72 comments

[ 3.0 ms ] story [ 130 ms ] thread
The 14” MBP has a 70 watt-hour battery so a 55W draw will drain the battery in just over an hour. And that’s just the CPU.
It's even worse. The Max turbo power on these CPUs is 157W. So if the cooling can keep up these will draw much more than 55W.
These will sit there drawing < 1W 99.5% of the time like all other laptop CPUs.
Especially with the big.LITTLE configuration. These machines simply have the option to push that power envelope if they choose.
I like that. I work on algorithmic performance, making the big-oh notation better often accompanied by a better hidden constant (both at the same time), and I would say...

Waiting for a machine is considered harmful. You are never to wait for your machine, it ought to wait for you. You yes want it running at 1% or so. Then it's at your beck and call, ready to fire up at a moment's notice. Instead of the other way around, you planning your schedule around the machine's work. I've done that, there's problems with this, for instance experiments failing half-way while I was out on a walk. I think one time I left it running overnight, that's a classic.

But if the computer can go all-out for 110 minutes, that's good, means it's a good sprinter, that's actually better than a good marathoner for the type of mathematical experiments typically carried out on a laptop. And for the laptop? Nobody cares about the energy efficiency when it's working hard. Then you plug it in. Or you think the problem over harder.

Like I've never had to do a long mathematical experiment without connecting to the wall. In 13 years as an algorithmist, doing all kinds of shit, not one time. Because then you can't use it for listening to music, or writing essays on Hacker News, or applying to anything, or using emacs. I guess the reason for that is lack of will, so that 110 minutes according to a child node of this poster is much less on an older battery, like 60% of that, so 66 minutes. Then, don't assume I start out on a full charge, I have to code up the algo first, that takes a lot of time at low intensity, a lot of battery, plus you have to finish up the experiment properly, save some files before shutting down. So I guess if you were allotted a limited amount of charge overnight and that's all you could use, then sure. Plus the TDP is lower than if you do really fucked up things with the cores, one time the jumping I think would use up more than TDP. I think I've broken the conceivable limit on one chip, one chip for sure. They assumed my algorithms were sane, reasonable, that's where they fucked up. And all these considerations, I could brainstorm more, but all the considerations reduce the effective battery. Nothing makes it bigger again, it's all a drain. Even writing about the drain like I am now causes drain, accounting for accounting should be accounted for in the total. In other words the Silva Paradox https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu... . In practice I behave totally differently on eg an airplane trip with limited battery than plugged into the wall. Might as well get another user account on everything, full on Dr. Jeckyl and Mr. Hyde with power versus battery.

.

.

And I had fucking old old batteries most of the time, my laptop was only a supercomputer due to algorithms, not software or hardware. Only a few of the algorithms were that good yet, when I had it, so in practice I couldn't say it was truly a supercomputer like this machine. Single thread, dual core, like a crappy plastic Macbook. It got looks at Stanford because your laptop is always visible in class (pre-covid) and a status symbol. Well mine was becoming the best, under the hood. And I did all sorts of stunts on it. It's not meaningfully different than now, that was 2.0 GHz, this is less than 3.0 GHz, clock speed for me is still king. Well I guess it depends. Yeah so in fact because of the clock speed factor, your experiments will run on one core usually, I wonder how that compares with the assumptions that lead to a 55W TDP.

I’d pay good money for a 5kg laptop without any battery at all.

I just need a computer I can move between the office and home.

I don’t need a screen, keyboard or touchpad either. They are a bonus (for meetings, for example) but I’d trade them away for a more powerful cpu any day.

You could build a Mini ITX PC. Some cases have even handles.
I have considered it. Problem is, my employer only allows certain manufacturers due to purchasing and service agreements. Building isn't an option.
Dell owns Alienware so there are some gaming PC:s. Don't know if it's still within corporation possibilities.
Sounds like you really just need sshfs and mount the machine from 2 locations.
Not sure what that is but it sounds linux-y. I work on windows and with desktop/graphics intensive development. Even a remote desktop on a low latency network connection in between is painful.
sshfs is a linux filesystem driver that uses ssh (secure socket host) as a transport mechanism. With proper setup you shouldn't feel latency except with very large files (can be mitigated with caching). I think there's a windows client available but an alternative Windows solution is using WSL2 to facilitate ssh.
Why not just get a mini PC and 2 wall warts? I wouldn't mind hauling this thing around [1], or a mac mini for matter. I'd love to see something similar with a battery so I can hibernate in between locations.

