Nvidia definitely has a lot of skeletons in their closets with respect to anti-competitive practices.

But even beyond that, any government would be within its right to start action against their dominance in the GPU compute space (e.g. making CUDA an open, independently managed, standard that Nvidia would have to do their absolute best to comply (or Huang would have proper liability, not american style).

Same with their schemes around sanctions busting. If I was American, I would be extremely pissed off with how they are being handled with kids gloves for what is essentially treason (i.e. from my limited understanding the highest penalty in the US would be capital punishment).

TSMC would love such a setup, although I feel Apple will demand a measure of loyalty (and to my limited understanding Apple hates Nvidia).

I don’t mean to underestimate or playdown China’s potential. Being from the former (russia-occupied) USSR, I think both the west and global south severely underestimate and misunderstand the nature of China/Russia (and how to deal with them).

What I am saying is that there are also inherent weaknesses to their economic and political systems that are often completely ignored; typically because they tend to be more medium/long term in nature.

Hell, you can see it now with older chips at bigger physical nodes where China is now a significant portion of global production.

Genuinely curious if you have any data on this.

Will the PRC chip industry face many challenges? Of course it will. However, the PRC’s track record of going from nothing to 5nm in a few years cannot be ignored by TSMC.

This is one example of “western” misinterpretation/misunderstanding of a regime such as China. While one should not casually dismiss their achievements, one should also be critical about their PR statement regarding 5nm.

To my understanding their 5nm approach has yet to be delivered (show me a product with a 5nm chip) and its fundamentally unsuitable for mass scale production.

It is reasonable to evaluate the role of “5nm” as a PR move and not as a working product.

I would speculate even their “7nm” chips may be less competitively viable than one would think based on their use in Huawei’s smartphones. I could be wrong though, it’s difficult to find good information on this topic.

And what if Nvidia does front a lot of cash (similar to how Apple bankrolls new nodes)?

Not saying this is happening or will happen, but it is a distinct possibility.

The development of PRC’s chip industry is not guaranteed to go on a exponential curve.

There are inherently some disincentives to PRC’s approach to managing their chip industry, lots of money chasing limited productive projects (you can bet some opportunistic people in China are going to try and piggyback the money faucet). Essentially no-bid type contracts that are going to create a culture of stagnation (at least partially). Lack of a broad customer base (e.g. Apple essentially bankrolling new nodes at TSMC); global demand will always be way bigger and more sophisticated than China, Russia and Iran.

Then you also have technical limitations like lack of access to hardware from ASML, more limited (and less competitive) tooling and design ecosystem.

I believe this is tied to the memory bus size (which is tied to the overall architecture design).

I believe 448 bit memory bus doesn’t actually allow 32GB, I think it would have to be 56 GB.

We’ll be lucky if it’s 2K USD.

Considering how much higher margins are for enterprise “AI” use cases, they could easily price this at $2.5 - $3K and they would do fine (if it’s too much you can pay $1999 for the 5080).

Yeah, since the fabs switch from planer designs, the node naming schemes have been marketing derived.

I don’t think it’s solely an issue with 60+ year olds, although I can’t speak for Korea as I don’t know anything about the country beyond general knowledge.

I heard a theory that because it’s a relatively small industry, all the senior technical marketing execs come from the same pool, so that’s why they all have “fascinating” naming schemes.

It’s not going to be affordable, especially at retail (desktop). AMD know they have they upper hand over intel in this particular area.

This seems to be targeted for 2033.

And costs are not going to go down. We’ve hit the point were new developments in chip fabrication do not lead to lower costs (perhaps cost per unit of performance, but that’s a different discussion).

You probably can, but it doesn’t look like it will be officially supported:

https://itc.ua/en/news/amd-processors-with-zen-5-architecture-will-not-support-windows-10-lenovo-manager/

The Nvidia 5000 series pricing will make the 4000 series look like charity. You know it yourself!

Quite a few games I play (lot’s of indie economic strategy/builder type games) would probably benefit from the extra cache, but I also don’t feel like upgrading my 5800X to 5800X3D.

My bigger issue is the lack of support for Win10 in Zen 5. Win11 is basically Win10 with monetization/extra spyware junk (ads, BS AI trash) and a shittier UI (no taskbar on top).

So maybe it would be best to just update to 5800X3D and go with that.

For sure, I am from a developing country too. :) My laptop has a Nvidia 760M from the early 2010s, I don’t really get 30 fps even on low end games.

My general comment was that expectations around 30 fps have changed a bit in the last ~15 years, where even non hardcore (well off) PC gamers generally appreciate 30+ fps. Even some of the newer iGPUs from both AMD and Intel are aiming for a somewhat better gaming experience (relative to earlier efforts).

Thank you! I currently use my 3080 dGPU for Stable Diffusion. I wonder to what extent NPUs will be usable with Stable Diffusion XL.

I think these days expecting higher than ~30 FPS is not the sole domain of “PC Snobs”, expectations have slowly changed over the past ~15 years,

What use cases are you planning to use the NPU for?

Especially if you need relatively low performance storage with high capacities (above 10TB).