China's strategic bet on open-source RISC-V chips | Hacker News Reader

China's strategic bet on open-source RISC-V chips

btimesonline.com

Hmm... Is China betting on RISC-V chips? China is too big to rely solely on a single chip architecture. I saw another attempt to develop a new chip architecture in China, called Loongson64. In recent years, I've seen a lot of contributions to libraries/software, such as the Linux Kernel, Chromium, LLVM, V8, and ffmpeg, by Loongson, a Chinese chip company that produces an architecture called Loongson64. Apparently, China wants to develop its own chip architecture, chip design, and chip manufacturing capabilities, and RISC-V is one of the options, but there isn't a single path to achieving that.

> China is too big to rely solely on a single chip architecture.

This doesn’t really make sense. Prior to the current ARM revolution, the entire world was on x86.

> I saw another attempt to develop a new chip architecture in China, called Loongson64

Without using a common chip arch, you go at it alone. That is a competitive disadvantage.

> Prior to the current ARM revolution, the entire world was on x86.

Hmm, that's kinda false.

Let's go back ~20 years. Workstation and datacenters were on Sparc or PowerPC. ARM was in every phone, MP3 player, digital camera or any other portable device. Gaming consoles and set top boxes and the like were on MIPS or the new Cell processor which was also PowerPC based.

The world definitely was never fully on X86, and ARM was always among us.

>Hmm, that's kinda false.

> The world definitely was never fully on X86, and ARM was always among us.

The "entire world" there wasn't meant to be absolutely literal. The point I refuting was that because of China's size, they cannot focus on a single architecture. That makes no sense. China overwhelmingly utilizes x86 now.

> Let's go back ~20 years. Workstation and datacenters were on Sparc or PowerPC. ARM was in every phone, MP3 player, digital camera or any other portable device. Gaming consoles and set top boxes and the like were on MIPS or the new Cell processor which was also PowerPC based.

That's all great and good, and while I never discounted a heterogenous world, this is both Intel and AMD's marketshare: https://www.statista.com/statistics/735904/worldwide-x86-int...

The world is on x86.

Not every processor is a full fledged PC or server. Don't forget about the various microcontrollers and other smaller processors across all other product categories, like components for cars, hvac systems, IoT devices and so much more.

Microcontrollers are totally other beasts, and for current China semiconductors level, I think they could use just one simplest case, like 8048 and other cases something like simplified x86, like Intel Atom (ARM have simplified MICRO-architecture, and Risc-V also have).

So, architecture could be one, but few MICRO-architectures, like IBM 7xx series was extremely wide (could talk about 360 series if you want), but mostly hardware differences hidden under software abstractions.

You do realise that Chinese company SOPHGO is selling a 64 bit 2.0 GHz RISC-V chip with 64 OoO cores with dual 128 bit vector pipelines (per core) and 64 MB L3 cache? Chinese company Milk-V is selling these in workstations with 128 GB RAM. Customer deliveries started last month and I know several people whose machines have arrived.

Greetings from Ukraine, country at war now, and one of main foundation whale of USSR at not too far past.

I'm not too old, but I born in USSR and seen from inside, how it steal western technologies and where it stuck.

We now chosen democracy way, mostly because Russians are Nazi and think Ukrainians not nation of their level, but my youth spent in hot speech battles, on how we will deliver own totally independent semiconductor industry.

Systems, you mentioned, are from my view, something in best case 2017 year desktop technology on steroids for West - not bad, but already outdated, and far from professional systems. Modern professional systems, should be something like latest IBM Power9 or Power10, somewhere near 3.5..4.0GHz cores, Terabytes RAM and hundreds megabytes of L3 cache.

And also very important, I have not seen anything comparable to IBM or even to Microsoft in China. Fortunately I also don't seen much enough from Russia.

So, what I release, this is really serious threat, they could kill many people, but it is not parity with West even within one industry.

And yes, you could say, I'm too subjective, but I think, for West now is more important to save Ukraine as independent country (sure, same I consider about Taiwan and Israel, which are also under serious threat), to show how West values allies, not to win race by total domination for all cost.

Привіт, Слава Україні!

I'm from New Zealand (and live there again now), but I've traveled in Ukraine a little, including Dnipropetrovsk (as it was still called at the time in 2014), previously I believe the largest "closed city" in the USSR and the place a lot of aerospace technology was made. The other previously closed city I've been to is Chernogolovka, not far (20km) from "Starry City". My ex-gf (1996-2004) grew up in Chernogolovka before moving to NZ in 1994, her parents were chemical engineers there. Now the town is perhaps best known for a line of soft drinks.

