News China's secretive Tianhe 3 supercomputer uses homegrown hybrid CPU — rivals US systems with 1.57 Exaflops of performance: Report

[Admin]

China's most powerful supercomputer uses MT-3000 processor with built-in accelerators.

China's secretive Tianhe 3 supercomputer uses homegrown hybrid CPU — rivals US systems with 1.57 Exaflops of performance: Report : Read more

 

[bit_user]

The key aspect of the MT-3000 architecture is that it packs both general-purpose and matrix acceleration cores into the same piece of silicon. To some degree, this integration mirrors the design philosophy behind AMD's Instinct MI300A CPU-GPU hybrid, suggesting a shift away from conventional discrete CPU-GPU systems towards more cohesive and efficient designs.

I'd argue there's no imitation or shift, here.

I'm immediately reminded of the Sunway SW26010, powering their TaihuLight supercomputer:

A First Peek At China’s Sunway Exascale Supercomputer

The trade war between the United States and China is not just a top-down political and economic one, but also a technical one. And one could argue that

www.nextplatform.com

I had even guessed the Tianhe-3 is using a derivative or descendant of these Sunway processors, although it looks like the MT-3000 is of a different lineage, in spite of its apparent similarities.

I would also point out that Japan's Fugaku supercomputer also opted for a pure CPU approach, utilizing Fujitsu's entirely-custom A64FX ARM processors, with 512-bit SVE pipelines, instead of GPUs.

https://www.anandtech.com/show/13258/hot-chips-2018-fujitsu-afx64-arm-core-live-blog
​

 

[gg83]

Those must be very expensive and energy hungry chips.

 

[Evildead_666]

Those must be very expensive and energy hungry chips.

45GFlops/watt amd 1.5ExaFlops total.
I cant do this on my phone right now, but its doable to find out the total wattage of the system approximately.
Then you just have to know how namy of these Mt-3000's there are in the system.

 

[P1nky]

45GFlops/watt amd 1.5ExaFlops total.
I cant do this on my phone right now

You can use your phone to ask any AI LLM asistent to calculate the answer for you. The supercomputer uses 34.6 MW.|

 

[Gururu]

Do any other countries other than U.S. and China have their own homemade supercomputers or does everyone else like Russia just use U.S. hardware?

 

[Pierce2623]

I see the new Chinese “7nm” called an advanced node and it’s pretty funny. Its performance/efficiency falls in between TSMC 10nm and 14nm. It’s also MUCH more expensive than even TSMC 7nm. Then the “5nm” is going to be 50% more expensive than TSMC 5nm yet it will perform like TSMC 10nm if they’re lucky.

 

[The Historical Fidelity]

Im gonna wait until actual verified figures come out for this supercomputer. As of right now, this articles filled with hearsay numbers and napkin math.

 

[bit_user]

Do any other countries other than U.S. and China have their own homemade supercomputers or does everyone else like Russia just use U.S. hardware?

Well, I did just link to the Fujitsu A64FX, powering Japan's Fugaku supercomputer. After the USA and maybe China, I think Japan has probably the most self-sufficient HPC industry, and it's not just Fujitsu. There's also:

• https://www.preferred.jp/en/projects/supercomputers/
• https://www.pezy.co.jp/en/products/pezy-sc3/

However, PEZY - the most interesting, with an exotic RF-coupled alternative to HBM - got hit with some massive corruption case that seems to have derailed their business, at least for quite a while. As for Preferred Networks, I'm not sure if their processors are for sale outside Japan.

I think Japan has really struggled to build a homegrown computer industry. Mainframes and HPC seem to have been the areas of greatest success, perhaps largely because they tend to be heavily-subsidized and the purchasing is more nationalistic.

Europe is attempting to achieve its own degree of self-sufficiency, here:

General Purpose Processor - European Processor Initiative

www.european-processor-initiative.eu

Specs-wise, I think it's fairly underwhelming. That doesn't mean it's pointless, however.

