One piece falls, one piece grows.

I already wanted to write about NVIDIA's purchase of ARM, but suddenly something changed the game, and so I decided to try to put things together. I mean, how shortly the illusion of winning the ISA war lasted.

First, what is an ISA and why does it matter?

Let's put it this way, each CPU speaks its own language. This "vocabulary" is called the "instruction set", and includes the name (in binary) of the instructions, describes the effect of the instructions and the input data, as well as the registers used by the instructions themselves, and other details.

When using an intel processor, you are using an ISA which is patented by Intel. If you are using a 64-bit Intel processor, you are using an AMD-patented ISA, which is why 64-bit linux packages have the name “amd64” in their name. X64 is also used as a name, but not IA64, or IA64e, which exist but have minor differences from AMD64. (some instructions behave differently in some cases).

Here a distinction should be made between architecture and ISA, but it would go into too much detail, so purists refrain.

In practice, at the beginning AMD used Intel's patented ISA, while from the 64-bit version AMD won, becoming the de facto standard after AI64 and the Itanium processor (developed together with HP on the EPIC concept) showed itself. a failure, after several vicissitudes.

Then there is ARM. It comes from an English company that produced computers such as the Acorn Archimedes, but also the spectrum and the BBC, famous in the 80s. She later stopped making whole computers and specialized in ISA design. If you want to make a processor compatible with ARM (it means that the code you compile will run, so you can make it an android, on a linux, or a windows for IoT) you have to pay a license (only for the most recent versions, the old ones are now free) and go.

There are also other ISAs, such as PowerPC, MIPS, SPARC, PA-RISC and others. The oldest is the IBM360 (still used for mainframes), which is over 50 years old.

This is a big obstacle for those who want to enter the world of processors, as the Chinese wanted to do. That is, you have to invent an ISA from scratch, which then becomes an ISA for mobile phones, one for computers, one for servers, and so on. 32, 64, 128bit….

The Chinese have tried several times, but intense barrage based on patent trolling has "convinced" them to give up. The longsoon in fact uses MIPS instructions, which binds the Chinese to manufacture the CPU at companies that have paid the license to the MIPS consortium (such as STM, to name one).

But that still means depending on a possible American veto .

So by buying ARM, which becomes American, the US thought they had conquered the monopoly of the CPUs. Depressing, you will say. Depressing, you will say.

But it lasted a few days .

hurray

What does it mean? It doesn't mean that tomorrow you will find RISC-V computers in every supermarket. But RISC-V has one feature: the ISA is opensource. It is now in the Linux kernel trunk, in the gnu Libc libraries, and in the GCC compiler. In fact, there are already cards that run linux for RISC-V.

The fact of being opensource is fundamental. It is fundamental because any nation / company today can produce its own processor.

But this ISA has a peculiarity. The specifications are now defined in practically every aspect, it covers 32-bit uses (typically embedded), 64-bit ones, and also defines uses for Hypervisor and Supervisor, reaching up to 128 bits.

In a nutshell, with RISC-V you could theoretically build everything from GPUs to mainframes to IoT for cars. And you don't have to pay any licenses.

It is a huge advantage, for a reason: the Chinese already have their processors for supercomputing, such as the Sunway, but in addition to imitating the architectures of DEC (patents now owned by intel) and therefore limited to the Chinese territory, and 'a processor made for supercomputers. It cannot exist on your cell phone or IoT.

Zaoxin takes common extensions with Intel from VIA Technologies, so in the event of a confrontation with the US it would face severe problems in Western markets. The FeiTengs are ITANIUM compatible (for the first series) or SPARC, for the second series. Still, that is, within the reach of US sanctions. The third generation uses the ARM64 ISA. Even worse.

China, that is, has several companies capable of producing CPUs, such as Ingenic and Unisoc , but also in the case of the former, which has an XBURST architecture, the ISA is MIPS32. Vulnerable to US attacks.

In short, once ARM was also bought by NVIDIA, a US company, the games were over for the American government.

But no. RISC-V breaks all games, because as ISA it is completely free from patent war risks. Anyone can go to their favorite silicon foundry, bring a CPU design made with the RISC-V ISA, and get a CPU compatible with any other CPU made with the same ISA.

And as if that weren't enough, RISC-V has a core and extensions, scaling from small chips (like that of a smartwatch) to supercomputer and mainframe processors.

The paradoxical thing is that RISC-V is developed in the USA, and HiFive is a spinoff of the University of Berkeley, where RISC-V was created. But the point is different: the point is that with the release of the complete specifications for computers and servers, RISC-V fully opens up the games in the world of CPUs.

Because here's the thing: As long as RISC-V made us small appliances like controllers and embedded stuff for cars, everything was quiet. But if you get to a main board to make a PC, it means that other specifications are considered stable.

In short, the USA believed they had won the game forever, and the asteroid comes from the USA.

Are the Chinese interested in this ISA? Yes a lot.

See who the sponsors are here at https://riscv.org/

heating

Moral: anyone who believes that the US will hold technological supremacy for a long time is deluding themselves.

If it took Huawei a year (since the beginning of the trade war with the US) to announce its new mobile OS, it means that buying ARM does not guarantee anything not only in the long term, but not even in the medium and short term.

For now, Huawei has stocked up on CPUs, even those not tested, to arrive with full warehouses at the time of the total closure of the commercial window.

The question was "what after"?

Now we know. And we know what they can do in India, Russia, and anywhere else where they have a silicon smelter and want to make computers that don't depend on US-patented ISAs.

Go ahead and bet on the USA if you like. I would be more cautious.

Leave a Reply

Your email address will not be published.