News for hackers. Stuff that matters.
Re:Apple, ARM, and Intel by dijit on Tuesday June 16, 2020 @10:08AM (#23539299)
I mean, ARM is 'there' in most cases, the issue tends to be perceived latency of RISC vs CISC CPU's on desktop, not overall performance.

There are decent quality laptops shipping with ARM already, like the Thinkpad Yoga: https://www.theverge.com/2020/1/6/21050758/lenovo-yoga-5g-sn...

There's also the obvious: Pinebook pro.

Re:Apple, ARM, and Intel by DCKing on Tuesday June 16, 2020 @10:45AM (#23539712)
The RISC vs CISC discussion was important in the nineties but not anymore. There's absolutely nothing 'reduced' about modern ARM chips, and every Intel processor since the late 1990s has used significant RISC design elements.

In modern contexts, the RISC vs CISC discussion still gets brought a surprising amount. But it is simply not relevant anymore.

See also previous discussions on Apple SoCs: https://news.ycombinator.com/item?id=19327276

Re:Apple, ARM, and Intel by DC-3 on Tuesday June 16, 2020 @10:13AM (#23539346)
I think the Pinebook is an awesome piece of kit and if I could justify doing so I would absolutely buy one but from everything I've read it's not really something you'd want to do serious work on unless you also have the luxury of a beefy workstation to shell into. I suspect the same applies to the Yoga - perfectly alright for most of the day but just wouldn't have the firepower for big compilation jobs or serious multitasking.
Re:Apple, ARM, and Intel by dijit on Tuesday June 16, 2020 @10:29AM (#23539511)
Your original assertion

> Unfortunately, though Apple will likely find very great success with their ARM Macs, it does not necessarily follow that the architecture will gain significant ground outside the Apple-sphere.

Does not imply heavy workloads, compilation being offloaded to cloud is something already being done though.

I suspect that ARM will cannibalise most workloads, since it's shown very strong early competence already in photo and video editing on the iPad Pro, I highly suspect advancements in that direction will come to quallcomm CPU's in future.

Re:Apple, ARM, and Intel by DC-3 on Tuesday June 16, 2020 @10:44AM (#23539690)
My original comment discussed the prospect of ARM becoming the leading architecture for the entire range of machines that at the moment tend to be x86 - that's laptops, desktops, workstations, and servers. For this to happen, ARM needs to be capable of dealing with intensive, CPU-heavy tasks like compilation. As the Torvalds quote in the linked article explains, ARM servers cannot become the norm without ARM desktops becoming the norm and vice-versa.
Re:Apple, ARM, and Intel by dijit on Tuesday June 16, 2020 @10:51AM (#23539782)
Sure, but it's not all or nothing.

ARM can eat the earth but it will not be a shift in a single year, I was suggesting a likely way forward and that is bottom up, coming from low end laptops to higher end laptops and eventually workstations and servers.

Just because it's not for -you- right now because you need local compilation does not mean a JS coder or a normal office worker would be unable to use an ARM based machine.

Nothing prohibits ARM from ever being workstation grade, it's just not the focus right now- but you seem to imply that because there's no workstation capable CPU that there never can be one, which I find disingenuous.

Re:Apple, ARM, and Intel by CountSessine on Tuesday June 16, 2020 @01:04PM (#23541420)
I think you’re right, but I also think that he’s right in that there’s a real chicken and egg problem here.

Just because it's not for -you- right now because you need local compilation does not mean a JS coder or a normal office worker would be unable to use an ARM based machine.

I don’t know what workflow other JS coders have, but mine starts with local development in Node. Does V8 produce code that’s as optimal for ARM as x86_64? That’s a pretty big deal to me, and can in fact be very enabling. The local development experience really needs to be good to build a platform, otherwise development will be halting and piecemeal the way embedded development usually is.

Re:Apple, ARM, and Intel by GoblinSlayer on Tuesday June 16, 2020 @01:34PM (#23541781)
> JS coder

> never tested on android

just as planned

Re:Apple, ARM, and Intel by CountSessine on Tuesday June 16, 2020 @11:44PM (#23547516)
> JS coder > never tested on android just as planned

???

I’m back-end, not front-end. I think I mentioned Node, right? Why would I test backend code on a phone?

Re:Apple, ARM, and Intel by DC-3 on Tuesday June 16, 2020 @11:35AM (#23540383)
I agree it'll happen in time but my point is that Apple adopting ARM will not necessarily cause an instant chain reaction across the industry. My concern is whether my next laptop (perhaps 4 years away) will be ARM - and I'm not sure that there will a suitable machine on the market (I need Linux support) even if there is a perfectly capable range of MacBooks.
Re:Apple, ARM, and Intel by GoblinSlayer on Tuesday June 16, 2020 @01:11PM (#23541508)
Is compilation a problem though? It's an unattended process, so it doesn't matter how long it takes, 1 hour or 2 hours, both equal infinity for all practical purposes. I thought ARM has problems only with FPS games, which are not parallelizable and sensitive to single core performance.
Re:Apple, ARM, and Intel by simias on Tuesday June 16, 2020 @10:10AM (#23539322)
What do you mean by latency here exactly?
Re:Apple, ARM, and Intel by dijit on Tuesday June 16, 2020 @10:27AM (#23539494)
I'm not exactly sure what I mean and that's part of the issue, when using an ARM CPU with desktop software which has not been optimised to hide UI "lag" there's something that feels like it just.. lags.

You can feel it on a rpi if you use it for a day, it's not that it's underpowered, you can grab a 15yo celeron that is much slower and single core and it will have less perceptable latency. The Celeron will "feel" as if it's working, the ARM chip will "feel" as if it's laggy.

I can't quantify this behaviour, but it's consistent across all ARM CPU's I've ever used with traditional desktop software.

