I tend to think that the small node, high-density, high-speed chips have to do with use of AI in future warfare. The need for compute power to do training and then inference on drone crafts, such as the loyal wingman concept. It seems this is where warfare is headed given the success of current drones in Ukraine.
That there is a need for CHIPs act would appear to me that there really is a palpable fear of the potential of the Chinese military within the US defense sector. That China has been on a technology hunt (and quite successfully) for the last two decades is well known. They had students learning some pretty super tech in US universities and have been repatriating and recruiting other scientists to China via the 1000 Talents Program. That there was so much technology exchange between the countries leads me to believe that the rise of Xi and the military buildup beginning in 2012.
So while the US has the stronger military for the moment, there appears to be a push and squeeze in the South China Sea by the US to provoke confrontation. However, military advantage may favor China in 10 to 20 years, so that nonmilitary means of managing China will be required.
Elon Musk's recent intervention shutdown of Ukrainian drone communications on behalf of the Russians will lead to greater reliance on local computing. Drones are usually piloted remotely, and the US had a lot of success countering drones by interfering with their communications. Letting a drone act autonomously if it is cut off from home is likely the next step. It's like the move from laser guided munitions to fire and forget munitions. Cruise missiles followed terrain, but independent drones should be able to integrate data from multiple sources like GPS, a compass, the sun or stars, terrain, ground video and target video. I don't expect it to come together all at once but over the next five or ten years.
There are definite advantages to denser chips. There are size and power constraints on board the typical drone. As the computing on board gets more demanding, it makes sense to stay as far ahead of the curve as possible. It's hard not to have a sense of where drone tactics are heading. It also makes sense to minimize the on board chip count. It means that there are fewer things to break and fewer items to stock and repair. Connections, especially, are failure prone, so the fewer the better. Anything that can be done to simply operations and repair will pay off.
I can't read the full Ben Thompson article without yet another subscription, but I can answer your question about why four big chips aren't just as good for hairy edge military applications as one chip of the same size with four times as much stuff on it. The reason is speed. Clock frequencies on chip have gotten stuck at about 2 GHz, and may have to be held below that to keep systems immune from jamming. The signals that pass between chips are 100-1000 times slower because they must be amplified, put on big wires and then detected again, and they have much further to go.
That kinda messes up our desire to put 100 kg of high explosive into the open hatch of a T-something tank by giving missies and smart shells a navigation and vision system fast enough to make all the final corrections needed in the last seconds of their mission.
This assumes what may be a fallacy in starting with a new 5nm chips that is equal to 4x 10nm chips. Are 4x 10nm chips even needed, or can the task be accomplished with better software engineering, e.g. reducing the number of bits per variable, or better algorithms, etc.? In extremis, all the possible improvement can be wrung out and used on the 5nm chip, and possibly the chip needs all the processing power it has to run the software, but are we anywhere close to that in practice?
Not assuming anything, I'm just answering Brad's question. He didn't seem to be aware of the performance cost of going off-chip to use older technology. Ziggy's point about power consumption is also relevant, Rethinking the algorithms and design could of course make this an entirely new problem, but that happens slowly with military spec technology, unlike in the consumer space.
With respect to the 1->4 chip thing note that chips have both transistors and wires connecting those transistors. A high density chip will have huge quantities of both. If you split that into 4 less dense chips you might end up with the same total number of transistors, but you've severely constrained the number of wires between transistors on different chips to a fraction of the external pins on each chip.
External pins are a scarce (and costly) resource, so to divide the problem this way you need to find 4 equal-sized chunks of logic in the original chip that don't need to talk to each other very much. If you can do that then maybe the four chips is a fine solution, but generally the applications that really benefit from the high density are the ones where no such division is possible.
¨Yet military and national-security thinkers are obsessed as the modern day equivalent of pre-World War I specialty steels and chemistry that did interesting things with the creation of the triple nitrogen bond. I really do not know enough to have an informed view of this. ¨
It´s a prestige competition. Some of these things MAY pan out as a giant game-changers, but like all potential technologies, particular computer tech, what we have here is a hype cycle.
¨His view is that the road to Chinese success in building the semiconductors of the future is to develop, organically, its own engineering community that can substitute for what ASML and Applied Materials make¨
I do, in fact, know that Hexopodia is the Key Insight. What´s missing for the Chinese is that it´s.... a country run by the Communist Party, and Communist Parties tend to stifle R&D, even if they spend lots of money on it, so China is starting out behind. (The Bolshevik cadre believed that they needed revolutions in rich countries so that technology & capital could be imported into poor communist countries. They assumed everything would then (´naturally´!) level out. This turned out to be wrong, and that failure killed the Soviet Union, even with their commitment to education, infrastructure & R&D. They were effectively pushing on a string.
