Stories
Slash Boxes
Comments

SoylentNews is people

SoylentNews is powered by your submissions, so send in your scoop. Only 16 submissions in the queue.
posted by hubie on Tuesday September 17 2024, @09:15PM   Printer-friendly

Arthur T Knackerbracket has processed the following story:

For some context, Loongson is a Chinese fabless chip company grown out of the country's state-sponsored efforts to develop domestic CPUs. Its 3A6000 chip launched in late 2023 is claimed to match AMD's Zen 3 and Intel's Tiger Lake architectures from 2020. While the company has mostly played in the CPU space until now, the GPU offerings represent a new push.

Their current model, the 9A1000, is a pretty tame GPU aimed at budget systems and low-end AI tasks. But the 9A2000 is allegedly taking things to the next level.

According to Loongson's Chairman and General Manager Hu Weiwu, the 9A2000 delivers performance up to 10 times higher than its predecessor. He claimed it should be "comparable to Nvidia RTX 2080," according to ITHome.

[...] That said, Loongson has another card to play. At the same briefing, Weiwu also provided a teaser for their next-gen 3B6600 CPU, making some lofty performance claims about its architecture. He touted "significant" changes under the hood that should elevate its single-threaded muscle to "world-leading" levels, according to another ITHome report.

Previous leaks suggest this processor will pack eight LA864 cores clocked at a stout 3GHz, along with integrated LG200 graphics.

As for the launch, Weiwu gave a tentative first half of 2025 target for initial production, with mass availability hopefully following in the second half of next year.

Loongson has typically played more of a supporting role in the CPU arena and is yet to make a dent outside of China. But if this 3B6600 chip can truly hang with the heavyweights of x86 and Arm in per-core performance, it would mark a major step up for the company.


Original Submission

Related Stories

Chinese GPUs Surpass Nvidia Chips by Nearly Tenfold in Supercomputer Simulation 16 comments

Chinese scientists have significantly improved the performance of supercomputer simulations using domestically designed GPUs, surpassing systems powered by Nvidia's advanced hardware:

Professor Nan Tongchao and his team at Hohai University achieved the performance gains through a "multi-node, multi-GPU" parallel computing approach, using Chinese CPUs and GPUs for large-scale, high-resolution simulations.

The study highlights how U.S. sanctions aimed at limiting China's access to advanced semiconductors may have inadvertently spurred innovation, leading to technological self-sufficiency and reduced reliance on foreign hardware.

Also from Interesting Engineering:

The stakes are particularly high in fields that depend on extensive computational resources. Scientists frequently rely on large-scale, high-resolution simulations for real-world applications such as flood defense planning and urban waterlogging analysis.

These simulations require significant processing power and time, often limiting their broader application. For Chinese researchers, the challenge is compounded by the fact that the production of advanced GPUs like Nvidia's A100 and H100 is dominated by foreign manufacturers and the export restrictions imposed by the US.

Also at South China Morning Post.

Previously:


Original Submission

This discussion was created by hubie (1068) for logged-in users only, but now has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 4, Interesting) by corey on Tuesday September 17 2024, @10:51PM (8 children)

    by corey (2202) on Tuesday September 17 2024, @10:51PM (#1373083)

    It’s always interesting to wonder how the Chinese manage to catch up to western competitors so quickly. Took western companies decades to get to this point.

    • (Score: 5, Insightful) by vux984 on Tuesday September 17 2024, @11:27PM (1 child)

      by vux984 (5045) on Tuesday September 17 2024, @11:27PM (#1373089)

      Even without outright IP theft, simply knowing what someone did and roughly how they did it can save you a LOT of time. And then of course, with China especially, you can't really rule out outright IP theft too.

      • (Score: 5, Funny) by bzipitidoo on Wednesday September 18 2024, @01:17AM

        by bzipitidoo (4388) on Wednesday September 18 2024, @01:17AM (#1373122) Journal

        In the 19th century, the US committed massive theft of British IP, improving much of it. Railroading was a big area for that. The Brits complained, to no avail. In the 20th century, the US whined about Soviet "theft" of ideas. Now it's whining about Chinese "theft" of ideas, as if the rest of the world is too stupid to build on ideas and not just copy.

        I wonder if these Chinese chips will have backdoors much like Intel's Management Engine. If they don't, China could stake out moral as well as technological high ground.

        Face it, this intense drive to make abundance into scarcity, treat the immaterial the same as the material, is a dumb idea.

    • (Score: 5, Insightful) by higuita on Tuesday September 17 2024, @11:32PM

      by higuita (2465) on Tuesday September 17 2024, @11:32PM (#1373092)

      for one expert, isn't that hard to analyze other people hardware and take ideas (or copy!), specially if you don't care about copyright and patents... and even those, with money or good pressure from the Chinese government may be solved and still cost lot less than innovative development. After all, copying is always cheaper, that was one of the premises of patents, protect the first one that had the big research and development cost to repay and profit from it. This would promote R&D for new products as the profits after could be big.(of course patents were quickly abused and specially software patents are a total nonsense, even worse for that long).

      They may lack experience and fall in errors that the older players already learned, but they can also avoid carrying old baggage that other keep for historic reasons and incremental improvements approach.

      But that is not all, US dropped to a cycle of profit above all and stopped innovating and doing R&D as in the past, Asian students are great and smart too, many learn in western universities and companies, and sooner or later carry the knowledge back to their countries.

