It's kind of amusing to watch Intel struggle to develop their 10nm process, and simultaneously fall on their face trying to respond to AMD's new product breakthroughs punch for punch.
It very much reminiscent of how IBM failed to progress with their PowerPC architecture in the 2000s, and by 2006 Apple had replaced all their IBM chips with Intel ones. The funny thing is that today, Apple is working hard to bring their own ARM chips to Macs due to the fact that ARM development is moving much faster than x64 at this point, and it will be much simpler for software developers to create cross platform apps for iOS & Mac.
Trust me when I say that things are going to get shakey for the computer chip industry in the next few years.
Intel is fucking up but that has nothing to do with ARM, which is a failed abortion on desktops and laptops and it will forever be an abortion.
Connor Adams
This. As shit is x86 is, it's still decades ahead of ARMshit.
Nathan Barnes
To begin with, the ISA is separate from the implementations. While ARM isn't exactly the best ISA ever conceived, x86 is still in its own league of shit ISA-wise.
Second, though, Apple's ARM cores are already beating Intel's, as evidenced by pic related.
Pic related has nothing to do with SSE/AVX 128/256 performance. Can you find us a comparison of something like SW x265 10-bit 4K video encoding using the slow preset? I think at minimum something like that would be a fair metric up to 8 threads at least.
99% of code does not use AVX, so who cares. Also it's not like ARM doesn't have NEON. Or SVE, for that matter.
Caleb Moore
ARM code, sure. But on x86 at minimum 4x32-bit simd instructions are fed through 128-bit AVX assuming it's faster than AVX2.
AFAIK ARMs neon stuff can't hold a candle to x86 but that was back in the snapdraon 801 era. Did ARM really improve THAT much? You'd be talking about like a 20X fold increase in SIMD stuff.
Jacob Perez
>But on x86 at minimum 4x32-bit simd instructions Point still standing, 99% of code doesn't use SIMD, so who cares. To the extent that library routines like memcpy uses SIMD, it's already baked into the test.
Jackson Ortiz
Source? Because last I heard, HSA was essentially dead and intel implemented AVX-512 specifically to ward of HPC from just buying a bunch of teslas and using power9 cpus to connect them altogether.
Speaking of which if ARM is so ahead of x86 why don't we see it in the top500 super computer list?
Michael Smith
lets not forget current avx-512 x265 encoding optimization that can also be seen even with a 10-bit precision encoder.
>intel implemented AVX-512 specifically to ward of HPC from just buying a bunch of teslas and using power9 cpus to connect them altogether. They did, but judging by the TOP500 list, they're not succeeding very well, since that's exactly what numbers 1 and 2 do (except they use Voltas instead of Teslas). >Speaking of which if ARM is so ahead of x86 why don't we see it in the top500 super computer list? Because noone's making ARM processors for that market, quite simply.
Why this focus on HPC, though? OP's context is clearly primarily on mobile/desktop.
Nathan Gonzalez
>15% gain Not H.265, but that's not nearly enough to compensate for the huge advantage the A12 has in H.264 encoding in .
Grayson Mitchell
>reference h264 >h264 at all fuck off retard
Jeremiah Cooper
>not using hardware encoding Why test or even do something so inefficient in the first place?
Eli Stewart
>a failed abortion
so a successful birth?
Adrian Robinson
Because x86 has the highest performance/watt for that market. Even if apple forced it, how would the get mac os running without a huge performance penalty? Emulating x86 syscalls for example doesn't even surpass 1 threaded emulation afaik. Or are they just going to shoehorn iOS on a fucking laptop?
Dominic Miller
>H.265 is the only workload that matters What kind of mentality is this even? Let me guess, you're a """professional streamer""".
Because hardware encoding is extremely inneficient. It requires 10X the bitrate of proper SW encoding. Are you really okay with 1GB/minute video files?
Jaxon Hall
Because it never got AVX-256 let alone AVX-512 support you fucking mong. You know the thing that makes x86 chew through double precision math with great parallelism so fast you'd think it was a GPU?
