News China's Loongson Unveils 32-Core CPU, Reportedly 4X Faster Than Arm Chip

[Admin]

Loongson launches the 3D5000, a domestic server processor with 32 cores and support for eight DDR4 memory channels.

China's Loongson Unveils 32-Core CPU, Reportedly 4X Faster Than Arm Chip : Read more

 

[DaveLTX]

Despite loongson claiming LSA is developed on their own, it's MIPS with their own AVX like extensions

So yes it's modified mips

 

[bit_user]

Thanks for the continuing coverage. I think it's worth keeping tabs on China's tech industry.

Loongson claims its 32-core domestic chips deliver 4X higher performance than rival Arm processors.

That's immediately quite suspect, so I plugged in the MyDrivers link to Google Translate and here's what it claims they said:

​

"In terms of performance, the SPEC 2006 score of Loongson 3D5000 exceeds 425. The floating point part adopts dual 256bit vector units, and the double precision floating point performance can reach 1TFLOPS (1 trillion times), which is 4 times of the typical ARM core performance."​

(the bold is theirs)​

​

That's a far more nuanced statement. For one thing, the typical ARM core has dual 128-bit vectors, which gives it an automatic 2x. I don't know where the other 2x comes from, but it's not hard to imagine theirs has dual-FMA units, whereas their basis for comparison doesn't. That still doesn't get us quite to 4x, but now we're in the ballpark.

It's way off the mark to generalize 4x the vector fp64 throughput to an overall 4x performance increase, however.

According to Loongson's provided numbers, the 3D5000 scores over 425 points in SPEC CPU 2006

That's the claimed single-CPU score. The article also mentions scores of 800 and 1500, for dual- and quad- CPU configurations.

For comparison, the last set of SPEC2006 scores I could find for an AMD CPU are the Zen 2-based Rome Epyc:

https://www.anandtech.com/show/14694/amd-rome-epyc-2nd-gen/10​

I think the score they're quoting is a simple average of the sub-scores, in which case the 128-core dual-CPU Rome Epyc delivered 3434. That's about 2.3x what they claimed to achieve with the same core-count.

Meanwhile, the processor's stream performance with eight channels of DDR4-3200 memory crosses the 50GB mark.

This probably refers to the OpenMP-based Stream Triad benchmark, which adds 2 streams of numbers and writes out a 3rd. That means actual memory traffic is 3-4x whatever number they're quoting, so 150 to 200 GB/s. That aligns well with a raw bandwidth of about 205 GB/s from 8x DDR4-3200. In multi-CPU configurations, it particularly stresses the inter-processor link. Presumably, the number their quoting is just from a single-CPU config.

It would be great to know what process node these chips were designed for.

 

[bit_user]

Despite loongson claiming LSA is developed on their own, it's MIPS with their own AVX like extensions

It's certainly more than that, but I'm not an authority on the subject. My best understanding is that they borrowed much of the MIPS system architecture, while the instruction set architecture is substantially different.

If you have a good source on the matter, please share it.

 

[The Historical Fidelity]

It would be great to know what process node these chips were designed for.

I suspect it is SMIC 14nm since the 3A5000 was originally designed to TSMC 14nm design rules and this new cpu is just 2 3A5000’s glued together. Since SMIC 14nm is an unsanctioned copy of TSMC 14nm and loongson no longer has access to TSMC services, the cpu design would be compatible with SMIC’s node so this seems the most likely assumption.

 

[shady28]

I suspect it is SMIC 14nm since the 3A5000 was originally designed to TSMC 14nm design rules and this new cpu is just 2 3A5000’s glued together. Since SMIC 14nm is an unsanctioned copy of TSMC 14nm and loongson no longer has access to TSMC services, the cpu design would be compatible with SMIC’s node so this seems the most likely assumption.

Pretty sure that TSMC never had a 14nm. They went from 16->12->10 (Apple only) -> 7

Worth noting that Intel 14nm is about 10-12% higher density than TSMC 12FFC and about 20% higher density than SMIC 14nm.

This is basically on a SMIC node that is just a smidge better than the old 16nm node that TSMC was using in 2013.

 

[anonymousdude]

Pretty sure that TSMC never had a 14nm. They went from 16->12->10 (Apple only) -> 7

Worth noting that Intel 14nm is about 10-12% higher density than TSMC 12FFC and about 20% higher density than SMIC 14nm.

