New Huawei 96GB GPU
Key topics
The tech world is abuzz with Huawei's new 96GB GPU, sparking debate about its potential for gaming and AI applications. While some commenters, like commandersaki, wonder if it could handle gaming, others, such as smallmancontrov and jsheard, point out that its architecture is likely optimized for AI, not graphics, making it a poor fit for gaming. The discussion also touches on the limitations of Huawei's GPU due to its incompatibility with CUDA, with OsrsNeedsf2P arguing it may hinder China's AI progress, although aswanson and chickenzzzzu suggest alternative abstraction layers like Vulkan could bridge this gap. As the AI landscape continues to evolve, this thread shines a light on the complex interplay between hardware, software, and industry standards.
Snapshot generated from the HN discussion
Discussion Activity
Very active discussionFirst comment
38m
Peak period
66
0-6h
Avg / period
11.6
Based on 81 loaded comments
Key moments
- 01Story posted
Aug 30, 2025 at 6:37 PM EDT
4 months ago
Step 01 - 02First comment
Aug 30, 2025 at 7:15 PM EDT
38m after posting
Step 02 - 03Peak activity
66 comments in 0-6h
Hottest window of the conversation
Step 03 - 04Latest activity
Sep 2, 2025 at 11:42 AM EDT
4 months ago
Step 04
Generating AI Summary...
Analyzing up to 500 comments to identify key contributors and discussion patterns
Want the full context?
Jump to the original sources
Read the primary article or dive into the live Hacker News thread when you're ready.
I guess you could try to make a GPU with just programmable compute but Intel attempted that and it didn't go very well.
Even useful models like gemma3 27b are hitting 22 t/s on 4bit quants.
You aren't going to be reformatting gigabytes of PDFs or anything, but for a lot of common use cases, those speeds are fine.
https://www.hardware-corner.net/huawei-atlas-300i-duo-96gb-l...
Memory bandwidth for the M4 Max is 410 GB/s with the 32-core GPU or 546 GB/s with the 40-core GPU. It’s 819 GB/s in the M3 Ultra config.
But don't ignore this at all. This industry can change in 1 or 2 years very quickly.
It wouldn’t have a market if it wasn’t for the tensions between the USA and China, unless it’s super cheap.
Support for ECC
@150Watts
It is super cheap, compared to what's available if you want 96GB for ML inference in a single card (ignoring the other aspects one might care about). I'm seeing it on Alibaba for 1200-1500 EUR which is like 7-8 times cheaper than I can buy a RTX Pro 6000 for locally.
I highly doubt these will overtake the modded 48gb 4090 usage in China, but still it's a clear indicator that the chip embargo has lit a fire under Chinese industry to accelerate their own offerings in the semiconductor sector. It'll be interesting to see how these evolve in the coming months/years, as I suspect their iteration cycles will be shorter than US counterparts.
I'd very much like whatever card can run LLMs with Ollama or vLLM without bankrupting me and hopefully with somewhat low power usage.
Nvidia L4 cards seem to fit the bill when it comes to the power usage and getting things done, but the costs are way out there, not functionally different from H100s (I can afford neither).
So I'd very much welcome the Intel B60 Pro cards or honestly anything I could actually buy online. Until then, I'm stuck throwing money at OpenRouter and other API providers every month.
A 24GB affordable GPU can easily power an entire house worth of AI work, from real time voice chat, image generation, simple tool calls and task running, reminders, alerts, smart home integrations, etc.
IMHO a large set of potential use cases is being held back by Nvidia's high prices.
Also worth trying to better compare on efficiency. I don't know how this shakes up, but TOps/W is another figure of merit that also matters a lot, maybe more than absolute TOps count. I don't know how good 1.86TOps/W measures up here, but knowing that this is 150W is quite time where-as a 4090 or whatever is way way more reasonable than how most cards get built.
I am not saying whether retaining an edge is good or bad or that I have a different answer if one thought it was good. Just curious what you guys think.
There's a few hurdles for China to overcome first, most notably catching up on high-end manufacturing processes, but it's naive to assume that won't happen eventually.