China has exactly the same problem. They imported the upgrade technology from the Soviets, and then decided they had enough (to build socialism in one country?), and stagnated. So they flipped sides and started to replay the process with the West. That worked for the same reason people try and rob banks: that´s where the money is. The problem is, of course, then they have to start making a lot of more effort in terms of building R&D communities and they´ll continue to face plant trying to develop this stuff.
It´s just fundamental to the whole issue that vampires always need more blood and a ready supply of it, regardless of how self-impressed the elite cadre is with themselves at this point. Cut off, reduce or dilute that supply and they will get ill.
On our side, we have a problem with not having spent enough on R&D and creating results with that R&D. Thus we have the current situation with tanks, for example. 30 years of non-development of armored vehicles, concurrent with lots of easy wins in the anti-tank space and armor looks fairly obsolete. (It´s not, I don´t believe, but there needs to be a lot of catch-up work done.) That in turn leads to the situation with the Ukraine: neither side has enough air assets to create a situation where they can compensate for the weaknesses of the armor, and the armor isn´t good enough to do it on it´s own. (German armor was quite bad at the start of Barbarossa, but they had air superiority to start with - the Ju87 is what enabled all those giant encirclements.) So essentially the UKR/RF war has reverted back to 1917/1918.
We need lots of development of all kinds of things, but, of course, the elite right actually loves authoritarians and they simply don´t want to pay to do this. That is what´s keeping China in the game.
To revert back to the original question: I don´t see it either, and I see a lot of arguments for vapor possibilities which are not going to pan out (because the batting percentage of purported Silicon Valley game-changers is very very low).
(If there is a big chip win in view, it´s with all the quantum stuff, I´d think.)
To go back to the Hexopodia episode - I (still!) think what trips up everyone´s predictions w/ regard to Chinese development is still that it´s a Communist country. They´re in a debt trap, sure, but they could just run around and write all that down and off. Basically they could drink a big glass of salty water and poof, all their paperwork problems would go away. That wouldn´t solve any of their other problems, but it would solve that one. (That is effectively what they did after 2008 - they did 50% per year inflation, which is one way to write all that stuff down.) Debt is a problem that a country where all property is still legally property of the state could just wish away.
Doesn´t mean they will. Worldwide, authoritarians of left or right tend to share the same streak of stupidity regarding paper debt and legal tender.
Problems that can´t be fixed by having your country run by a Communist party are corruption, lack of initiative, failure of vision, an aging population, most social problems (they don´t want to admit they have them), a lack of foreign investment (if the foreigners choose not cooperate), and a lack of export markets (ditto).
elm
unfortunately, our elites tend to have differently bad head-up-ass problems
Agreed with you everything here, except your assumption that China is a communist country like the old Soviet. A few facts: The owner of Huawei had reached a deal for selling all of its business to Canada's NORTEL, and NORTEL gave up the acquisition after senior management changed. They paid big money to IBM for management and culture change after they decided to continue their business. HUAWEI currently has employees over 200,000 and about half of them are RESEARCH engineers.
Current China has lot of private enterprises like Huawai. While the government is favor the state enterprises but it doesn't limit their growth. I don't see a similar situation in the old communist Soviet.
the government doesn't limit the private business growth if the later are in compliance with the law and regulations. You may argue that the law and and regulation are made and could be changed by the communist party any time, but still we see the private businesses contributing most part of the China's economic growth in the past few decades.
Jack Ma is very hard-working and intelligent man, who wanted to and strove as hard as was humanly possible to be "in compliance with the law and regulations". And yet he failed. And nobody has given me any explanation for why he failed, other than someone in Xi Jinping's immediate orbit decided that he did not like Jack Ma.
If I were you, Alfred, I would be careful here. Politics changes and can change very quickly in autocracies, and people who have been the loudest advocates of the old party line often find themselves in difficulties...
Thanks Brad. Agree that Jack Ma is a very hard-working and intelligent man. In fact his creation of Alibaba was against the government regulations at the time and could get punished if he could not handle with the authority. Jack Ma risked his life for ideals. The government banned his Ant Group going to public in stock market since the online lending services would have very high leverage over the government allowed for a regular commercial banks.
Alibaba and its Ant group still run well and I believe that Jack Ma still is the largest owner for the business, even I'm not sure.
Government responsibility is one side, and the other side may have to playing evil.
Isn't making something more expensive to import called "infant industry protection?" :)
I tend to think that the small node, high-density, high-speed chips have to do with use of AI in future warfare. The need for compute power to do training and then inference on drone crafts, such as the loyal wingman concept. It seems this is where warfare is headed given the success of current drones in Ukraine.