This is basically on a SMIC node that is just a smidge better than the old 16nm node that TSMC was using in 2013.

You are correct. TSMC never did formally name a node 14nm. Their 16nm and 12nm constituted the "14nm-class" node.

 

[DaveLTX]

It's certainly more than that, but I'm not an authority on the subject. My best understanding is that they borrowed much of the MIPS system architecture, while the instruction set architecture is substantially different.

If you have a good source on the matter, please share it.

The programming manual mentioned by chips and cheese shows the lineage of the LSA clear as day and night, that and they were using MIPS recently so that gives a sign it's based off MIPS with their own additions

 

[The Historical Fidelity]

Pretty sure that TSMC never had a 14nm. They went from 16->12->10 (Apple only) -> 7

Worth noting that Intel 14nm is about 10-12% higher density than TSMC 12FFC and about 20% higher density than SMIC 14nm.

This is basically on a SMIC node that is just a smidge better than the old 16nm node that TSMC was using in 2013.

Good catch, I meant TSMC 12nm

 

[thisisaname]

It is easy to make claims and I do not think any third party benchmarks are going to be run any time soon.

 

[DaveLTX]

It is easy to make claims and I do not think any third party benchmarks are going to be run any time soon.

Loongson’s 3A5000: China’s Best Shot?

We at Chips and Cheese have covered two Chinese CPU architectures: Zhaoxin’s x86 compatible Lujiazui architecture found in the KX-6000 series of CPUs and Phytium’s ARM compatible FT663 architecture found in the D2000 series of CPUs. Zhaoxin was founded in 2013 as a joint venture between VIA and...

chipsandcheese.com

Except there is?

Previewing China’s Loongson 3A5000 with Performance Counters

Loongson’s 3A5000 represents another domestic CPU effort from China.

chipsandcheese.com

 

[bit_user]

I suspect it is SMIC 14nm since the 3A5000 was originally designed to TSMC 14nm design rules and this new cpu is just 2 3A5000’s glued together.

One article I saw about the 3A5000 showed their roadmap, with its successor being slated to use 12 nm. That makes me wonder if this current CPU is made on a 16nm-class node.

 

[bit_user]

The programming manual mentioned by chips and cheese shows the lineage of the LSA clear as day and night, that and they were using MIPS recently so that gives a sign it's based off MIPS with their own additions

The way the actually characterized it was:

​

"Loongson used to make CPUs based off the MIPS ISA, but the company recently switched to a homegrown ISA called Loongarch. This “new” ISA retains many of MIPS’s semantics, but uses incompatible encodings. Loongarch also gets extended to better support Loongson’s goals of making a viable domestic Chinese CPU. "​

​

The conclusion starts:

​

"By using incompatible encodings, Loongson can say they have a new ISA and develop it independently from MIPS. Calling it Loongarch rather than MIPS means they don’t have to deal with rights for the ISA, even if Loongarch and MIPS share a lot of semantics to the point where you can use MIPS64 manuals."​

​

It's interesting they say "many" of MIPS' semantics, which implies points of deviation. Unfortunately, I don't see an itemized list of the differences, but it indeed sound like they basically started out by copying MIPS.

Both quotes are from this article: https://chipsandcheese.com/2023/02/26/loongsons-lsx-and-lasx-vector-extensions/

It is easy to make claims and I do not think any third party benchmarks are going to be run any time soon.

Just today, they published a full analysis of the 3A5000.

https://chipsandcheese.com/2023/04/09/loongsons-3a5000-chinas-best-shot/​

Back in January, they published yet another article on it (that @DaveLTX also linked). I haven't used Patreon before, but I'm thinking of signing up just to give them a few $ per month. So far, theirs is the best in-depth CPU & GPU analysis I've seen on the web.

 

[thisisaname]

Loongson’s 3A5000: China’s Best Shot?

We at Chips and Cheese have covered two Chinese CPU architectures: Zhaoxin’s x86 compatible Lujiazui architecture found in the KX-6000 series of CPUs and Phytium’s ARM compatible FT663 architecture found in the D2000 series of CPUs. Zhaoxin was founded in 2013 as a joint venture between VIA and...

chipsandcheese.com

Except there is?