The desire to "buy local" might also be one of the main things keeping Tachyum still alive, who is claiming to be building a HPC-capable cloud processor:

https://www.tomshardware.com/pc-components/cpus/tachyum-prodigy-production-starts-in-2024
​

 

[bit_user]

Im gonna wait until actual verified figures come out for this supercomputer. As of right now, this articles filled with hearsay numbers and napkin math.

China has stopped submitting entries to the Top500 list. So, their developments are going to remain more in the unofficial domain.

 

[aldaia]

Do any other countries other than U.S. and China have their own homemade supercomputers or does everyone else like Russia just use U.S. hardware?

Strictly speaking no country in the world has a home made supercomputer. China is probably the one closer to that idea, although I doubt that DRAM and NAND are produced locally.

Lets take Frontier, #1 Top500 supercomputer. Its based on AMD EPYC CPU's and AMD Instinct GPUs. AMD may be a US company, but the actual chips are fabbed by TSMC in Taiwan. Not to mention the huge amount of DRAM and NAND chips required, that most probably are produced in South Korea.

If we take number #2, things get a bit better, Intel Xeon CPU is probably fabbed in the US , however, as far as I know, the Intel GPUs that power it are fabbed by TSMC, not Intel. And again don't forget the DRAM and NAND chips.

CPU/GPU may be the glamorous part of a supercomputer, but it's only the top of the iceberg. DRAM (for both CPU and GPU) and NAND for storage, represent the majority of the silicon.

 

[bit_user]

Strictly speaking no country in the world has a home made supercomputer. China is probably the one closer to that idea, although I doubt that DRAM and NAND are produced locally.

Lets take Frontier, #1 Top500 supercomputer. Its based on AMD EPYC CPU's and AMD Instinct GPUs. AMD may be a US company, but the actual chips are fabbed by TSMC in Taiwan. Not to mention the huge amount of DRAM and NAND chips required, that most probably are produced in South Korea.

Depending on how far you want to go with this, basically nothing is made in a single country. Even if we're talking about chips, wafers, substrate, photo masks, photoresist, various other chemicals, and raw materials all tend to be imported. Not to mention the fab machinery, chip layout & design software, simulation software, etc.

again don't forget the DRAM and NAND chips.

Micron is US-based and makes both of these. I don't happen to know which fabs make what, but they do have some fabs in the USA.

 

[aldaia]

Micron is US-based and makes both of these. I don't happen to know which fabs make what, but they do have some fabs in the USA.

Samsung + SK Hynix (both Korea based) are the memory top dogs and account for more than 70% of the production although Micron has a respectable 20% share. All 3 have more than 90% of the memory market.

Micron is US based but most of its fabs are based in Singapore, Taiwan, Japan and even China. As far as i know, Micron only has 2 fabs in the US: Fab 4 in Idaho is a R&D facility while FAB 6 in Virginia was built in the 90s and accounts for a tiny fraction of Micron production. Its current leading-edge chips are made in Japan and Taiwan, but Micron is aiming to bring advanced memory production to the U.S. starting in 2026.

 

[aldaia]

Europe is attempting to achieve its own degree of self-sufficiency, here:

​

General Purpose Processor - European Processor Initiative

www.european-processor-initiative.eu

I really hope this time we do better than on previous attempts to build a European Supercomputer. Unfortunately, past experience with some projects I participated/contributed tells me that won't be easy.

The first and probably the closest one to being successful was the attempt to build a supercomputer based on transputers. https://en.wikipedia.org/wiki/Transputer
In the late 80s early 90s transputers where ahead of its time. They pioneered the concept of clusters and MPPs. Other innovations where hardware based context switching 2 decades before Intel Hyperthreading and integrated communication links for message passing. Unfortunately the failure to deliver a superscalar transputer in mid 90s killed all prospects of having a competitive supercomputer.

