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 All Models Is Apple ****ing themselves over the ThreadRipper 3 yet?
- Thread starter macjunkie2013
- Start date
- Sort by reaction score
Apple designed a OS capable on running on x86 hardware, as ive pointed out there is basis to complain if it affects you.
also as i pointed out i understand what their supposed reasoning would be for not changing it up until they need/want to, but it doesn't change the validity of the complaint.
once again complaining aside the original comment was pointing out the provable existence of the limit not a complaint one way or the other.
Apple did not sell the OS to run on random x86 hardware. There's nothing wrong with it.
Apple sold the OS.
like any other non-contractual transaction the seller doesn't get to dictate terms to the buyer.
there's only something wrong with it to the extent that the people it affects negatively that feel its wrong.
but wrong or right is once again irrelevant.
the fact the is limitation exists which was the point i made.
And since Apple OSX can't support a 64-core Threadripper 3 - does it matter? Amazing how quiet the AMD fans are when you point out that Apple OSX is limited to 64 threads - so that the idea of a 64C/128T processor is not supported on Apple OSX.
Of course Apple could make a Mac Pro with a 64C/128T processor - and tout that under bootcamp you could use all of the threads with Windows or Linux. 😉
(Sorry for posting a reponse that's relevant to the OP, rather than just bickering. 😉 )
Last edited:
It's an excellent point. One would think that if apple did go with the 64core AMD, they would be quite capable of extending the limit to more cores. Do you foresee some difficulty peculiar to their kernel that would prevent it working with more cores should apple seek to make it do so?
Putting macOS on a 64-core Threadripper is like putting a Fiat 500 engine in a Mustang. You can do it, but it was not made for that.
[automerge]1579406676[/automerge]
Apple likely already has it in their lab.
I canceled my Mac Pro purchase Until Apple switched to AMD ThreadRipper Period end of story.
Alot of folks imagined Apple was dutifully working on a new Mac Pro back in 2015-2016 too. It didn't happen.
Apple working on a brand new scheduler that iOS (and the rest beside macOS) don't need is probably not high priority. The vast majority of the rest of the Mac line up doesn't need it either. All of the current Mac Pro line up doesn't need it either.
The number of CPU sockets is almost completely immaterial to the issue of the OS scheduler design. If living in the past ( a decade or two or more in the past ) there was some tighter indirection correlation between main CPU core count and socket number. In the very late 2010's and now in 2020 that correlation is exceedingly weak.
Even more so the "CPU" package isn't like the CPU packages from that earlier time period either. There are more subsystems in addition to CPU cores in the package. The overall transistor budget is far higher and don't have to even stick with a single die in the package to have something in the affordable range ( the packaging tech is far and away improved ).
Decades old "rule of thumb" don't really apply. The computer architecture world has changed.
High core counts and/or NUMA issues that the OS schedule would need to account for can appear inside of a single CPU package. The count being higher than a simple native integer register has been talked about in the thread before. NUMA is another issue (the layout of the work over the processors can be an issue).
AMD's current Zen design negates most of NUMA by making the accesses to memory equally as slow (in the single package set up). Almost a forced serialization. But the data structure to track processor objects will still be there.
If Apple requested that AMD give them 24 , 32 , 40 , 48 core ThreadRippers at the Zen 3 stage then they could could "lazily" do nothing and still ship product with probably better $/performance on high core count to Intel's options ( if Intel falls flat in 2020-2021 ). They'd probably ask that it wasn't kneedcapped on memory capacity also. Those core counts and removing the keecapping would be sufficient for next Mac Pro in 2021 (or so) time period if Apple choose to go that way.
Most other high end workstation OS vendors have minor forked their OS to cover the higher core counts. ( to cover more than substantively different hardware bases. Order of magnitude different core counts , memory capacities (with non uniform access ) , and various CPU package interconnects ). With macOS to date it has always been one kernel with a few parameters to cover everything offered. It is doubtful that Apple is going to minor fork macOS to cover what most of the rest of the Mac product line isn't going to do for at least a 5+ year window. It was a strecth even to get Apple to add inter-GPU support (which just came after being present in other systems for a very long time. ).
Apple is going to quickly change the kernel scheduler to optimize it for their lowest unit selling product? That would be highly unusual.