[1] https://system76.com/desktops/meerkat

Have to get either Dell or Lenovo (don't ask). And while they do make mini PCs, they tend to be POS clients or similar, some times with even worse thermals and power than their best laptops.
https://www.dell.com/en-us/work/shop/cty/pdp/spd/optiplex-70...

Configurable with a 12900 & up to 64GB of DDR5. Seems to be pretty much what you're asking for.

Thanks. The drawback of it is that you can't have a laptop to that occasional travel/meeting. I think between "movable small desktop" and "massive tower desktop and tiny laptop to remote into the desktop" the latter actually wins, since you then have a laptop you can bring to a meeting too.
I've carrying a Mac Studio to the office for a few weeks now and it's a pretty reasonably arrangement actually.

It's liberating not having an atrocious keyboard and a tiny screen in front of you all the time.

Intel do some pretty powerful NUC models now, and you (could) get something similar with an AMD model.

>It's liberating not having an atrocious keyboard and a tiny screen in front of you all the time.

That's why we usually work with our laptops docked to external displays and peripherals. You don't?

Yes, it would normally be docked, but you still end up using the screen unless you're in "clamshell mode".

You also end up obliged to use the keyboard at some point, if that's because you took it to a meeting or whatever. Without a laptop, you no longer need an excuse.

Who cares? MBPs will never see this CPU anyway. Assuming that a CPU is at 100% TDP all the time is a bit naif, too.
Yeah, with Intel's recent definition of TDP the CPU can stay at 300% TDP.
Interesting choice to double down on a power hungry chip when targeting mobile. Personally I would prioritise battery life and a quiet laptop above other considerations. But perhaps the gaming laptop segment they’re targeting is big enough that it makes sense?
Don't forget workstation laptops that are basically always plugged in.
my work laptop's CPU sucks 45W: https://ark.intel.com/content/www/us/en/ark/products/201837/...

lasts about 90 minutes when not plugged in (it also has a nvidia card to drive a bunch of monitors). is basically only not-plugged in while checking a quick e-mail or something. i have USB-C bricks and cables scattered throughout the house. doesn't particularly bother me.

With severe throttling it might become an energy efficient but slow chip when needed.
This is kind of a rebrand of the 12900K so it didn't cost much to develop. They have P class and H class processors for laptops that you could actually use on your lap.
>This is kind of a rebrand of the 12900K so it didn't cost much to develop

More correct would be binning instead of rebranding, and no, better binned chips do cost more to weed out from the rest of the silicon, regardless of being identical in design to the rest of the (cheaper) chips, since they can clock higher within the same power envelope due to winning the silicon fab lottery.

The fact that it can draw a ton of power doesn't make it "power hungry".
The mobile workstations which use such CPUs have usually also a much higher idle power consumption, not only a much higher peak power consumption, in comparison with low-power laptops.

For example, my laptop is an older Dell Precision mobile workstation, with a Skylake mobile Xeon, 32 GB ECC RAM and an NVIDIA Quadro GPU.

It has an idle power consumption of 25 W (with the screen light being on), which means that even when staying idle the battery would last only 4 hours (with the screen light off the battery would obviously last longer, but the 4 hour limit would be valid e.g. when I would just use the laptop to read documents).

In this Dell mobile workstation, the CPU power limit for infinite time is set to 60 W, even if that laptop CPU has a nominal TDP of 45 W. Unlike in thinner laptops, in such a mobile workstation the CPU can indeed dissipate continuously for an indefinite time 60 W without throttling. Taking into account the power consumption of the remainder of the laptop, when running at 100% CPU utilization, the battery will be drained in less than an hour.

The new Alder Lake i9-12950HX will be used in the new mobile workstations Dell Precision 7670 and Dell Precision 7770.

These might have a lower idle power consumption than my older model, but I doubt that it will be less than 20 W, when also having a discrete NVIDIA GPU and 64 GB RAM.

I expect that the steady-state CPU maximum power consumption will also be higher than the nominal 55 W TDP, maybe at 65 W, while the turbo power consumption of i9-12950HX is 157 W, according to the rumors.