For consumer and prosumer gear that they can get it done is already obvious, cf. people generally having no problem with buying DJI, BambuLabs or Anker.
I would be astonished if the backroom deal wasn't "If you take Taiwan now, we'll have to stop you, if you wait until we're self sufficient, we won't interfere."
Stop looking at China through a Western lens. No one knows what China will do, so this statement is false. Considering their history and culture, they will first use all other tactics to take over the island. They said they want to do it by 2049, and they could succeed without firing a single bullet, as the commenter below noted.
> I would be astonished if the backroom deal wasn't "If you take Taiwan now, we'll have to stop you, if you wait until we're self sufficient, we won't interfere."
None of the cutting edge nodes are, or will be, produced in the US by TSMC. They are all produced in Asia, and the fabs in US will be X years behind.
I don't doubt you, and I think you'll be right for some time.
I also think it's foolish to count out Intel. They're down, but so was AMD. More pointedly, this is not Intel's first time playing "This works or we go under"
No, it is not. Do you know why?
It's an opinion, clearly. Clearly, as few claim to be prescient. Dismissing my opinion, because I am not psychic and prescient is a very, very strange thing to say and do. You may say "that won't happen, your opinion will turn out to be wrong", but you cannot say my opinion as a statement is false, unless you are claiming I am lying about my own opinion?
And really, it is exceptionally silly to say "No one knows what $x will do", because of course not it's the future. We're all employing prediction trees, when we offer opinions on future events. Saying "no one knows the future" is just plain silly in this case. What are you even trying to assert? That we should all just never use our life experiences, knowledge, to attempt to provide some idea of what may come? Absolutely absurd! All of what I've just said is also understood as part of normal discussions of the future, so please try to keep this in mind.
Because trying to invalidate opinion by saying "you can't tell the future" makes no sense
And beyond that, after you discount my opinion because you claim people cannot tell the future (eg, no one knows what China will do), you immediately provide your own rendition of "what China will do".
What?!
So presumably, what you really mean in your first paragraph is that only you may predict the future outcome of events? I suggest, and I mean this honestly, that you drop this weird tactic from future debates. You cannot invalidate opinion in this way.
Moving on, it is strange to claim I am using a "Western lens". Are you trying to claim that China is somehow a land of pure people, free of all aggression and expansionist drive? And which will engage in no warlike actions? Which will not use force when it suits them? Such a rendition of any grouping of people is truly bizarre, and it is the only possible way your statement may be read. It is also very strange for you to throw this in.
You seem to be using trigger words, and pre-packaged conceptualized methods in an attempt to invalidate things people assert or say. Throwing 'Western lens" around is an attempt at impinging my worldview, it is logical fallacy, an ad hominem attack.
Please drop these sorts of tactics. If you want to realistically refute something I am saying, just refute the specific thing. Don't use ad hominem attacks. Don't refute a method (opinions of future actions), then employ them yourself.
Back to the meat of it. Surprisingly, for China, you and I seem to agree here, for you claim that China will try to use other tactics. Not will, but try as in "first use all other tactics". No kidding, ya think? Everyone tries other tactics first. Look at how many years Russia spent trying to subvert the Ukraine, before invading.
So you're not disagreeing with me. Not one bit. Because when I say China will invade Taiwan, I know all of this. Pretty much everyone you talk to knows that China has spent decades trying to subvert and take over Taiwan via sneaky, tricky subversion of Taiwan's political system. This isn't news to anyone, they've been trying for decades and endlessly failed. They've already tried those other tactics. Forever. They've failed. Over and over again.
And no, they aren't closer than ever before. Not much has changed in this regard.
So my assertion is that all of that will fail, as it has failed for decades. And that, as I said:
"China will 100% invade Taiwan."
Back to TSMC. There is more to the world than TSMC. There are other FABs coming online. There will be more money spent. And that's the whole crux of my comment.
Because the US does not want a war with China, any more than China with the US. Yet the US absolutely, positively, will not give control of Taiwan to China ever, under any circumstances, as long as the very prosperity of the US depends upon it.
Not going to happen. Not via political means. Not via a direct attack. Not via invasion. Never, never, never.