The Mac Pro 2019 is pretty good evidence that Apple isn't interested in chasing the folks who primarily just want to the most affordable workstation container wrapped around the highest x86 core count. That probably won't change in future iterations either.
Is the argument you're using is akin to:
You buy a car and try to use the car as a boat but the manufacturer is at fault when the car fails miserably at being a boat.
?
I think Apple using AMD or not is more about profit to the company bottom line, above all else.
I'm thinking the amount of profit Apple would get from using AMD, is going to be less than the amount of money Intel gives Apple for their volume purchases of chips.
Back in the 90s more than a few systems were still 32 bits and not many had even 256GB of RAM. Not sure how many systems even had L3 caches. ( and L1 and L2 caches were relatively minuscule by todays standards).
One of the factors is that the scheduler just needs larger data structures to handle larger number of processors (and processes/threads ... because if there are more then most likely folks will invoke more ) then all of the objects have to be models and tracked. If you can fit the core Scheduler data structure in a L3 cache and mark it as "don't remove" then can get in and out faster. That isn't even an option if don't even physically have an L3 cache. Or if have an extra small non pageable kernel memory block to work out of .
The Scheduler also has to be 'fair' and not let any process/thread get starved out on time. There are lots of priorities to juggle but don't necessarily have O(n) time (where n is the number of objects being managed ) to get to them. Where n gets very large ( 10k, 100k , 1M , 2M ) that can be relatively a long time. SMT ( hyperthreads) adds t to the pile too.
So one factor is the "extreme" (at the time) core count scheduler tends to have a bigger footprint than what the mainstream processors and memory hierarchies want to allow. For example, for a while had folks clinging onto 32-bit kernels because of the incrementally smaller footprint worked better with limited RAM capacity of the system.
no because hackintoshing a PC or using VM's use the same architecture as the Mac whereas the example you gave doesn't
to properly use your analogy it would be like:
You buy a car and the dealer you bought it from or just a fan of the car manufacturer is claiming without evidence the manufacturer could sue you for using roads they didn't approve even though those roads are public (or private but you have legal access to them.)
Looks to me like you start losing efficiency somewhere around 8 cores and it goes downhill from there (note that the scale of the X axis isn't linear...) - doubling from 8 to 16 only gives a 50% speedup.
Misdirection. Again nothing even remotely to do with CPU socket count.
Apple had Mac OS running on hardware in the same scope of what they were already selling . The move to Intel was not to come up with new Mac models they really didn't sell before.
There were no major architectural coverage shifts in those forked other than instruction set and bit endian orientation. "Plan B" stealth projects.
P.S. The more likely "Plan B" stealth macOS port it is to ARM (Apple flavors ) not AMD. There is absolutely zero > 64 thread pressure there at all. So no OS scheduler change needed. If Apple is dogma linked to past actions.... that would be be even more pressure not to cover the highest end of the AMD line up.
Extremely likely that "Plan B" would not be for the Mac Pro product space though.
Last edited:
As I pointed out to cube it matters the people that it possibly affects, sure that is not a large portion because right now the only way you're affected is if your few high-end users that either use hackintosh or VM and not every possible VM either so its a niche of the niche.
but like everything technology technology and can move incredibly fast, prior to the new Mac Pro the highest core actual Mac was 12 now is 28 who knows when the move to 64+ cores will happen ( presumably only Intel unless Apple opens up to AMD)
realistically the only reason it's affected me as it affected upcoming purchase, rather than holding off for the 3990X I ordered a 3970X which I will make do with until i can upgrade again.
weird must be the circles you travel in, most AMD fans I know either don't care about Mac OS but the ones that do are fine hackintoshing within that limitation.
believe it or not the majority of the AMD CPU's are within it.
I only care because when I build a new machine I tend to like to go the highest end I can ( within reason)
when buying a new actual Mac I typically buy the base model and the highest-end model I actually can justify for my business for comparison purposes. ( in this case it was mid range processor hiring graphics and a mediocre amount of RAM just shy of $20K)
my thought process being... here is as good as it can be and here's a starting point if I want to do the upgrades myself, think long-term because the machines amortized over a few years so can the upgrades be.
in the end I returned both, the $20K machine did outperform everything I had leaps and bounds including the hackintoshes ( the base model did not, it beat every other Mac i have though)
but the fact that I could build part for part of the base model for about half the price threw a wrench into that because as I pointed out in previous threads the "Apple tax" usually is only a few hundred dollars not that offensive.
it means I'm just buying it for the pretty case or perhaps some future afterburner card all of which isn't worth it without some kind of guarantees from Apple.
so once I had firmly decided on a hacintosh it strictly comes down to performance per dollar thus ryzen.
