Linus Torvalds does not believe that servers on the ARM architecture will replace x86. “Selling a 64-bit model is idiocy”

Last week, ARM Holdings announced that it is developing a new micro-architecture for server processors. The computational core to be used in it is codenamed Ares, and the promises should give a 60% increase compared to the current platform. With each generation, productivity should increase by another 30%.

The server market is not yet the largest for ARM. Now the processors on its architecture are used in mobile and embedded devices. The performance leap that the company promises to server manufacturers will be higher than what Intel and IBM have done over the past few years.
')
However, Linux creator Linus Torvalds was skeptical about the announcement. He believes that the future of the new architecture is not so bright.

I can guarantee that while everyone is engaged in cross-platform development, the platform will not be stable and successful. Some people think that the set of instructions is not important for the “cloud” - you are developing at home, deploying and everything (by “at home” I mean not in my real sense at home, but in my working space).

This is complete rubbish. If you are developing on x86, then you want to deploy on x86, because you can run what you tested at home. So, you’ll be happy to overpay for hosting on x86, just so that it coincides with your working environment, and the resulting errors were broadcast more easily.

Therefore, providers will get more money from x86 servers and keep them in priority. Any options from ARM will be secondary, and most likely they will throw off all silly nonsense, like the frontend, static HTML and all that.

One of the declared advantage of the ARM architecture, which allowed it to win in the mobile market, is power consumption. The company believes that this will reduce the cost, and productivity will not be worse. This combination will help her compete among servers. But Torvalds thinks that success in the market is determined by other reasons.

In his opinion, the determining factor is precisely “home development”. Small companies are testing workloads on ordinary cheap PCs, and when loads grow, these computers take on the role of real servers. And only with a huge increase in loads, companies are transferring everything to the cloud, not wanting to change the architecture in order to avoid problems.

This is exactly what killed the vendors of RISC processors and made x86 the king of the mountains among the servers. To the extent that everything else is just an error. A couple of decades ago, this would have sounded like a fantastic invention.

Linus believes that entering the server market without creating a preliminary development environment and not “flooding the market with cheap devboxes is complete idiocy.” In addition, he doubts that the benefits that ARM calls are really considered advantages. According to him, all current servers on this architecture are in reality slower, more expensive and are unlikely to save so much energy.

The creator of Redis Salvatore Sanfillipo did not agree with Torvalds. He believes that most developers do not think of constant immersion in the computational cores and do not attach any importance to the reproducibility of the environment at the architecture level. According to him, the translation of Redis to the ARM architecture did not cause the problems that the creator of Linux scares:

Redis, which in itself is a low-level code, runs quietly on ARM, all tests pass, and there is no stability problem. And since the code written in C many years ago, when no one thought about ARM, works almost out of the box, the more so with Ruby or Node applications nothing happens when they are uploaded to ARM servers.

The architecture company also responded to Torvalds. They agree with his opinion that development in the framework of one environment works much better, so they announce their own development platform, probably this week.

Iron producers are also moving towards the changes that Torvalds says are necessary for the future of the new architecture. For example, Apple is rumored to be preparing to release Macs with ARM processors, Qualcomm is developing ARM processors for laptops, and Microsoft supports ARM development for Windows 10.

Torvalds himself in his next post also lowered the degree of skepticism:

Let's see what will happen in reality, but now I like ARM more than I did before.

Until I see the widespread distribution of iron that people can use to develop and deploy, I will hold onto their judgments. I just heard too many promises about iron, which after release was not needed anywhere.

Hopefully ARM will not be too hard on rescaling. Maybe they will succeed, but, honestly, I doubt it. It takes a lot of time and effort. No need to aim at 64-128 cores until you can make normally at least 8 cores. What they have not yet demonstrated.

But you never know, maybe they will surprise me.

In discussing the issue on Reddit, it was noted that most of the known architectures could not compete with x86.

m88k? Dead or moved to something like PowerPC. i860? Is dead. i960? Is dead. PA-RISC? is dead. AMD 29000? Is dead. IA64? Is dead. Alpha? Is dead.

However, the discussion agreed that the current market for server architectures is divided into three systems. ARM - as the weakest and cheapest. Power9 is the most powerful, but incredibly expensive. x86 among them is the golden mean in the price-quality ratio:

Developing and deploying to ARM is fine if you use Rasp. Pi. But you need something more powerful. With Power9, the situation is reversed: the cheapest system is the Talos II. Therefore, you need many thousands of dollars to build a normal devbox on Power9. Of course, it is good, but it is too expensive for normal development needs.

Therefore, x86 falls just in the niche of mass consumption - laptops and desktops cost up to a thousand dollars.

But in the second quarter of this year, Raptor Computer Systems plans to release a desktop with a 4-core CPU on the Power9 architecture for $ 1,200. Therefore, if this trend continues - ARM will increase productivity, and Power9 will reduce the price - competition may arise again in the mass segment.