China will never ever be allowed control of Taiwan, until the US no longer needs it.
And so yes, there is an understanding between the US and China. You and I and everyone very much should want there to be an understanding. We should all want the US to have all the fabs it needs.
Because the alternative is a lot of death and destruction.
The closest parallel is, if a country cannot feed itself, and its stomach is filled by the bread of another land? And you invade that land? You will immediately be at war.
Instantly
This is entirely the same. So you should very much hope I am correct.
They will surpass us on chips just like they surpassed us on EVs. The leading edge of chip design is very complex so it will just take more time than EVs. But it is inevitable.
Even if China could get their hands on all the NVIDIA GPUs they wanted they would still try to make their own as fast as possible.
Since you are from this domain:
1. Why will they master it? Because they dedicate their industrial strategy and hence resources to it like they did in the other technological domains and flood the market?
2. Is the only way out a strict decoupling from the Chinese market in these domains? Or would it be a strategy that involves protecting domestic industries with other levers?
I understand that advanced chip making has been done, and is an engineering problem. By generations I assume you mean cohorts.
However, one must not forget the subsidy lever China is using to distort competitive advantage on a financial level. As long as we do not level the playing field in a strategic sense, we will loose on the market long term.
In theory you could gain the knowledge to design an early 1990s CPU at the logic gate level by reading some books and doing a bit of research, on the other hand actually manufacturing such IC would take considerably more effort.
Design and manufacturing are both engineering problems. Throw enough people and money at the problem and it will get done eventually. What we're banking on in the West is by the time China catches up to where we are now we'll be on to the next thing. Always one step ahead. What I'm saying is, unless we refocus our society on STEM, those days are numbered.
Canon, Nikon, ASML all used to have competitive lithography machines.
Now it’s just TSMC and Samsung at the edge, and only ASML supplies the latest lithography machines.
China will probably catch up quickly but the pace will be nonlinear and illusory. They will hit diminishing returns just like everyone else has.
They’ve probably stolen every bit of semiconductor IP they can through economic coercion or espionage.
All they can do now is out-innovate everyone else and that will take a long time. But who knows, their pace of advancement since Mao died has been impressive.
The legendary Zeiss is producing the lithography lenses for ASML, so it looks like China is pouring lots of effort to photography lenses to bootstrap their lithography lens capabilities.
I don’t know about the other parts needed for chip fabbing but I kinda expect then to encourage and subsidize other technological fields related to it as well.
Intel 3 has been shipping since last year and is only very slightly behind TSMC N3.
TSMC is almost certainly doing far more volume on their leading node though.
The US has export restrictions on certain computing devices to certain regimes which included the Sony Playstation 2, a gaming console from the double noughts [0]. Apparently the military thought it could be used to create nasty weapons. Two decades later and nobody cares whether a PS2 is shipped to Iran. We still track FPGAs I guess, though I haven't checked what's on the ITAR/EAR list in a while.
Embargoes typically work until the embargoees(?) develop the technology to build or acquire what they need. If AI is only a strategic advantage because of hardware alone, then yes. But Deepseek kinda maybe killed that idea. China has never been the first mover. They optimize. But it looks like today, AI embargoes to China will get the US months at most.
[0] https://www.pcmag.com/news/20-years-later-how-concerns-about...
1. If you look at a world map you will quickly notice that the island of Taiwan is adjacent to the asian subcontinent and mainland China. If you had one of those physical globes with a world map on them, you would not even see the United states of America when looking at Taiwan. This fact alone should make you realize that the US has no business whatsoever interfering or warmongering about anything going on over there.
2. The reason for all this is to, by any means, try prevent any increase in "power" for other countries in an attempt to prevent their time in the sun on the global economic field from fading away.
3. The reason why this will fail is because the strategy of trying to force others to be your friends and to blackmail/strongarm them into submissive allies does not work in the long run. Anyone who has attended the social battleground of kindergarten should have seen this to be a human fact.
4. My tip to all world leaders is as follows. Try not act like hyperagressive mentally challenged warmongering pssies and everything will be fine.
/Michael Ah. Sweden
7 more comments available on Hacker News