Oliver Morales
>Even if apple forced it, how would the get mac os running without a huge performance penalty? Why would Mac OS have any intrinsic performance penalty on ARM? Seeing as how iOS uses the exact same kernel, it should be quite clear from empirical evidence too that it doesn't.
Brody Reed
everything has been fucked since x86 and atom massively failed in the phone and tablet space
we are fucking having to re-invent the wheel and recompile fucking everything and i still dont have what i want ...
>You know the thing that makes x86 chew through double precision math with great parallelism so fast you'd think it was a GPU? In the real world, a V100 has 7.5 TFLOPS of DP through where as the absolute fastest x86 processor by that matric, the Xeon Phi 7120X, sports 1.2 TFLOPS.
Julian Allen
Because the programs that run on mac depend on system components programmed specifically to execute on x86 metal. Or what, you think they're gonna port mac os to ARM? Shit man, even windows couldn't pull that shit off and all they had to show for it was RT.
Connor Ramirez
>Or what, you think they're gonna port mac os to ARM? You seriously believe that Mac OS isn't portable to ARM, or even that Apple isn't already running ARM versions of it internally? >even windows couldn't pull that shit off and all they had to show for it was RT. ...but, they did in fact pull that shit off, and the result was RT. Which there's nothing wrong with in itself. The problem with it is that noone is interested in using it since the whole point of running Windows is to running legacy, thus x86, programs.
Jose Howard
Right but this is a huge jump before we saw even AVX on sandy bridge xeons and executing all this on a CPU is significantly easier than optimising for cuda/opencl stuff.
Caleb Sanders
So apple is going to magically get devs to code to ARM mac os? lmao
Kevin Morales
Goes full circle doesn't it. 2008-2017 was Intel's hayday 1998-2007 was amd
Joshua Long
Are you seriously saying that Core CPUs have higher DP throughput than a Xeon Phi?
Caleb Miller
It hasn't been all that long since they switched ISAs last time around.
Eli Hernandez
They've been preparing for this for the last decade, you're absolutely clueless. Its a little thing called SWIFT. Developers are ALREADY making compatible applications.
Julian Anderson
>SWIFT You mean .NET ?
Daniel Jones
Not at all, just saying even lower performing server xeons have vastly improved DP performance, of course phi is better but sadly got eating by volta/tesla whatever the fuck they're using now.
Oh boy, you guys have no idea how bad powerpc mactrash was. These 0.8GHz macbook "pros" are only a shadow of the terrible things apple has concocted.
t. retro macfag
Robert Martinez
>executing all this on a CPU is significantly easier than optimising for cuda/opencl stuff Is that why every HPC application uses SIMD CPU implementations instead of GPUs? Oh, wait...
>Oh boy, you guys have no idea how bad powerpc mactrash was. These 0.8GHz macbook "pros" are only a shadow of the terrible things apple has concocted. No doubt, but how does that contradict what you replied to? All those posts said was that Apple has experience, and even recent experience, switching ISAs.
Christopher Green
You already run x86 versions of IOS apps when you use the xcode simulator. Really how hard would it be to have the other way around with arm versions of mac apps?
James Myers
You think that shit's easy? Devs complain about how fucking hard it is to make x265 efficiently use more than 8 threads meanwhile some lowly fucks are going batfuck insane making opencl/cuda utilize GPUs at least half load efficiently across all cores.
Apple has done EVERYTHING to bury it deep underground. They know what did.
Thomas Ross
Holding out for a bright RISC-V future
Isaac Collins
Intel will be on TSMC 7nm by next year, and 5nm by 2020, screencap this.