Next one was the ACRI project. If transputers pionered MPPs at a time where supercomputing was dominated by vector supercomputers, ACRI was too late for its time. It was devised as competition for vector supercomputers, at a time where supercomputing was already moving to MPPs. In the end it didn't matter, the project was a total failure plagued with management issues. This is the story of a European supercomputer that never happened.
https://archive.gyford.com/1997/wired-uk/2.01/es/waste.html

Last and more recent is the (multiple) montblanc projects. https://www.montblanc-project.eu/
They where technically succesful, but again nothing close to a European supercomputer came out of it. In the end its main contribution was to pioneer the concept using ARM for supercomputing and creating a supercomputing software stack for the ARM ecosystem. It was the japanese with the Fujitsu A64FX and Fugaku supercomputer, who benefitted the most, they reused most of the software stack.

I see no reason why this time will be different, but I'll be happy to be proven wrong.

 

[bit_user]

The first and probably the closest one to being successful was the attempt to build a supercomputer based on transputers. https://en.wikipedia.org/wiki/Transputer
In the late 80s early 90s transputers where ahead of its time. They pioneered the concept of clusters and MPPs.

I once programmed a parallel machine that was built on a spiritual successor of the Transputer. It implemented SIMD via instruction-broadcast to each of the processors, rather than SIMD referring to multiple elements within the vector registers of a single core.

https://en.wikipedia.org/wiki/Super_Harvard_Architecture_Single-Chip_Computer
​

I see no reason why this time will be different, but I'll be happy to be proven wrong.

Well, if they can just de-politicize the staffing and management of it enough, maybe...
: D

 

[aldaia]

I once programmed a parallel machine that was built on a spiritual successor of the Transputer. It implemented SIMD via instruction-broadcast to each of the processors, rather than SIMD referring to multiple elements within the vector registers of a single core.

​

https://en.wikipedia.org/wiki/Super_Harvard_Architecture_Single-Chip_Computer​

Interesting. I happen to know the SHARC DSP because I worked in the compiler of its successor, the TigerSHARC. I did some work on the instruction scheduler.

 

[bit_user]

Interesting. I happen to know the SHARC DSP because I worked in the compiler of its successor, the TigerSHARC. I did some work on the instruction scheduler.

I think we got tired of waiting for the TigerSHARC and that's why we switched to a Philips chip. It was 5-way VLIW and really not programmable in assembly language. It had a pretty good C compiler, though.

 

[aldaia]

I think we got tired of waiting for the TigerSHARC and that's why we switched to a Philips chip. It was 5-way VLIW and really not programmable in assembly language. It had a pretty good C compiler, though.

Guess it was the Philips TriMedia.

That was the period where DSP companies switched all to more complex designs (VLIW, Clusters, etc) that where not practical to program in assembly anymore. They where like crazy looking for automatic instruction scheduling technologies.

 

[bit_user]

That was the period where DSP companies switched all to more complex designs (VLIW, Clusters, etc) that where not practical to program in assembly anymore. They where like crazy looking for automatic instruction scheduling technologies.

TI really lead the way, there. Their compilers were really aggressive at optimizing. One of the few times I hit an honest-to-goodness compiler bug was in TI's optimizing C++ compiler for their floating point DSPs. That said, I really did like how many hints you could give the compiler, to help it do unrolling vectorization, etc.

 

[Pierce2623]

Samsung + SK Hynix (both Korea based) are the memory top dogs and account for more than 70% of the production although Micron has a respectable 20% share. All 3 have more than 90% of the memory market.

Micron is US based but most of its fabs are based in Singapore, Taiwan, Japan and even China. As far as i know, Micron only has 2 fabs in the US: Fab 4 in Idaho is a R&D facility while FAB 6 in Virginia was built in the 90s and accounts for a tiny fraction of Micron production. Its current leading-edge chips are made in Japan and Taiwan, but Micron is aiming to bring advanced memory production to the U.S. starting in 2026.

Micron did announce a new flagship fab in Boise, where they’re based. It’s not completed yet, though.