That is an offtopic response, because the discussion is not about whether anyone will buy Chinese GPUs. It's about whether Chinese GPUs can be competitive for scientific computing. Whether they will ever will be or not, they are not now, which we know because they are still hungry for our hardware. And we can reasonably believe that to be true and not merely propaganda both because of sales, and effort spent redirecting purchases through other nations.

In the really real world, efficiency matters. If you need more nodes then both your up front and ongoing costs are higher as your system takes up more space and uses more power. Even if China can design a GPU as well as Nvidia (let's go ahead and assume that they can; I think they probably can soon, if not already) that won't make their performance per watt as good unless they have it fabbed by someone with superior process technology.

I'm not clear on what you're claiming here. If the article is worth anything, which I don't take for granted, what it says is that they got a speedup for a specific workload by organizing their system with more GPUs per node. That would give a speedup for their workload no matter whose GPUs they used. The article doesn't claim it has anything to do with the performance of their GPUs.

"non-paywalled, syndicated link"

*visits page*

"Please enable JS and disable any ad blocker"

You could say the same thing about Intel.

But that's not an endorsement of the idea that China will succeed, and the point isn't even to bag on Intel, although that's fun.

ASML is a multinational company. The technology needed to do modern photolitho required the efforts of multiple nations to create. Why should we believe China can do it alone when so far, no one nation can do so?

That works for battery technology, where the big limiting factor was investment activity. It doesn't work for litho processes where the level of technology is very high. China's limiting factor in this regard is process technology. Even if they figure out how the chips work, which they have decades of experience at doing (decades ago, I worked as a sysadmin for an IC design company whose designs were being copied in China) and are therefore plausibly quite good at it, they still couldn't fab them.

Can they eventually catch up to the efforts of the rest of the planet to create new process technology? Maybe, but I doubt it. The rest of the world has been sleeping on battery tech (literally the only thing China is ahead of everybody else on right now) and depending on China for development and production, which is why they are leaders there. It has absolutely nothing to do with sending our battery tech to China or not. Same for Solar, and yet Chinese companies still don't produce the most efficient solar panels because it doesn't matter. (A couple of efficiency points don't really affect their plans; they would be welcome of course but they aren't required because solar panels are cheap and they have a lot of land.)

That requires placing a device on the train, because the train is in motion, or placing a whole lot of devices. The current situation only requires one low-power device someplace vaguely near the rail line.

Think harder. You know effectively nothing about trains, so do some searches so you can know something before posting again. Start by looking up "caboose" and what functions were performed there before they were replaced by FRED.

But I had it on good authority* that China is innovative and has caught up to or even surpassed us and therefore doesn't need our chips!

* Some clown on social media who's probably a Chinese bot

That's not new, though. The whole point of ChromeOS is to deliver Chrome, and Android has a System Webview based on Chrome. They both already have the browser integrated with the OS, so now instead of two such operating systems there will be one.

That's cool, it was always a good way to store some emergency contacts in a way that would work even if your phone died and you had to replace it.

That's a good workaround, but it's not what's actually called for.

LibreOffice Calc is full of compromises like that. Look how hard it is to split up a string with a formula, you need to use a regex! That's insane. (There's a split function for scripts, but not a formula.)

I'm clear about what the complaint is.

My point is that the word choice is bullshit.

If you want real, you go see live and if not unamplified acts, at least those which aren't using so much as an effects pedal. Or, for that matter, the reverb knob on the amp. You're not listening to canned modern music where the "real band" may or may not have skills or talent.

If you're willing to listen to music which was cooked up in a lab, what's the difference between that and listening to fully computer generated music? It's not like the fans have an actual connection with any of the band members, who are just taking up some of the slots in their head which could be filled with real people they actually know. (Yeah, there's no slots, but humans can only actually maintain a developed internal representation of so many people at once, which yes varies from person to person.)

Great, then this will confuse statistically nobody.

Thanks, those are very interesting links with solid information.

wat