The other mobile workstation vendors, e.g. Lenovo, HP, MSI, will have similar models.

Well yes, there's not going to be a free lunch. 64GB of ECC DDR4 is likely to consume 24W all by itself. This has basically nothing to do with the choice of CPU.
Gaming and desktop replacement. Plenty of people basically move their laptop between a desk at home and a desk at work.
In SFF PCs like the NUC this makes a lot of sense.
These CPU's appear to be for a niche market, the semi-portable mobile workstation.
Why? This exact class of powerful Intel chips were placed in MacBook Pros before the M1 and those were not exactly niche nor workstation class.
No, HX class didn't even exist in the Intel Mac days. This is a 300W chip being heavily throttled down to <100W for 2-inch-thick "laptops".
Those were “HK” in the 16 inch MBP. Intel never met a SKU it didn’t like.
Which shouldn’t be niche these days. What developer (for example) doesn’t sit 8h per day 98 days of 100 plugged into a wall socket and using external keyboard/screen?

Seems it would be a more common laptop use case than actual mobile use.

I take a train to a far away city sometimes (and I am due to take it more often from October on) and there isn't always a socket available, or I have to share it with someone else in turns (one socket for two or three people).
I need a chip like this for remote work. I work from my at-home office and company policy doesn't allow desktops at home so we work with H/HX series Intel processors instead. Battery life or weight don't matter at all.
Laptop OEMs like Eluktronics or XMG have been putting desktop chips in laptops since forever. Have you checked them out? Sounds exactly what you need.
I have not felt the need for a very fast laptop given reasonable internet. I just do any heavy computation (builds etc.) on a remote server.
You don't. You need better software.

Unless you're encoding video or compiling kernels 24/7 on your local machine.

This is the equivalent of asking for a faster horse instead of a car.

Let me guess, you are a web dev?

We have large C++ code base, and our build times are super slow. With beefy machines, it is tolerable. With laptops, it takes hours.

And yes, writing this in C would decrease our compilation times.. But would make the development and debugging times much longer due to the lack of safe abstractions.

So you listed the one use case that I agreed with and...what?
You might want to express yourself clearer then.

You pretty categorically said "You don't", and said "compiling kernels 24/7", which is something no one does. The implied message there is that the vast majority of people don't need lots of CPU, and there is some "better software" which can eliminate CPU requirement for this. Please correct me if I misunderstood your point.

My reply was an anecdote that when working on C++ codebase, the beefy machines are required. The implied message was that a sizeable fraction of people _do_ need the lots of CPU, and no amount of "better software", whatever it is, can help. Given the website we are on (HN), there is a large number of people reading your message who routinely actually use languages with long compile times.

Therefore, your statement of "You don't" is not correct, I really do need it, despite not encoding video or compiling kernels on my local machine.

Hopefully this makes it clearer for you.

Well that better software in this case was called Delphi :D blistering fast compile times and equivalent runtime performance, just slightly more verbose syntax. Pretty much enough power to do what you need for anything.
Been there, did that, nope. I am not going to writing manual destructors again.

This is not a desktop app, this is a data processing app with pretty complex algorithms. For example, we actually have to make a decision re scoped_ptr (non-refcounted) vs unique_ptr (refcounted) object, use arena allocators, and so on.

Whew, how big is your codebase and what kind of code is it? The biggest C++ project I worked on took about thirty minutes to compile from scratch on a 12 core Macbook. If you didn't need to recompile dependencies the build only took ten minutes or so, a good chunk of which is linking time.
I use VSCode remote SSH to maintain a very large C++ code base running on my 20 core office PC. It works perfectly. No powerful laptop required. Highly recommended.
A lot of people are doing the same a my org.

Unfortunately, we have a number of visualization tools, and while various remote desktop technologies work, they are still not as good as running on a local PC.

Not to mention that sometimes you are in the field, and there is simply no good internet around. But this happens much less now ever since COVID.

>You don't. You need better software.

Oh, the dream no one ever can achieve ;)

someone should make cloud desktops for this sort of thing?
Misleading marketing. It has 8 cores and 8 "cores". Hybrid workloads can really kill performance on big.LITTLE architectures.
Intel's efficient cores are not like the little cores of a big.LITTLE. They are still very potent CPU cores in their own right. We're still talking out-of-order cores & all that jazz ( https://www.anandtech.com/show/16881/a-deep-dive-into-intels... ).

