

ARM chips have many advantages, among which three stand out:

  • Ability to support heterogeneous and dynamically changing workloads. All thanks to the scalability that the chips provide, as well as their cost-effectiveness, as mentioned above.
  • Solutions based on ARM chips can be adapted to perform a variety of tasks, including very highly specialized ones.
  • In addition, the equipment is smaller than usual, so that the area is less.

The last point is especially important, since another trend has now emerged – the deployment of so-called small data centers by telecom operators and cloud service providers, which contribute to the active penetration of ARM processors into the market. In addition, the chips are in demand by hyperscalers and various edge computing vendors.

If you are reading our blog, you may be interested in these texts:

How we chose the best of the new NVIDIA graphics cards
Why data centers install dynamic uninterruptible power supplies
Is it worth building your own server enclosure?

Complicated history of ARM processors in the server market with a happy ending

The end of this whole story of the penetration of the ARM architecture on the market is still far away. But one thing is clear – this industry is developing more and more actively. A number of major chip makers have either already started making ARM-based server processors or are planning to do so in the foreseeable future. But it was not always so.

The fact is that the first server ARM chip was announced 10 years ago. This was done by Calxeda, which unfortunately had a lot of problems along the way and closed in 2013. However, she was the first to claim the production of such chips, talking in 2011 about a 32-bit processor based on ARM Cortex-A9. He allowed to create 2U servers with 120 processors (480 cores in total).

Then came the AMD Opteron A1100, which happened in 2016, which is two years later than the company originally planned. But these processors did not become popular and in demand. Then Qualcomm introduced its own version of the ARM processor for servers. Rather, the company regularly published news about the development, on which it spent about four years, but in the end the project was closed.

Marvell at the Hot Chips 32 event revealed details about the third generation of Thunder-X ARM processors, but they also did not become particularly in demand and popular. The problem was not the processor, but the fact that companies that buy server hardware, such as hyperscalers, were not ready for the new technology. They treated it with distrust, and if they acquired a new type of system, then only for testing. It all ended with the departure of the chief chip architect from the company, then the dismissal of more than 100 engineers.

But now, it seems that the “dark times” for ARM in the server industry are over. So, just a few days ago, NVIDIA announced its own 144-core Grace Arm processors, with a Grace Hopper hybrid. The Grace Hopper system combines both Grace (ARMv9) processors and Hopper accelerators. The Grace Superchip processor for AI and HPC workloads has 144 cores. In SPECrate2017_int_base, it showed 740, which allowed NVIDIA to talk about one and a half times superiority compared to a pair of AMD EPYC.

We should not forget about Huawei with its Kunpeng 920 chip. On its basis, the company produces a series of TaiShan servers designed to solve various problems – peripheral computing, high-load applications, etc. We already wrote about these servers after we tested them.

Okay, but what about servers based on ARM chips?

There are not many statistics, but at the end of last year, specialists from the analytical company Omdia published a Data Center Server Tracker report, according to which the supply of servers with ARM processors reached an all-time high. As it turned out, every twentieth server, which is about 170,000 pieces of equipment, is based on a new type of chip. The statistics was analyzed for the supply of equipment from July to September 2021.

In addition to the advantages of chips and servers of this type, as mentioned above, the growth was also due to the impact of the global shortage of electronic components. For this reason, many large companies began to look for equipment based on alternative CPU architectures.

In general, we have already written that several components are needed for the mass introduction of ARM platforms. The first is a good architecture, the second is a factory that is capable of producing chips according to a modern technical process. The third is innovation to create an energy-efficient crystal with a large frequency. And lastly, a huge circulation is required in order for the cost of such a chip to be competitive and the production of the processor to be profitable.

Most of the companies that produced ARM chips and equipment based on them experienced problems with just the last component. The prices were high, and the system performance was about the same as that offered by Intel, and at a lower price. But now, with the growing popularity of ARM servers, the situation is clearly changing for the better. The production of chips is becoming more large-scale, the industry requires more and more processors and equipment based on them.

So it may well be that ARM servers have a bright future, which is already somewhere nearby.

Similar Posts

Leave a Reply