      China is specially promoting R&D and new products and having most of the world manufacturing help promote new products and having people full of money that in turn are pushed to grow bigger, not just being rich, as in the many in US, Elon Musk is one of few that do use most of his money to create new things... what the leaders/owners of the other big tech are using their money on?! In china, even being right doesn't mean you don't have to obey, just look to Jack Ma, alibaba owner)

      Anyway, while many of the rapid grow of China is copy and undercut western companies, western companies also drop most of the R&D over profits and outsourced their production to China, further improving their capabilities to learn and improve other people design.

      Their weakness is the software, that have the chicken and egg problem, you need hardware to do the software, you need software to use and test the hardware... and while they have still many software problems, linux did helped a lot due to the open nature, allowing cheap and easy development of software components working on a good and working system. Some will do a good job, other the bare minimum to make it work... the first ones will grow with time, the later give the bad reputation to china quality, but with time they will slowly die. If the western don't stop the profit cycle, China and all Asia will take most of the R&D and innovation in existent (manufactured based) market

    • (Score: 2, Insightful) by Anonymous Coward on Wednesday September 18 2024, @12:59AM

      by Anonymous Coward on Wednesday September 18 2024, @12:59AM (#1373114)

      But have they got the whole picture? These days it's not just performance, it's also power consumption. Looked through tfa and nothing on the power draw of these Chinese designs. If they are as fast and draw 2x the power of existing designs, then why bother?

    • (Score: 4, Interesting) by mattTheOne on Wednesday September 18 2024, @02:52AM

      by mattTheOne (1788) on Wednesday September 18 2024, @02:52AM (#1373142)

      Well the 10xx series is like 7-8 years old now. If they used that as the baseline and just made it faster with process node tech it could be equivalent to a 20xx. It'd be more interesting if it actually had features like raytracing or physics.

      The node improvements are more impressive/interesting in China

    • (Score: 4, Interesting) by driverless on Wednesday September 18 2024, @03:00AM

      by driverless (4770) on Wednesday September 18 2024, @03:00AM (#1373146)

      It’s always interesting to wonder how the Chinese manage to catch up to western competitors so quickly.

      Oh, that's easy, when you control the media you can make your tech do anything you want. In particular Loongson have a long history of promising world-beating results that fizzled and were quietly buried after a year or two, only to be replaced by the next world-beating result.

    • (Score: 2) by Ingar on Wednesday September 18 2024, @09:12AM

      by Ingar (801) on Wednesday September 18 2024, @09:12AM (#1373179) Homepage Journal

      The 2080, wasn't that that card that was on par with the 1080, but with RT?

      The 2xxx series was a joke.

      --
      Love is a three-edged sword: heart, mind, and reality.
    • (Score: 3, Insightful) by gnuman on Wednesday September 18 2024, @09:21PM

      by gnuman (5013) on Wednesday September 18 2024, @09:21PM (#1373288)

      It’s always interesting to wonder how the Chinese manage to catch up to western competitors so quickly. Took western companies decades to get to this point.

      It took hundreds of years from Newton to Einstein. Now you can just learn both in a few semesters.

      It's not a big deal to design a chip that already existed. It's more trouble to print it and make it work. And then there's the software.

      Secondly, if you are catching up, you tend to be always catching up unless the other side stops innovating. So Chinese are not "catching up" anymore than the "western competitors" are falling behind. The race is not static.

      Finally, seeing is believing. Loongson has overpromised and not delivered in the past.

  • (Score: 2) by captain normal on Wednesday September 18 2024, @03:10AM (2 children)

    by captain normal (2205) on Wednesday September 18 2024, @03:10AM (#1373148)

    I have to admit that I'm stuck on that phrase. How can they produce chips without some type of fabrication facility?

    --
    The Musk/Trump interview appears to have been hacked, but not a DDOS hack...more like A Distributed Denial of Reality.
    • (Score: 2, Interesting) by Anonymous Coward on Wednesday September 18 2024, @03:29AM

      by Anonymous Coward on Wednesday September 18 2024, @03:29AM (#1373156)

      They are most likely being fabbed by SMIC. Most chip design firms are fabless, even Intel is about to spin off as well.

    • (Score: 0) by Anonymous Coward on Monday September 23 2024, @07:19AM

      by Anonymous Coward on Monday September 23 2024, @07:19AM (#1374118)

      I have to admit that I'm stuck on that phrase. How can they produce chips without some type of fabrication facility?

      A "fabless" semiconductor company designs chips in house but outsources the fabrication.

      This is in contrast with the more traditional structure where a company does both the design and fabrication in house, think of chips from companies like NXP, Analog Devices, Texas Instruments, Fujitsu, Sony, etc.

  • (Score: 2) by jon3k on Wednesday September 18 2024, @09:30PM

    by jon3k (3718) on Wednesday September 18 2024, @09:30PM (#1373289)

    It won't. The only way this is possible would be with incredibly expensive GPU with insane power draws. I will call it now that it won't compete either in terms of absolute performance, power:watt or cost:performance.

    But for China, that might be good enough. In the short term they can just spend endless money on producing these and boiling the ocean to power them to try and be anywhere near competitive to modern datacenter GPU from NVIDIA and even AMD.

(1)