oh, boy, instruction set fanboyism.
I only mentioned x86 because that is the relevant issue in which I pointed out my two-player space thing above.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
oh, boy, instruction set fanboyism.
I only mentioned x86 because that is the relevant issue in which I pointed out my two-player space thing above.
Errr,
One. Not having to fix firmware where the CPU asks for 20-70% maybe 100% power over the TDP specs that Apple as given for the CPU.
https://www.macrumors.com/2018/07/24/throttling-fix-2018-macbook-pro-improvements/
The current stuff can be contained to work, but it isn't what Apple wants. They are looking for stuff not stuck on the same process for last 3-4 years. ( that was AMD GPUs for a while but they aren't at the moment).
Two, Sky High pricing. Intel was pitching the W 3175X at one point for $8K it came out at $3K. Was Intel blowing smoke to confuse AMD or smoking large tacks of legal weed over in Santa Clara. It is kind of tough to tell.
Apple doesn't like paying more than they have too. (e.g. sold "FirePro" GPUs in Mac Pro 2013 so less than what AMD was selling discrete FirePros for. Not cheap as mainstream , but less. )
Apple isn't manically occupied with first hot rod status. Bread of solutions probably at least as much.
Something competitive would probably work for them. AMD doesn't have to completely bury Intel, they just need to be in the same range.
Apple puts no CPU stickers on the their computers at all. Their tech specs pages doesn't even tell you the processor model number. If Yoyodyne Propulsion Systems made viable CPU (using Apple's criteria) for the Mac Pro they'd probably use it. They not caught in a fan boy loop.
I think I can bash AMD if I want to. And, it doesn't mean I am a fanboy or for Intel.
I think everybody, especially, @namethisfile forgets, in the discussion about AMD, is that with Zen2, and Chiplet architecture, they can make specific, custom chips only for Apple.
APU based on 4 CPU Chiplets, and two GPU chiplets, for BGA package, for iMac Pro? No Problem. And whats best - this design ticks every box of HSA design.
Its only up to you, to speak with knowledge, or to speak without it, and look clueless.
For past 5 months AMD CPUs have been outselling Intel 2:1
. Yeah, nieche.
AMD simply hasn't got good products until Zen. Right now - they do. Thats why they are selling. End of the story.
APU based on 4 CPU Chiplets, and two GPU chiplets, for BGA package, for iMac Pro? No Problem. And whats best - this design ticks every box of HSA design.
Sure, you can bash AMD if you want to. But if there is no reason for bashing - you will look ridiculously clouless.
Its only up to you, to speak with knowledge, or to speak without it, and look clueless.
For past 5 months AMD CPUs have been outselling Intel 2:1
. Yeah, nieche. AMD simply hasn't got good products until Zen. Right now - they do. Thats why they are selling. End of the story.
And that sounds suspiciously like the arrangement they already seem to have around Radeons. Weird versions of Polaris and Vega keep popping up on Macs.
I wonder if they could even throw on a few Apple ARM cores in addition to the x86 ones.... That might close the deal with Apple. Apple could move stuff they've had on T2 onto the CPU.
Not according to some. One even implied the OS is irrelevant. There's even the belief ARM is ready to go today. Yet despite all of the advantages ARM is supposed to offer today we don't see any ARM based Macs. Why is that?
For power management, operating of the internals of the iMac - yep, it might be possible. That ARM chip could have direct control over the System-on-Package, for security reasons.
And at the same time, a recipe for potential disasters, with reliability...
You DO know that ARM is an instruction set/programming language, and not a processor, right?
The A series isn't any sort of standard processor these days.
But that is when they were trying to share the same die as the CPU. I think going forward they have seen that multiple die packages are going to get more common. Lots of aspects of their December architecture day pointed at that. When they are not solely in the game of "how to we split the transistor budget between the CPU and GPU" mode then dedicated GPU dies would be a strategic path they'd have to take much more serioursly.
The flaw they clung far too long was that x86 cores made good GPGPU cores. Errr no. They aren't on that path anymore. Haven't been for a couple of years now by all indications from Gen11 details they released. Gen 11 isn't the complete story but it is a clear way point.