If you have a multithreaded workload that can actually hammer that many cores, you will get decent scaling out of them ( https://www.pugetsystems.com/labs/articles/Intel-12th-Gen---... )

Still, it's a "Core" core and "Atom" core together, leading to all kinds of unexpected consanguinities. As with M1 benchmarks, what really suffers is the 20%-80% load area, together with the lack of AVX-512 which 11th gen had that might be a deal breaker to some. Also, having 8 efficiency cores is just silly, if you just want to run background tasks with minimal power consumption, how do 8 cores lead to that?
The only thing I can figure out to do on these hybrids is to constrain system junk to the e-cores using a systemd directive, and put all my user session processes on the p-cores in the same fashion. Also necessary to boot with interrupts on the p-cores only, unless you want your nvme and ethernet device performance to be terrible. I'm not sure the complexity is worth it.
There's still 8P cores, so no the 20-80% load area is not suffering.

> Also, having 8 efficiency cores is just silly, if you just want to run background tasks with minimal power consumption, how do 8 cores lead to that?

Who said efficient cores must only be for low power background tasks? These efficiency cores are small, very small, so Intel can pack a lot of them into the die. And they provide far far bigger returns on multithreaded workloads than SMT does. As in, given a choice between 2 more P cores (so 10P cores total) or these 8 E cores (so 8p+8e), it's a no brainer to go with the 8 E cores. You get larger multithreading returns with a similar die size cost. And the background power efficiency is just a freebie bonus.

They rate these things like planes are rated: "1900 mile range, 400mph speed" really means "1900 mile range" OR "400 mph speed" because the range is only at best economy speed, and the top speed will never produce range.

Same thing here: stated perf OR stated low power. Power use at max perf: 157W

Nice. I might get one to replace my current mobile workstation. Then replace thermal paste with liquid metal and it's good to go. Should stay silent unless running some benchmarks.
Are all the cores working at full speed at those 55 watts ?
No, at full speed it consumes 175W.
It's a little sad to see Intel trying so hard to be relevant. AMD has really upped the game, and Apple has surprisingly made Intel #3, all in just the last few years.

Intel is only relevant when they throw power at their CPUs, which isn't something you want with mobile chips.

I genuinely hope they become more competitive soon, since it'll mean better performance and prices for everyone. However, the high power chips really need to go.

Don't be sad, since intel changed from a engineering to a pure money-making firm, it went downhill...they deserve it. So many shady things, blatantly lying to customers and faking benchmarks.

I hope they are a getting beaten down until they realize that the are an engineer firm, and if the US is smart, they eventually take over Intel -> "country important products"

Intel - From Inventors of the CPU to Laughing Stock:

https://www.youtube.com/watch?v=JH2nXMv6yZI

Not sad at all. I am glad to see that the market place is working. Punishing companies like Intel that decided to focus more on top management enrichment than quality products. Intel had the chance to do the iPhone CPU and turned it down. Intel was the #1 chip manufacturer in the world but ended up falling behind. Intel had the chance to be the #1 ARM manufacturer in the world but decided not to. And yet the Intel CEO’s who made these decision were all rewarded with massive bonuses. So much for accountability.
If you're willing to spend USD500 on a new Intel CPU for your new system, you might as well buy the new Xeon W-3300 series CPU (3rd Gen Xeon Scalable).

The cheapest W-3300 (W-3323) series is based on Ice Lake architecture, that means it can now suport accelerator for polynomials multiplication and SHA-2 for crypto operations. The lowest specs CPU has 12 cores/24 threads and only cost double compared to Core i9 (around USD1000), however it can support up to 4TB RAM (~30 times more memory than the core i9 max 128GB RAM).

But again we're comparing mobile CPU with workstation CPU, i.e. 1/4 of the power, 55 W vs 220 W. If any laptop manufacturer is adopting the Xeon CPU in a laptop workstation (e.g. Thinkpad Extreme), I'll probably buy the laptop in a heartbeat. If we can have a powerful gaming tablet, why not a powerful laptop workstation? [1]

[1]Introducing The World's Most Powerful Gaming Tablet: The Rog Flow Z13:

https://rog.asus.com/articles/rog-gaming-laptops/2022-rog-fl...