Previewing China’s Loongson 3A5000 with Performance Counters

Loongson’s 3A5000 represents another domestic CPU effort from China.

chipsandcheese.com

Yes and they do not look that good.

Even though Loongson has gotten their cores up from 1 GHz to 2.5 GHz, no one runs desktop or even laptop CPUs at clocks that low. Because of its massive clock speed deficiency, Loongson can’t even get in to the same performance ballpark as recent desktop CPUs. It even struggles against Neoverse N1 running at 3 GHz.

 

[Unlessround_Isfunny]

This is a great, informative post. Thank you for it.

 

[The Historical Fidelity]

One article I saw about the 3A5000 showed their roadmap, with its successor being slated to use 12 nm. That makes me wonder if this current CPU is made on a 16nm-class node.

It could be, but I’ve seen other articles state 3A & 3C 5000 chips are made on 12nm. I believe TSMC design rules are similar between the 2 nodes so can be easily ported to either.
https://www.tomshardware.com/news/loongson-3a5000-3c-5000-to-launch-in-2021

 

[drajitsh]

I wonder why they have 5 HT links. To have a glueless 4P system they need 3 links while for a 8P system they need 7 (n-1).

 

[tomscomments]

Despite loongson claiming LSA is developed on their own, it's MIPS with their own AVX like extensions

So yes it's modified mips

not acording to phoronix and most technical sites, since loongArch is a new architecture
LoongISA is MIPS, LoongArch is not.

 

[bit_user]

not acording to phoronix and most technical sites, since loongArch is a new architecture
LoongISA is MIPS, LoongArch is not.

There was a big debate in the Linux kernel community about Loongarch being treated as its own architecture, precisely because they copy-and-pasted so much code from the MIPS tree. What I think this highlights is that a lot was borrowed from MIPS, though probably more system-level architecture (which is what the kernel is principally concerned with) than ISA.

Basically, the claim of Loongarch borrowing from MIPS isn't without merit, even if you can make the case that quite a lot of it is different.

 

[tomscomments]

There was a big debate in the Linux kernel community about Loongarch being treated as its own architecture, precisely because they copy-and-pasted so much code from the MIPS tree. What I think this highlights is that a lot was borrowed from MIPS, though probably more system-level architecture (which is what the kernel is principally concerned with) than ISA.

Basically, the claim of Loongarch borrowing from MIPS isn't without merit, even if you can make the case that quite a lot of it is different.

Before LoongArch, loongson bought MIPS (around 2000; so more than two decades).
They became the most important developpers for MIPS.
Because they didn't have access to all patents, they maintained MIPS developpement who became later loongISA. LoongISA est MIPS enhanced. Loongarch is not based on MIPS microarchitecture

LoongArch have ;
-SIMD and binary translation
-Hig performance microarchitecture
-Mips Apps don't natively run on LoongArch, they need to be emulated mike arm or x86
-Loongarch and MIPS 64 are not compatible if running under linux.

Did LoongArch used somme open source mips (whom themselves worked on or riscV open sources codes ? highly probable
Does it make it a MIPS chip ? cerainely note.

According to Gentoo ! " LoongArch is not binary compatible with either MIPS or RISC-V, although the ISA and ABI show heavy influence of the two."
according to other major open source specialists, LoongArch is a new ISA.

1. Introduction to LoongArch — The Linux Kernel documentation

www.kernel.org

What loongson do is no different from other majaor compagnies
intel 8008 derived from datapoint 2200 instruction set and later became X86

That's just the story of software or hardware, starting from a base doesn't mean you are the same
Chromim emerged from blink and blink from khtm
ChromeOS is built upon Geentoo linux

That's always been the game between tech compagnies
MacOS was built on nextstep. and its mach kernel nd they later wen away from it.
Apple still use XNU kernel (derived from mach)
MacOS is still based on darwin and darwin is a mix of BSD, mach Nexstep and other opensource libraries
And you what ? HarmonyOS is in the same path. Adding proprietary technologies on existing ones.

China do nothing different from other major tech compagnies
The humanity didn't feel the need to reinvente the wheel invented by persians. But all the wheels used in the world since 5,500 years ago are a derivative.
We still use a derivated ink invented by egyptian and chinese
And we still use a derivative of chinese printing concept.