There is possibility that a meteor will hit the Apple space ship. I suspect there could be some talking smack about how Apple could do discrete GPU but that is about as detached as Intel's x86 GPGPU play. There is even less volume there than there would be for a Mac "A series" CPU. As pointed out their whole GPU strategy has been a shared memory ( not just flat address space but shared integrated memory). It is doable if throw gobs of money at it but where is the return on investment (ROI)? It is utterly vacant there. Apple is hemming and hawing about making Mac Pro systems because of volume issues. Even more expensive components isn't going to help.
That will probably turn into another blow gobs of money on the Apple Car decision.
The mac pro 2013 put video out and TB on the same buses (even HDMI) and they cut PCI-E slots + NO SATA (with only 1 pci-e sdd)
More then one disk is an needed + maybe room for cheaper 2.5 sata ssd's. Some people may even want one 3.5 bay for a say a BIG HDD
6TB $200 HDD vs 4TB ssd $700 vs 5.2 TB 48K pci-e
Wasn't this Knights Corner? I just feel like we've been here before.
Well, I'll certainly keep an eye on what they're doing. But I think it would have to be a heck of a release to pry Apple away from AMD.
(Not being x86 based probably explains how this is different from Knights Corner.)
Is any of the new stuff Intel announced using a shared memory space? I was really disappointed when their Vega i7 wasn't a shared memory space. That would have been a slam dunk for Apple.
I know you're joking, but it's not entirely irrelevant that they would need a bunch of other ARM chip sizes possibly for whatever car/VR/AR headset they're working on. I think given where they are going it's not impossible that they'd want to have an internal strategy for being able to produce A series chips at many sizes.
And if they ever start building their own in house servers for iCloud/whatever...
Even if A series is different than standard ARM processor, it still is bound by typical ARM "problems" which make this architecture low power.
Architecture is designed to work with very low Amperages. I remember David Kanter talking about this issue with Apple engineers and they said to him that essentially the designs do not allow to put more than 40 Amps, which is ridulously low, for desktop usage. Threadripper, 32 core monster requires 200 Amps of power, on desktop. Because of this very reason, we may never see ARM CPU in a workstation. In spaces where you can make a lot of cores, on low amperages, with relatively high IPC - sure. But we are years from this point.
And the fact that 16 core ARM Server is 5 times slower than simple Threadripper 16 core server shows the mountain ARM has to climb, as a whole. Apple CPUs are not "that" much faster than standard A76 cores. Unfortunately.
Context, and links?For past 5 months AMD CPUs have been outselling Intel 2:1
OEM sales, DIY sales, servers, gamers, first tier vendors, ????
It really means nothing if gamers building systems buy AMD 2:1, but 99% of the HPE/HP/Dell/Lenovo/Supermicro pre-built systems have
![intel_inside_2014[1].png](https://forums.macrumors.com/data/attachments/734/734628-40c9f8bbc75d4c8f294b3c66d6384548.jpg "intel_inside_2014[1].png")
I'm totally with you on that. But what I'm saying is I don't think the instruction set is the cause. No more than x86 ensures a CPU is fast, there are plenty of slow x86 CPUs.
With that in mind, even at the scales we're talking about, the entire portable line including the Macbook Pros is still in striking distance.
And Apple is one of the few in a position to scale that hill. But the challenges are why we're talking about AMD. And the inverse has been hampering Intel. Intel is designed for the high power end of the scale, and has tons of trouble scaling down.
It means a lot. Because of very simple reason. Mindshare. AMD does not have mindshare, yet, and Intel has all of the mindshare. Both companies are playing the long game, here.
Gaming, and DIY is the most reactive market to the situation. Pre-built market is least reactive to the situation. Cause and effect. Two different perspectives. All based on mindshare.
The only thing that is stopping AMD from pushing more hardware - it is money. They simply have no money, yet, to give enough support, to push the hardware down the throats of consumers, like Intel does. In time, it will change.
And hopefully we will have healthy 50:50% ratio between both companies on the market.
Instruction set limits software. Circuit design limits hardware, that runs the software.
Which one limitation is more important, in the grand scheme of things?
Who cares that your software has limits, if your hardware can push 200 Amps, and not be limited? Software can mature, over time.
