Hacker News Reader

Comment by wifipunk | Hacker News Reader

Intel's Core Ultra packs new tech, but can it really match Apple's M3's ecosystem and energy efficiency?

I think ecosystem is a big thing to call out here. A decent proportion of why the M* CPUs are so efficient is that the OS is built around them, with tasks being appropriately scheduled to CPU cores based on developer hints, and things like the widgets framework taking into account minimising power used to update data which doesn't need it, and background updates for other apps being batched rather than each app waking up for network traffic at random. While it's possible to write inefficient software for macOS if you follow the happy path you'll like come out the other side with something that respects the user's battery.

> A decent proportion of why the M* CPUs are so efficient is that the OS is built around them, with tasks being appropriately scheduled to CPU cores based on developer hints

It helps, but its far from the whole story. Asahi Linux developers have reported exceptionally good battery life on Linux. IIRC they were getting around 7 hours before the GPU driver was implemented and it was continually burning ~60% all-core CPU just to render.

I believe the hardware itself (and the firmware) handles a lot of the power management. Aside from which, Apple's CPU cores do seem to be straight up more power efficient than anybody else's.

7 hours is still quite a bit below the 20-22 hours with MacOS though. It may seem a lot given how bad batteries are on non-apple computers, but it's still less than 1/3 of what it could be.

7 hours is incredible if the CPU was really at 60% sustained usage.

>Aside from which, Apple's CPU cores do seem to be straight up more power efficient than anybody else's.

That reads TSMC 3nm seems to be more power efficient than any other process.

I might be mistaken, but I believe Apple's first gen M1 chips (TSMC 5nm) are still more power efficient than AMDs Zen 4 chips (also TSMC 5nm), although the difference isn't huge (~10-20% ?)

Does the ARM architecture have much to do with it?

Do Mac app devs really do much to reduce power consumption? It doesn't seem like anything special to me, but I haven't actually shipped anything Mac-specific in a long time.

Most of the power on a given day will go to web apps anyway, which ofc the dev doesn't optimize for Mac, and it seems like the Macs do that more efficiently too.

On what do you base these conclusions? There are no numbers here.

As far as the components, Apple Silicon also contains an NPU and GPU. Thus far, there is no reason to believe that Intel has anything significant here aside from chiplets, a technology in use elsewhere already.

"Chiplets" doesn't adequately describe the situation. The packaging tech Intel is using for these new laptop processors is far more advanced than the packaging tech AMD has been using for any of their chiplet-based consumer CPUs.

The packaging tech actually has to be less advanced to be economical.

"That was the first generation. Xilinx worked with TSMC on CoWoS. Their codename was CoWoS. It’s a funny name for TSMC’s silicon interposer. That was a first-generation advanced package technology... I talked to one of their VP. I talked to them many, many times, until one time, I had dinner with [a Qualcomm] VP, and he just very casually told me, he said, you know, “If you want to sell that to me, I would only pay one cent per millimeter square.” One cent per millimeter square. He said, “That’s the only cost I will pay for it.” I said, “How come you didn't tell me earlier?” He said, “You should know that. Why I should tell you? You should know that.” ...Our second generation called InFO meet that criteria and it was sell like a hotcake. So that one word saved my life and the InFO was why Apple was hooked by TSMC."

https://www.computerhistory.org/collections/catalog/10279267...

The packaging AMD uses for their desktop processors is cheaper, and much worse in terms of performance and power efficiency: moving data between chiplets is too expensive to work well for a laptop processor. (That hasn't stopped them from shipping the latest version in a BGA package for use in massive gaming laptops to compete against Intel's desktop-silicon-in-laptop-package parts, but the idle power draw still sucks.)

IMO Intel has over-engineered Core Ultra. The packaging is very advanced but what if that complexity isn't needed? Never send a bunch of chiplets to do a single die's job.

In the short term, this makes them much less dependent on their first EUV process, since that's now only used for a small chiplet of just the CPU cores, while the rest is on a more affordable and reliable TSMC process. That risk reduction may have been critical to being ready to ship this year.

In the long run, I think they also need this kind of packaging to become affordable enough to use across their consumer lineup, not just for the exotic data center parts. Making a big change like this mobile-first and bringing it to desktop in a subsequent generation worked out alright for them with Ice Lake and their 10nm process, so I'm not surprised to see them do it again.

Reading the article. I'm not making any hard claims.

I think the M3 is most likely still going to be better, but I want it to have more competition.

Reply on news.ycombinator.com