In science, you don't have to constantly change concepts. You can invent yours based on others. This is called the history of humanity

 

[bit_user]

According to Gentoo ! " LoongArch is not binary compatible with either MIPS or RISC-V, although the ISA and ABI show heavy influence of the two."

I think that's basically what I'm saying.

I don't like to repeat myself, but since it's been 8 months, let me cite the same analyst I did in post #13:

"Loongson used to make CPUs based off the MIPS ISA, but the company recently switched to a homegrown ISA called Loongarch. This “new” ISA retains many of MIPS’s semantics, but uses incompatible encodings. Loongarch also gets extended to better support Loongson’s goals of making a viable domestic Chinese CPU. "
​

The conclusion starts:

"By using incompatible encodings, Loongson can say they have a new ISA and develop it independently from MIPS. Calling it Loongarch rather than MIPS means they don’t have to deal with rights for the ISA, even if Loongarch and MIPS share a lot of semantics to the point where you can use MIPS64 manuals."
​

It's interesting they say "many" of MIPS' semantics, which implies points of deviation. Unfortunately, I don't see an itemized list of the differences, but it indeed sound like they basically started out by copying MIPS.

Both quotes are from this article:

Loongson’s LSX and LASX Vector Extensions

Loongson used to make CPUs based off the MIPS ISA, but the company recently switched to a homegrown ISA called Loongarch.

chipsandcheese.com

 

[tomscomments]

I think that's basically what I'm saying.

I don't like to repeat myself, but since it's been 8 months, let me cite the same analyst I did in post #13:

​

"Loongson used to make CPUs based off the MIPS ISA, but the company recently switched to a homegrown ISA called Loongarch. This “new” ISA retains many of MIPS’s semantics, but uses incompatible encodings. Loongarch also gets extended to better support Loongson’s goals of making a viable domestic Chinese CPU. "​

​

The conclusion starts:

​

"By using incompatible encodings, Loongson can say they have a new ISA and develop it independently from MIPS. Calling it Loongarch rather than MIPS means they don’t have to deal with rights for the ISA, even if Loongarch and MIPS share a lot of semantics to the point where you can use MIPS64 manuals."​

​

It's interesting they say "many" of MIPS' semantics, which implies points of deviation. Unfortunately, I don't see an itemized list of the differences, but it indeed sound like they basically started out by copying MIPS.

Both quotes are from this article:

​

Loongson’s LSX and LASX Vector Extensions

Loongson used to make CPUs based off the MIPS ISA, but the company recently switched to a homegrown ISA called Loongarch.

chipsandcheese.com

​

yes you are right
you can read my previous post about it. It is not specific to loongarch. All compagnies do the same and it's not something bad or stealing.
Before China rise as a technological power, we constantly read patent dispute between aerican or sometimes european or japanse tech compagnies
In fact patents, when not used wisely, become a weapon to discredit others invention. It's part of economical war wheater between american compagnies, between chinese compagnies or between chinese and us compagnies

Chineese usually fight against each other for patents
https://technode.com/2023/03/02/hua...eged-infringement-of-four-registered-patents/ (huawei sued xiaomi)
Like their us contrepart
And just like their us contrepart, that end with a nice deal 😎


Conrad Suze was the first man (known) to produce what would be the modern computer later
He patended what would be the von newman architecture wich was later used by Turing and its design are still used in 2023 machines.



https://www.cnbc.com/2017/01/20/apple-sues-qualcomm-for-1-billion.html (apple sues qualcomm)
oracle sued Google

in the tech world everyone sues everyone and that's why lawyer is very lucrative job
I just want to insist it is not a specifi chinese thing. Weaponizin patents has long history.
Britts used it against the states after the independance and we had to steal it from them (textile machinery)

Patents are something recent in human civilisation?
The most impressive civilisation in terms of scientific inventions is egypt

They even invented the binarty system that would be a basic for computing know

Ancient Egyptians: The Original Computer Scientists?

Renowned by the Greeks as masters of geometry, very little remains of complex Egyptian mathematics. Used to build structures, manage the…

medium.com

And we owe to muslims algorithm (the name of its inventor).

So once again, nothing new here.
i do believe we need to protect IP. But not excessively weaponizing it. Science is about transmission, it's always been the case
we'll end up producing more lawyers than stem graduates with this mentality