Angel Bell
>picture
>Nothing can now be believed which is seen in a newspaper. Truth itself becomes suspicious by being put into that polluted vehicle. The real extent of this state of misinformation is known only to those who are in situations to confront facts within their knowledge with the lies of the day. . . . I will add, that the man who never looks into a newspaper is better informed than he who reads them; inasmuch as he who knows nothing is nearer to truth than he whose mind is filled with falsehoods & errors. He who reads nothing will still learn the great facts, and the details are all false. >Thomas Jefferson
Noah Rodriguez
Intel 10nm superpower by 2030!!! Just WAIT!
Ryan Reed
intel is an israeli company. not using intel chips is antisemetic, user.
Not him but truth starts in lies which is why we TRUST the press to lie to us. Think about that.
Jackson Green
It's intel we're talking about, They never have a "nice competitive state", they're either ramming the ass of everyone or getting their asses rammed hard. And well, they will find a way to get a new much better arch probably 2 or 3 years from now, most likely by buying it from someone else.
Luis Torres
I bet they all did rose to power because the press decided to pick a side instead of delivering the actual truth, thus disarming those people.
Michael Murphy
In the context of this thread, the V (Vector) extension for RISC-V is really interesting. It hasn't been finalized yet, but it seems to hold great promise of bringing the best of CPUs and GPUs, and even features which none of them have, into one ISA. I wouldn't be surprised if the V extension actually becomes the killer features of RISC-V, taking hold in HPC and offline-rendering applications first and working its way down from there.
Hudson Lopez
Popular philosophy aside, the point was simply that it's not just "evil drumpfs" who complain about the press.
Oliver Williams
No, they probably talked smack about them 24/7. What really got them into power was bonafied pragmatism and refined charisma skills. In real life you just have to appease to the right people to get what you want regardless of what others say.
Cameron Ward
tru
Kevin Ward
And that's a bad thing why? I prefer coffee over tea, does that make me antitea? If so, who cares? Nobody likes everything
Brandon Perry
listen, goy. the usa will not let intel fail. israel is america's greatest ally.
Ian Johnson
I'm not american
Thomas Ramirez
didn't say you were... israel is america's greatest ally. the usa won't let intel fail. ego much?
6ghz housefires with thermals just as bad as they are now due to shit Monolithic design and their mcm glued on crap is a joke much worse than bulldozer
Eli Stewart
>how would the get mac os running without a huge performance penalty? Emulating x86 syscalls for example doesn't even surpass 1 threaded emulation afaik Re-read OP and try to grasp a basic history lesson: Apple's machines used to run on Motorola 68k. Apple switched to PowerPC in the mid 1990s. That lasted ten years before they switched to Intel chips.
Thinking Apple can't switch from Intel to ARM is beyond silly. They can. They've done it before. They can run old applications unbearably slow with emulation and just refer you to the new native version and make promises of improved emulation until they declare "legacy" is no longer supported. Again: They've done it before with great success. They know how.
Apple probably has a ARM version of Mac OS ready right now.
Lucas Baker
>apple will go full ARM Why there's people still believing this Bloomberg BS? Didn't they learn anything with supermicro fiasco?
Anthony Robinson
It's a bad name for what are essentially power rangers.
Robert Bennett
never read about that on bloomberg and I don't have to. It's simply a logical step for them. They already make ARM CPUs. If they are able to put their own chip in their Macbooks and still sell them then they will.
Jaxon Moore
The rumor was instigated by an Bloomberg article, acctually.
Josiah Hughes
Several sources are instigating that rumor independently. Whether actually true or not, it's hardly a leap of mind to imagine that they might do it.
Grayson Gonzalez
>needing more engineers/overhead to design yet another grade of A-series chip suitable for a desktop system >needing more resources/area/machinery to fabricate said chip >needing more developers/overhead to implement an x86 translation layer in subsequent OS X versions and make sure it works >losing compatibility and potential sales from customers with edge case requirements for no actual tangible reason >other than "because we can" and some geekbench/datasheet metric that means nothing for real world use and value to people who are actually paying money for it not just shitposting about it on an imageboard people who think an architecture swap is just some simple switch and retarget compilers kind of job are fucking retards
Nvidia build ARM CPU for HPC plus NVLink Nvidia build Risc-V CPU for replace ARM shit,
Nvidia Machines for the future, Intel BTFO.