I worked on compilers & JITs for Samsung R&D in Moscow from July 2014 to March 2018, living there from April 2015. I visited old friends/colleagues there at New Year 2020.

I don't intend to visit Russia again any time soon, for obvious reasons, but I'm thinking about spending some time in Odesa (and spending my remote-working salary there) as soon as it is practical.

I've been following democracy in Ukraine, and the fight against the left-over USSR culture of corruption, for a long time. Here's a video I captured (at home in NZ) from a webcam in Ukraine showing my friend Natalia voting in Kryvyi Rih (where I spent a week in 2014) in the October 2012 elections.

https://www.youtube.com/watch?v=1tbGrcqHKFY

Ok, that's a very long diversion from the topic. Just want to show I didn't suddenly become an expert on Ukraine or Russia or UA/RU relations on February 24 2022 )))

> Systems, you mentioned, are from my view, something in best case 2017 year desktop technology on steroids for West

The SG2042 is very much like the Graviton 1 that went into production (A1 instances) at AWS in November 2018, just over five years ago. The biggest difference is the SG2042 has 64 cores while the Graviton has 16. The C910 and A72 cores are very similar designs.

Note that the RISC-V instruction set did not yet officially exist until July 2019, and it takes around five years to design a CPU core and get it into a chip in a system, you can buy.

MUCH faster RISC-V SBCs using the SG2380 are coming out late this year, at Arm A78 level. That doesn't exist yet in Arm SBCs. The first A76 RK3588 boards were in around May 2022, and the Raspberry Pi 5 arrived at customers in bulk only around 2-3 months ago.

When the SG2380 arrives later in the year RISC-V will be less than 2.5 years behind ARM.

Sure, both are behind Intel and AMD at present. That's going to change by about 2026. There are multiple companies in the USA with Apple M* level RISC-V chips under development, several of them employing senior ex-Apple designers, and all of them with a lot of people with recent Intel and AMD experience.

Btw, I also have access to an Elbrus, which I've experimented with, so I've got some idea about them too. I also had many colleagues at Samsung who previously worked on Elbrus compilers at MCST.

    Mac-mini:~ bruce$ ssh elbrus
    brucehoult@sumireko:~$ uname -a
    Linux sumireko 5.4.0-6.9-e8c2 #1 SMP Mon Mar 6 23:32:47 MSK 2023 e2k E8C2 Elbrus-MCST GNU/Linux
    brucehoult@sumireko:~$ head /proc/cpuinfo 
    processor       : 0
    vendor_id       : Elbrus-MCST
    cpu family      : 5
    model           : 9
    model name      : E8C2
    revision        : 2
    cpu MHz         : 1200
    bogomips        : 2400.00
    
    processor       : 1
    brucehoult@sumireko:~$ date
    Ср фев  7 08:10:41 MSK 2024

I've also seen Mt Elbrus with my own eyes (but haven't climbed it!), from Pyatigorsk (and the top of Beshtau, 95km), from the top of Dombay ski lifts (60km), and from somewhere on A157 (50km)

Привіт, Слава Україні!

Героям Слава!

Well, first, I know few people in Pyatigorsk, and they said me, Beshtau is in strict translation from Tatar "five mountains" or in Russian "5 gor = Pyatigorsk" :)

I now cannot say you something optimistic about Odessa, except from, in about May, there will be warm enough, so it will be possible to work from Solar batteries and live without heating.

In Kiev, now is safe. Yes, sometimes could fall parts from sky, but rule of two concrete walls working good.

But ok, let's return to matters we talking about.

> Note that the RISC-V instruction set did not yet officially exist until July 2019 ... > vendor_id : Elbrus-MCST ... > cpu MHz : 1200

You remember me Russian anecdote, when somebody say (names are fictional) "Ivan won Big bucks in Preferans!!". Other answered: "not Preferans but Poker". Third: "and not won, but fail and now owe his house".

What all these really mean, we should very carefully look, to not compare apples with carrots.

What mean 1200 MHz in 2024? - Easy - this is low power technology, intended for military use, even not desktop.

Where could be used low power technology in 2024? It is also easy for me in Kiev - as I just sometimes hear loud bang - rockets or drones controlled by those tech shot down under Kiev, and unfortunately NOT shot down in Odessa and in Dnipro :(

BTW you may not interest in such economy things, but in around 2008 Fitch presented credit rating for Nasdaq members, and it said, AAA+ have only Microsoft, imagine! What this mean? Nothing special, Microsoft have cash to pay for all credits bodies just now, but other companies have money to pay percents for 1 year and more, but have not cash to pay body just now.

BTW revision:2 also said much. For real commercial chip it is fortune to see revision 5, but usual for market something more than 10.

