Comment by mdasen | Hacker News Reader

Comment by mdasen | Hacker News Reader

I think this is a bit where Intel has an opportunity to catch up a bit. 3nm is costing 40% more than 5nm and we'll see how long it will take AMD to move to 3nm. It took AMD 2 years to move to 5nm after Apple started shipping 5nm phones.

Intel is set to launch Meteor Lake on Intel 4 later this year or early 2024. That will put AMD and Intel on a similar process-level footing. If Intel can get Intel 20A out the door around the same time TSMC gets 2nm out the door, we might see Intel overtake AMD on process. Intel's latest roadmap has 20A in the "2023-2024" time frame which seems a tad unlikely at this point and 18A for "2024+". It's possible that Intel will get ahead of TSMC on process if they actually hit their roadmap given that TSMC is looking at 2025 for 2nm. In that case, AMD might be facing a tough battle.

If Intel can control its costs better than TSMC's charges, that will make a big difference. If TSMC's charges for 2nm are going to be 75% more than 5nm, that's going to eat into margins for someone like AMD.

However, while Intel does seem to be making some good moves today, it's probably too soon to completely buy into their roadmap. Yes, they're refocusing their culture in a positive direction. It's still reasonable to have some doubt on Intel's roadmap.

> It took AMD 2 years to move to 5nm after Apple started shipping 5nm phones.

They didn't have much choice. Apple booked most of the production capacity for those two years:

https://www.notebookcheck.net/Apple-secures-80-percent-of-TS...

Intel has already lost Apple. ARM has not been very successful on Windows yet, but Nuvia hased processors are supposed to be launched end of this year. If Nuvia chip is anything close to M1 also then both Intel and AMD are going to have tough time.

> If Nuvia chip is anything close to M1 also then both Intel and AMD are going to have tough time.

Who’s going to write a “Rosetta for Windows” that works seamlessly though?

Isn’t that already available on Windows?

From reading the reviews my impression was that it was too slow.

> Who’s going to write a “Rosetta for Windows” that works seamlessly though?

Microsoft? https://learn.microsoft.com/en-us/windows/arm/apps-on-arm-x8...

They haven't shown any signs of doing a competent job on the $1,800 ARM Surface Pro 9.

> My frustration with this computer wasn’t a workload thing. It didn’t start out fast and gradually slow down as I opened more things and started more processes. It was peppered with glitches and freezes from start to finish.

I’d have only Slack open, and switching between channels would still take almost three seconds (yes, I timed it on my phone). Spotify, also with nothing in the background, would take 11 seconds to open, then be frozen for another four seconds before I could finally press play. When I typed in Chrome, I often saw significant lag, which led to all kinds of typos (because my words weren’t coming out until well after I’d written them). I’d try to watch YouTube videos, and the video would freeze while the audio continued. I’d use the Surface Pen to annotate a PDF, and my strokes would either be frustratingly late or not show up at all. I’d try to open Lightroom, and it would freeze multiple times and then crash.

https://www.theverge.com/23421326/microsoft-surface-pro-9-ar...

ARM SP felt sluggish even without running x86 software when I held it in the store.

Number of cores sold is I think quickly tipping towards the data center. The data center doesn't need x86 compatibility.

The other bulk of thr market is mobile, which is almost entirely arm already.

Microsoft would need to do their part in the OS, but Apple reportedly didn't do all of Rosetta in house. (IBM now owns the start-up reported to have done the core.)

> Apple reportedly didn't do all of Rosetta in house

You're talking about the 2011 version of Rosetta that allowed Intel Macs to run PPC software.

>QuickTransit was a cross-platform virtualization program developed by Transitive Corporation. It allowed software compiled for one specific processor and operating system combination to be executed on a different processor and/or operating system architecture without source code or binary changes.

https://en.wikipedia.org/wiki/QuickTransit

You think Apple wouldn't even partially go back?

Why would they? Right now they have the best laptop and tablet chip, bar none, and total vertical integration. They've invested heavily in M1 and M2 and M3 is due out next year. What would an Intel chip offer, except uncertainty?

On the compatibility side, Boot Camp is less useful when Windows on Arm does its own x86 emulation and works well enough in Parallels. On the graphics side they wrote their own DirectX 12 wrapper. If they can port some AI libs over, running them on 64 GB of shared CPU/GPU memory would be pretty amazing.

Maybe in 10 years, if Intel generationally leapfrogs them again like with the PowerPC era. But short of that, what would they have to gain?

I meant building their ARM M3, M4, and so on CPUs

Like using Intel as a fab?

Yes, aren't they going into that business lately?

Yes, I think so. Sorry I misunderstood you.

Intel has already lost its role as chip provider with Apple. As a fab, if it outperforms TSMC as the parent suggests, it could provide manufacturing for Apple's designs.

Of course, we'll have to see if Intel keeps up after finally clearing its 10nm hurdle.

The M series isn't as great as it seemed. It had an initial advantage of a smaller node, and benefits from everything being on-die. It's huge, so it relies more on the chip tech getting smaller than traditional CPU solutions.

But how can they even find the people that are willing to work cheap and long enough to compete against TSMC in Taiwan? TSMC is having a lot of issues to scale up their Arizona operation.

You pay the right salaries and the people will come. Granted you’re not going to get cheap labor working borderline slavery hours like you can in Taiwan but maybe large scale automation has a role to play.

You are talking about the most automated industry in the world with productivity through the roof. Don't believe everything they tell you.

Why assume that Intel's cost at this resolution is anything less than TSMC's? It likely will be much more expensive for Intel as they lack the experience and are producing in a country with higher basic costs than Taiwan.

> AMD might be facing a tough battle

The Ryzen 7940HS, which is a pretty good value, is on TSMC 4nm, rumored to already cost close to $20k/wafer.

Even if they double the transistor count, a single SKU should theoretically be cheaper to produce, since you're reducing the footprint by 4. Of course, this is assuming yields don't fall too badly on 2nm (which they probably will, 2nm feels like alien technology at this point).

Reminder that those "nm" numbers do not refer to any physical dimension.

TSMC N2 is estimated to be something like 30% denser than N4.

Reply on news.ycombinator.com