Notably these issues don't seem to exist on Android or iOS, someone once told me this is due to the UI elements having a higher priority than anything else on the system.

Re:Apple, ARM, and Intel by DCKing on Tuesday June 16, 2020 @10:57AM (#23539872)
If your experience with "desktop UI on ARM" is based on the RPi and other ARM SBCs, the overwhelmingly likely explanation for this perceived lag is lacking I/O performance. Running a Raspberry Pi (or any system) from any cheap flash storage setup like SD cards or cheap eMMCs degrades desktop responsivity a lot. The ARM desktop hardware most of us have experience with has significant I/O problems, but that's due to them being cheap hardware, and not due to ARM.

I/O is actually the real prevalent bottleneck for desktop interactivity on all modern hardware. Even older x86 systems with spinning platter hard drives are set up for a more responsive desktop experience due to the I/O performance being suited for the desktop. Modern Android phones and iPhones also have much better I/O setups for interactive use (UFS for high end Android, Apple actually uses NVME because responsiveness is so crucial) .

Certainly CISC vs RISC has nothing to do with it as that discussion really has no bearing on the situation anymore on discussions about modern ARM and Intel (see my other post).

Re:Apple, ARM, and Intel by zozbot234 on Tuesday June 16, 2020 @11:11AM (#23540063)
> Running a Raspberry Pi (or any system) from any cheap flash storage setup like SD cards or cheap eMMCs degrades desktop responsivity a lot.

Not really, if the system has enough free RAM to use as disk cache. For many scenarios, that suffices to run quite acceptably.

Re:Apple, ARM, and Intel by DCKing on Tuesday June 16, 2020 @11:19AM (#23540168)
RAM as disk cache is only useful for things you have already loaded. It doesn't help your boot times, new program launches, or loading new content in your program.

I'm not saying SBCs are "unusable" for desktop use cases, I'm saying I/O is the likely explanation for desktop responsiveness issues they encounter on SBCs.

Re:Apple, ARM, and Intel by dijit on Tuesday June 16, 2020 @12:32PM (#23541074)
Nobody here is talking about boot times, I was talking about responsiveness.. it's easy to test your theory, you can boot USB on the new RPI, and USB3.0 has quite a high bandwidth, definitely comparable to SATA2, nearly SATA3
Re:Apple, ARM, and Intel by DCKing on Tuesday June 16, 2020 @12:46PM (#23541231)
There's a good reason that they're making it more possible to boot a Raspberry Pi from USB. It's more reliable storage and has better performance for many use cases :)
Re:Apple, ARM, and Intel by _bxg1 on Tuesday June 16, 2020 @10:33AM (#23539563)
Do those ARM systems have graphics chips, or are they rendering the UI on the CPU? Modern x86 processors have integrated graphics for GUI basics, video, etc. Linux desktops can theoretically run off of just the CPU, just... not super well.
Re:Apple, ARM, and Intel by zozbot234 on Tuesday June 16, 2020 @10:43AM (#23539682)
Even Raspberry Pi has an integrated GPU. They feel slower because they're lower-frequency in-order chips, and while the chips themselves have multiple cores many applications don't use them properly - more specifically, they don't render the UI in a dedicated thread.
Re:Apple, ARM, and Intel by fouc on Tuesday June 16, 2020 @06:43PM (#23545212)
RPI are cheap hardware not really designed for desktop computing I guess.
Re:Apple, ARM, and Intel by icedchai on Tuesday June 16, 2020 @10:57AM (#23539871)
Yes, I've noticed this as well. ARM Linux desktops feel sluggish. It is not IO, because I have done tests after everything has been cached, loaded once. I have a RPi 4 here, as well as some older Pine64 boards.
Re:Apple, ARM, and Intel by ianai on Tuesday June 16, 2020 @10:32AM (#23539553)
I forget the specifics, but risc winds up feeling slower because it’s a more “correct” arch than cisc. Been a while since this came up!
Re:Apple, ARM, and Intel by ianai on Tuesday June 16, 2020 @10:30AM (#23539525)
RISC cpus have historically been under-clocked and perceivable slower than their cisc competitors. Apple was running on risc cpus before going intel x86. They were constantly marketing their speeds as less important than speeds matter to x86 architecture.

I hope they’re able to devote two cores to UI/UX at a much higher frequency than the general compute cores. That would take some engineering know how, but could negate the performance complaints largely.

Re:Apple, ARM, and Intel by kmeisthax on Tuesday June 16, 2020 @12:46PM (#23541222)
"RISC" isn't a single architecture, it's a philosophy for designing architectures. The "RISC" Apple was using before x86 was called PowerPC, and it was designed mostly by IBM. It was also a horrible piece of overheating garbage that I'm still honestly surprised wound up in the Xbox 360 and PlayStation 3. The "RISC" Apple is using now, ARM, is a far more practical architecture that Apple has already beat Intel with in single-thread perf... at a lower clock rate, with strict power limits. Letting the A13 stretch it's legs with wall power and a big heatsink would completely crush Intel and x86 in every benchmark except multi-threading.

User-perceived latency has little to do with the clock rate of a processor, though. What you need is:

1. Immediate feedback upon user input 2. Progress animations that move, even if progress is indeterminate 3. High frame rates on your animations 4. No frame stutter or torn frames

Out of the four, low clock rate will mainly impact frame rates and pacing, but that can still be mitigated by using what you have as efficiently as possible.

Re:Apple, ARM, and Intel by Symmetry on Tuesday June 16, 2020 @10:48AM (#23539748)
That's just a matter of the process node Apple's PowerPC chips were on. When RISC chips came out their big selling point was that they could be clocked many times faster than CISC chips at the time could because it was easier to pipeline them and because you could put the entire CPU on a single chip.