This is because chip manufacture is iterative process, you spent at least 3 revisions to achieve predictable yield percentage, and then each revision will make yield percentage better and better economy.

I thought Loongsoon was just rebranded MIPS?

They've made enough modifications that it's no longer entirely compatible with MIPS. But it isn't far off, either.

Ironically, MIPS Technologies sold off all their remaining rights to the MIPS architecture a few years ago and is now trying to pivot to RISC-V...

> Hmm... Is China betting on RISC-V chips?

In a word: YES.

A ton of companies are already shipping RISC-V hardware, mostly in the microcontroller sector but rapidly moving to larger parts. (Off the top of my head: GigaDevice, Espressif, Allwinner, WCH, Bouffalo, Kendryte...) These devices are shipping in end-user products right now -- they have a lot more momentum than Loongson, who've been making their MIPS knockoffs since the mid-2000s with limited uptake.

It seemed to me like Loongson was sort of a "compliance" product. There are no doubt plenty of "we want to own the ecosystem end-to-end" forces who have cheque-writing authority within China. They'll bankroll homebrew parts even if they aren't really competitive to achieve that goal. Or it could be more based on their willingness to play a longer game-- they can take 20 years to get to competitive if they have to.

Serious question, x86 has obviously gotten us very far, but has legacy baggage, plus licensing restrictions.

Is RISC-V so much better that the future (whatever long horizon) is inevitable? Or are we likely to continue x86 forever? RISC for specialized (matmul, embedded) applications and x86 will remain CPUs?

If China injects $$ into the problem, will Intel and AMD have to follow?

Both amd and intel have hitched their wagons to ARM.

RISC-V is inevitable.

For legal reasons, not technical reasons.

This is despite having the technical chops, providing highest code density in 32 and 64 bit with highly competitive path length[0], all while remaining a much simpler ISA relative to the competition.

What actually matters is that it is an open ISA with huge traction that everybody gets a free license to. There's no need to ask for permission to use/implement it.

RISC-V is rapidly building the strongest ecosystem.

0. https://dl.acm.org/doi/pdf/10.1145/3624062.3624233

I can see a world in which the US transitions to ARM fully while everyone else transitions to RISCV.

Apple is not going to change to RISCV after spending a huge amount to move their entire lineup of products to ARM. Everyone else like Amazon and Nvidia are making ARM chips. Even Intel and AMD are rumored to be designing ARM chips. Qualcomm is also moving ARM to computers and servers.

Free world use both. But after totalitarian countries use riscv, it will cut and there are many ways to ban your adversities asset. Btw, adverse countries list are official.

> Apple is not going to change to RISCV after spending a huge amount to move their entire lineup of products to ARM

You could say the same about Apple spending a huge amount of money to move from 6502 to 68000 (which lasted 10 years), or spending a huge amount of money to move to PowerPC (which lasted 11 years), or spending a huge amount of money to move to x86 (which lasted 15 years).

RISC-V wasn't ready when Apple made their move to Arm64 in 2020 -- the base RISC-V ISA was only published in final form in July 2019 (RV64GC aka RVA20), many things needed for good desktop systems were added in November 2021 (RVA22), and Android for example is going to require the new RVA23 profile as its minimum configuration when the first devices come out around 2026.

Apple could move to RISC-V from around 2026 if they really wanted to. In an extraordinary move they've signed up with Arm until 2040 -- 16 years from now -- but who knows what will happen in that time period. Will Arm even exist? Will Apple be the only company still using the ISA?

At least with arm64 Apple is building their own cores and chips. Apple moved on from 68000, PowerPC, and x86 not because the companies ceased to exist but because they stopped making chips relevant to Apple, or the direction Apple wanted to move.

But never say it's too expensive for Apple to move ISAs. They've done it four times already, they can do it again.

They've done it for their computers only, which is just 10% of their revenue in 2024. If they do it again, they'd have to do it for the iPhone, iPad, Vision Pro, Macs, monitors that have A series chips, Apple TV, Watch, and all their smaller products.

I didn't say never. But 2040 is probably around the time they'd do it if they ever do it.

>I didn't say never. But 2040 is probably around the time they'd do it if they ever do it.

Doubtful. Once a decision is made, it would take full effect within 5 years.

And if they make the decision, they would make it when they determine that the ISA is ready (which would be soon, if not already in the past). Not a decade after.

The other products are easier, because Apple TV and Watch apps must be uploaded to the store as LLVM bitcode, which can be compiled to any ISA Apple wants, and for iOS bitcode has been the default for new projects for many years, though the user can change that.

Reply on news.ycombinator.com