Would you are if your software had no limits, but your hardware would have blatant wall over which you cannot go over? Hardware cannot mature over time, in any other way, than... buying a new version of said hardware.
And that's where we get to vague implications about the ARM instruction set no one has really explained. It's been implied several times in thread that there is some fault in the ARM instruction set keeping it from scaling. But that's not any opinion I've seen in industry and it doesn't align with other work being done on ARM.
I've poked around and I can't find anyone talking about issues in the ARM instruction set. The most I can find is issues with the canonical ARM CPU implementation sold by ARM that would prevent that specific chip from scaling. But none of those issues are with the instruction set. And Apple's custom implementation, from teardowns and interviews, is clearly moving past the canonical implementation.
Edit: The other problem I find in these arguments is they insist that ARM should not be able to sustain these levels of performance, while the reality is that ARM is. Ars Technica's review of the A12X does a pretty good dig in.
https://arstechnica.com/gadgets/2018/11/2018-ipad-pro-review-whats-a-computer/4/#h5
So am I supposed to believe that the ARM instruction set has some fault that prevents it from getting into higher end CPU territory with absolutely not justification, or am I supposed to believe my own lying eyes?
But again, Intel has the reverse problem. They're doing well on the workstation end, but the smaller you go the worse things get. The limit is my 15" laptop. That's the limit Intel is unable to contend with well.
To flip someone's argument around, if Intel chips were so great and fast, we'd all be running them in our phones right now. We aren't.
Last edited:
So, you don't have any context or links to support your claim. Not surprised.
[doublepost=1549066391][/doublepost]
My espresso machine runs a Linux OS - my guess would be on a low-end 32-bit ARM processor. It simply doesn't need a Xeon.
[quit now to avoid an automotive analogy
] If dual turbo 32-valve DOHC V-8s were so great and fast, all the soccer moms would be running them in their minivans right now. They aren't.Use the right tool for the job. That doesn't mean that there is anything wrong with the other tools.
That's actually exactly my point. I don't think any of this is a technology or instruction set problem, it's a business problem.
It's not worth Intel's time and money to compete in the low power space.
It's not worth any ARM vendors time and money (yet) to compete in the high end space.
It doesn't have anything to do with ARM or x86 instructions. Intel would have to front cash to get into low end, and someone would have to front enough cash on the ARM side to get into the high end. Apple might be crazy enough to do that, or they might stick with x86.
So far we've been in a place of "if it ain't broke, don't fix it." What's changing is Intel is clearly broken.
Perhaps we should encourage more precise discussions - x86 (32-bit) and x64 (64-bit) are very different instruction sets (x86 is a small subset of current x64), and the same for 32 and 64 bit ARM.Technically I never listed ARM64 either.
Instruction Set can be designed around power efficiency and can result in CPU design that will hit the wall, based on the diminishing returns, of the design. Power, Area, Clocks. Caches, Front - end, Back - end of the CPU. Instruction set is part of the silicon design. Im not saying that Apple will not hit certain frequency ranges. But I think everybody forgets that Frequency is just part the equation. ARM instructions can do 50% of the work that x86_64 can do in one cycle.
I do have sources, and links. Why don't you find them yourself, however? Mindfactory sales, for example. Motley Fool and Seeking Alpha analysts discussing the market share of AMD and Intel CPUs, for example?
Why do you have to be lazy, and never do research yourself?
They wanted to compete, and failed miserably. Why?
Because x86_64 Instruction set requires certain CPU design that is not as efficient as ARM is in low-power thermal envelopes. Intel CPUs are faster than ARM, if we compare 7W vs 7W thermal envelopes, in anything that is not Geekbench. If we will compare them in 3W vs 3W thermal envelopes - here ARM will run circles around Intel CPUs. Why? Because ARM designs in 3W's will clock much higher than Intel.
When compared 7W vs 7W = you will see the difference in IPC, Clocks, and sheer power of x86_64 and ability to do more than ARM each cycle.
Last edited:
That’s kind of the same vague thing you said the first time around, with more vagueness. You’re describing all the parts of a CPU, yes, but still not being specific.
koyoot said:
Again, vague vague vague. Which instruction? Is this a RISC vs CISC argument that RISC can’t scale? Is this a pipelining argument? Hyper threading? Vectorized math?