Brayden Roberts
99% of computer users in 2018 are brainless Facebook drones who just surf the web, check email and do other banal platform-agnostic things that can be done perfectly fine on practically any machine built in the last 10 years or even beyond. Why even bother discussing architecture at all in the context of the lowest common denominator where it's ultimately little more than just a different sticker on the case?
Parker King
You just described exactly what nu-Apple doesn't give a flying damn about :^)
Leo Lopez
nu-apple may be retarded but nu-apple still loves money, it's really hard for me to see how doing all of that bullshit and making the Mac less distinguished from iShit for what amounts to barely more than a footnote in a TV commercial is going to get them more of it
Caleb Young
How ist Swift any different to Obj-C in this regard? Unless the Obj C code contains Assembler both require "just" recompilation.
Andrew Ross
Those 99% of users don't use 99% of the code out there, though; rather on the contrary they use a very small amount of it, namely their web browser.
The vast majority of code, from simple UI, to database engines, to custom web serving code, compilers, CAD/EDA tools, the code parts of game engines, &c&c&c, falls into the "random control-flow integer code" category whose only relation to AVX is that it's useful in common routines like memcpy. Not to say that AVX-using code is totally irrelevant or anything, but it's pretty niche in comparison.
Joshua Lopez
its called either AMD64 or x86_64 you fucking retard there is no such thing as x64.
Jose Rogers
>>needing more engineers/overhead to design yet another grade of A-series chip suitable for a desktop system >>needing more resources/area/machinery to fabricate said chip Both of those go just as much for the mobile chips too, and the mobile chips absolutely dwarf the Mac chips in volume for Apple. >>needing more developers/overhead to implement an x86 translation layer in subsequent OS X versions and make sure it works >>losing compatibility and potential sales from customers with edge case requirements for no actual tangible reason Those are true, except that there is an actual tangible reason for Apple. Clearly they care about something when it comes to making their own parts, since they keep bringing manufacturing of various mobile components in-house, from the CPU now to the GPU, and various peripheral components like the M1, the T2 &c. I can't say I know exactly what they like about it, but there are various reasons one could imagine, like better control of the supply chain, better control over R&D, potentially reduced costs from integration, better planning abilities, who knows.
I too don't think that Apple is just around the corner of switching Macs to ARM or anything, but you can absolutely bet that the option is on the table for them and that they're continually evaluating it.
Joseph Price
Not him,and I haven't actually used either Swift or ObjC, but I'd imagine Swift to be more high-level, with less dependence on machine-specific data-types, and generally less ways to accidentally use non-portable constructs.
Elijah Perry
In Microsoft-land it is universally referred to as x64.
Aaron Gray
>In Microsoft-land it is universally... I'm gonna have to stop you right there Microsoft calls it x64 because they fucked up and made their first 64bit windows for titanium, so when amd64 came along, they couldn't just call it 64bit, because they used that already
Bentley Ramirez
Do the reasons matter? No matter what the reasons are, the fact remains that Microsoft consistently refers to it as x64.
Also you're wrong, because the first 64-bit version of Windows was for the Alpha, and the Itanium version wasn't called x86-64, so that option obviously remained.
Mason Phillips
If I have to choose, sure. Bandwidth and storage capacity have been improving at a tremendous rate when compared to battery technology, so it makes more sense to to put up with increased size rather than the battery rape of software encoding.
Adrian Green
It would be “magical” if it hadn’t been done before.
But Apple already “made” devs port their code from 8080 to PPC, and then “made” them again port their code to Intel. That apple was much smaller than today’s apple. If history’s an indication makes you think they can’t do it now?
Hell, most developers are already familiar with ARM due to mobile, and Apple already “made” Adobe and Autodesk port their code to ARM, while also developing a new interface with different input mode (primarily touch).