That there is a need for CHIPs act would appear to me that there really is a palpable fear of the potential of the Chinese military within the US defense sector. That China has been on a technology hunt (and quite successfully) for the last two decades is well known. They had students learning some pretty super tech in US universities and have been repatriating and recruiting other scientists to China via the 1000 Talents Program. That there was so much technology exchange between the countries leads me to believe that the rise of Xi and the military buildup beginning in 2012.
So while the US has the stronger military for the moment, there appears to be a push and squeeze in the South China Sea by the US to provoke confrontation. However, military advantage may favor China in 10 to 20 years, so that nonmilitary means of managing China will be required.
Elon Musk's recent intervention shutdown of Ukrainian drone communications on behalf of the Russians will lead to greater reliance on local computing. Drones are usually piloted remotely, and the US had a lot of success countering drones by interfering with their communications. Letting a drone act autonomously if it is cut off from home is likely the next step. It's like the move from laser guided munitions to fire and forget munitions. Cruise missiles followed terrain, but independent drones should be able to integrate data from multiple sources like GPS, a compass, the sun or stars, terrain, ground video and target video. I don't expect it to come together all at once but over the next five or ten years.
There are definite advantages to denser chips. There are size and power constraints on board the typical drone. As the computing on board gets more demanding, it makes sense to stay as far ahead of the curve as possible. It's hard not to have a sense of where drone tactics are heading. It also makes sense to minimize the on board chip count. It means that there are fewer things to break and fewer items to stock and repair. Connections, especially, are failure prone, so the fewer the better. Anything that can be done to simply operations and repair will pay off.
There is also power consumption. If you want a silent death sentry you can leave alone for six months, the fancier chip tech could use less power.
I can't read the full Ben Thompson article without yet another subscription, but I can answer your question about why four big chips aren't just as good for hairy edge military applications as one chip of the same size with four times as much stuff on it. The reason is speed. Clock frequencies on chip have gotten stuck at about 2 GHz, and may have to be held below that to keep systems immune from jamming. The signals that pass between chips are 100-1000 times slower because they must be amplified, put on big wires and then detected again, and they have much further to go.
That kinda messes up our desire to put 100 kg of high explosive into the open hatch of a T-something tank by giving missies and smart shells a navigation and vision system fast enough to make all the final corrections needed in the last seconds of their mission.
This assumes what may be a fallacy in starting with a new 5nm chips that is equal to 4x 10nm chips. Are 4x 10nm chips even needed, or can the task be accomplished with better software engineering, e.g. reducing the number of bits per variable, or better algorithms, etc.? In extremis, all the possible improvement can be wrung out and used on the 5nm chip, and possibly the chip needs all the processing power it has to run the software, but are we anywhere close to that in practice?
Not assuming anything, I'm just answering Brad's question. He didn't seem to be aware of the performance cost of going off-chip to use older technology. Ziggy's point about power consumption is also relevant, Rethinking the algorithms and design could of course make this an entirely new problem, but that happens slowly with military spec technology, unlike in the consumer space.
With respect to the 1->4 chip thing note that chips have both transistors and wires connecting those transistors. A high density chip will have huge quantities of both. If you split that into 4 less dense chips you might end up with the same total number of transistors, but you've severely constrained the number of wires between transistors on different chips to a fraction of the external pins on each chip.
External pins are a scarce (and costly) resource, so to divide the problem this way you need to find 4 equal-sized chunks of logic in the original chip that don't need to talk to each other very much. If you can do that then maybe the four chips is a fine solution, but generally the applications that really benefit from the high density are the ones where no such division is possible.
I guess the jury is out. It’s a Sputnik moment.
¨Yet military and national-security thinkers are obsessed as the modern day equivalent of pre-World War I specialty steels and chemistry that did interesting things with the creation of the triple nitrogen bond. I really do not know enough to have an informed view of this. ¨
It´s a prestige competition. Some of these things MAY pan out as a giant game-changers, but like all potential technologies, particular computer tech, what we have here is a hype cycle.
¨His view is that the road to Chinese success in building the semiconductors of the future is to develop, organically, its own engineering community that can substitute for what ASML and Applied Materials make¨
I do, in fact, know that Hexopodia is the Key Insight. What´s missing for the Chinese is that it´s.... a country run by the Communist Party, and Communist Parties tend to stifle R&D, even if they spend lots of money on it, so China is starting out behind. (The Bolshevik cadre believed that they needed revolutions in rich countries so that technology & capital could be imported into poor communist countries. They assumed everything would then (´naturally´!) level out. This turned out to be wrong, and that failure killed the Soviet Union, even with their commitment to education, infrastructure & R&D. They were effectively pushing on a string.