If anything, most devs will jump at the opportunity and race to market their software fas “designed for ARM” first in the new software ecosystem that will arise.
Go fight for israel goy, israel and america are one
Jaxson Bennett
> That apple was much smaller than today’s apple That's the reason why they were able to do that. On the other hand, there's Microsoft which tries to port everything to ARM for 6 years now, and still fails. The more software is out there, the more difficult it is.
Lucas Martin
Meanwhile Linux armhf has shitloads of stuff.
Logan Wright
INTEL GPU SUPERPOWER BY 2020
Julian Reyes
Apple never used the 8080.
The Apple 1 and II used the 6502, then they switched to the 680x0 family
Lincoln Sanders
No shit, that is where Intel is going. The SP xeons were so close to the Xeon phi, that the phi line is just going to be xeons.
Dylan Price
I believe that it will. Look up ispc.
Mason Fisher
yup. gimme mass market single board RISC-V machines.
Chase Long
>it will be much simpler for software developers to create cross platform apps for iOS & Mac Would it really make any difference?
Joseph Torres
I want to know if a particular vector width is optimal for a given process size. If so, and we are asymptotically approaching some 'future' then maybe a fixed vector width is not so horrible.
IMHO the danger with the vector extensions for risc-v are that there was a whole lot of experience creating risc-like architectures, so riscv while not cutting edge, does not have any huge thinkos in it. I don't have the same confidence in the vector extensions.
Hunter Watson
That would have been true if it weren't for the fact that they downclock by like one whole GHz under AVX-512 workloads. Given that, even the highest-core-count Xeons have like half the throughput of a Phi. Intel getting a SPMD compiler doesn't really count for a lot when their throughput is an order of magnitude less than a price-comparable GPU, though.
Colton Ward
>I don't have the same confidence in the vector extensions. For the record, the V-extension team has some quite experienced designers in it, among them a guy who worked on AVX-512 at Intel (and is well aware of its shortcomings).
>I want to know if a particular vector width is optimal for a given process size. One of the main advantages of the V extension is that it doesn't force any particular vector width. The VL register mechanic allows an implementation to choose pretty much any arbitrary implementation width while still working with all compiled software.
Landon Baker
>That would have been true if it weren't for the fact that they downclock by like one whole GHz under AVX-512 workloads.
By my reading, the axv512 unit is super decoupled from the main CPU/ integer stuff. As a mix of avx512 instructions comes through, the unit is bought online, which because of its insane power draw, causes the rest of the CPU to be down clocked - if lots of cores are busy. The avx unit itself runs at some reduced clock all the time. My assumption is that avx512 only really makes sense at 10nm and below. Future CPUs won't suffer nearly as much bringing the unit online.
Lucas Morgan
Adding to "experience" part, in every presentation of the V-extension, they've been very clear that they're drawing experience from both CPU-like SIMD implementations, GPU-like SPMD, and more traditional vector processors.
Caleb Evans
>The avx unit itself runs at some reduced clock all the time That's not the case in any current implementation, at least. The AVX-512 units are clocked with the core they belong to and while they're probably large enough to be quite physically separate, their dispatch logic is tied to the OOO machine in a normal fashion. >My assumption is that avx512 only really makes sense at 10nm and below If that is so, then there's something fundamentally wrong with it, since GPUs don't have that problem.
Ryder Jones
>their throughput is an order of magnitude less than a price-comparable GPU, though. Comparing high end to high end, I.e. a $10k CPU to a $10k gpu, the delta was way less than oom. Maybe 0.6. And that is with only 6 channels of ram on the CPU. This is what I had heard was pushing Intel to just use slightly optimized xeons instead of full custom phi's. I was under the impression that at the top end Intel was going to have a price, performance, and similarly power efficient option as GPUs with a normal CPU architecture.
Now my main source is the comments in rwt, where the people that play with this stuff hang out. These parts are above my play grade.