China has exactly the same problem. They imported the upgrade technology from the Soviets, and then decided they had enough (to build socialism in one country?), and stagnated. So they flipped sides and started to replay the process with the West. That worked for the same reason people try and rob banks: that´s where the money is. The problem is, of course, then they have to start making a lot of more effort in terms of building R&D communities and they´ll continue to face plant trying to develop this stuff.
It´s just fundamental to the whole issue that vampires always need more blood and a ready supply of it, regardless of how self-impressed the elite cadre is with themselves at this point. Cut off, reduce or dilute that supply and they will get ill.
On our side, we have a problem with not having spent enough on R&D and creating results with that R&D. Thus we have the current situation with tanks, for example. 30 years of non-development of armored vehicles, concurrent with lots of easy wins in the anti-tank space and armor looks fairly obsolete. (It´s not, I don´t believe, but there needs to be a lot of catch-up work done.) That in turn leads to the situation with the Ukraine: neither side has enough air assets to create a situation where they can compensate for the weaknesses of the armor, and the armor isn´t good enough to do it on it´s own. (German armor was quite bad at the start of Barbarossa, but they had air superiority to start with - the Ju87 is what enabled all those giant encirclements.) So essentially the UKR/RF war has reverted back to 1917/1918.
We need lots of development of all kinds of things, but, of course, the elite right actually loves authoritarians and they simply don´t want to pay to do this. That is what´s keeping China in the game.
To revert back to the original question: I don´t see it either, and I see a lot of arguments for vapor possibilities which are not going to pan out (because the batting percentage of purported Silicon Valley game-changers is very very low).
elm
so
(If there is a big chip win in view, it´s with all the quantum stuff, I´d think.)
To go back to the Hexopodia episode - I (still!) think what trips up everyone´s predictions w/ regard to Chinese development is still that it´s a Communist country. They´re in a debt trap, sure, but they could just run around and write all that down and off. Basically they could drink a big glass of salty water and poof, all their paperwork problems would go away. That wouldn´t solve any of their other problems, but it would solve that one. (That is effectively what they did after 2008 - they did 50% per year inflation, which is one way to write all that stuff down.) Debt is a problem that a country where all property is still legally property of the state could just wish away.
Doesn´t mean they will. Worldwide, authoritarians of left or right tend to share the same streak of stupidity regarding paper debt and legal tender.
Problems that can´t be fixed by having your country run by a Communist party are corruption, lack of initiative, failure of vision, an aging population, most social problems (they don´t want to admit they have them), a lack of foreign investment (if the foreigners choose not cooperate), and a lack of export markets (ditto).
elm
unfortunately, our elites tend to have differently bad head-up-ass problems
Agreed with you everything here, except your assumption that China is a communist country like the old Soviet. A few facts: The owner of Huawei had reached a deal for selling all of its business to Canada's NORTEL, and NORTEL gave up the acquisition after senior management changed. They paid big money to IBM for management and culture change after they decided to continue their business. HUAWEI currently has employees over 200,000 and about half of them are RESEARCH engineers.
Current China has lot of private enterprises like Huawai. While the government is favor the state enterprises but it doesn't limit their growth. I don't see a similar situation in the old communist Soviet.
the government doesn't limit the private business growth if the later are in compliance with the law and regulations. You may argue that the law and and regulation are made and could be changed by the communist party any time, but still we see the private businesses contributing most part of the China's economic growth in the past few decades.
Ah. But what are the laws and regulations?
Jack Ma is very hard-working and intelligent man, who wanted to and strove as hard as was humanly possible to be "in compliance with the law and regulations". And yet he failed. And nobody has given me any explanation for why he failed, other than someone in Xi Jinping's immediate orbit decided that he did not like Jack Ma.
If I were you, Alfred, I would be careful here. Politics changes and can change very quickly in autocracies, and people who have been the loudest advocates of the old party line often find themselves in difficulties...
Thanks Brad. Agree that Jack Ma is a very hard-working and intelligent man. In fact his creation of Alibaba was against the government regulations at the time and could get punished if he could not handle with the authority. Jack Ma risked his life for ideals. The government banned his Ant Group going to public in stock market since the online lending services would have very high leverage over the government allowed for a regular commercial banks.
Alibaba and its Ant group still run well and I believe that Jack Ma still is the largest owner for the business, even I'm not sure.
Government responsibility is one side, and the other side may have to playing evil.
As I said: be careful! Remember that Lin Biao was once Chairman Mao's principal comrade-at-arms and designated successor, until things changed...