Apple could do that but at the point they did they would update the kernel and associated bits to increase the limit.
no need to apologize ... :S i completely understand all i did was point out the proof that there was a provable core/thread limit and now im being bickered at by (trolls?) basically pushing the hypocritical argument of "If something really doesn't affect me, there's no reason I should care about it."
they just keep rewording it like it will score them some internet points by "proving me wrong" on a point i never really made.
Last edited:
AMD ThreadRipper 3 is not a good match to the top end end Mac products.
AMD went for core count max over TDP. That throws it off from easy placement in the iMac Pro.
AMD prioritized it behind EPYC and Ryzen so timing wise it was a bad fit for the Mac Pro 2019. Most likely it would arrive later than the Intel option (looking from when the design path would have been made back in 2017 or so). The memory capacity is kneecapped to make the EPYC more viable. Both moves by AMD are understandable because the major priority is not placement into Mac product, but more so in subsections of the the overall market. Solely selling to Apple wasn't going to solve their revenue generation and profitability problems.
The window to tie into AMD products could come with Zen 3 architecture but that looks to be 2021 for the Threadripper zone.
The Mac Pro has a decent chance of going back into Rip van Winkle mode. At that point may not have many options for a sizable number of years. ( Apple shouldn't do that, but ..... then there is their track record. ) . If the Mac Pro actually meets basic technical requirements for workload, but doesn't have the 'right' brand or doesn't have 'alternative universe' upper cap then this is a dubious stance.
If enough of the current Mac Pros get sold to give Apple confidence they have a viable product a new Mac Pro will show up in the future. There is extremely unlikely no written in stone requirement that it has to have an AMD processor. There are probably a faction inside of Apple pondering putting an Apple CPU in there ( if they can keep the system price high enough to pay for very high priced , super low volume custom one) or "wait and see" faction. There is good chance that a future Threadripper like product makes the cut, but close to 100% guaranteed or not buying isn't a prudent bet.
Agreed, but you still get significant throughput increases. Real diminishing returns. Meaning even if you only get 50% efficiency of the core, a 50% boost still has decent value. But you're right, the scale shows the efficiency loss is pretty real.
Wonder where they get their data. Perhaps that's with 64bit CPUs. I recall the academic work back in the 90s was you still got more than 50% efficiency (but I don't remember the kind of processing algorithms that was measured for) up to around 32 core. That was certainly for 32bit processors back then.
Also, it's important to point out, on modern systems we have dozens of processes going at the same time that don't need to be parallel but the overall system gets use out of running those separately on their own cores. I find systems with 6-8 cores on modern operating systems just are more responsive because so many of the little processes can always find a ready core.
Are you sure? 🙂
DCG is Data Center Group, which also involves Server group.
Read it and have fun denying that Intel is in trouble 🙂. And remember. Charlie is the guy who reported about how Intel's 10 nm process was dead.
So he cannot be right again, can he?!
You say with a straight face that AMD is not a good fit for Mac Pro because it prioritizes core counts and performance over TDP?
You say that with straight face knowing 56 core Intel monsters draw double the power of single EPYC AMD CPU while still being slower?
Using your argument - Intel is even WORSE option for Apple Mac Pro because its less efficient!
Like its harder to rewrite software, or, I don't know, update it, than to use slower, less efficient, less secure hardware, just because its more convenient, and be two, three generation behind anything that is on the market? 😉
Remember, Intel has become completely irrelevant in terms of raw performance. It will cost Apple sales.
Last edited:
Who is talking about sockets? The point is that Apple prepares for the future regardless of what they will actually offer.
Another shoe to drop. Appears to be unrelated, but it is 100% related to the topic.
Why?
Beucase if Intel would be able to deliver next gen nodes, they would have their GPUs on their own process.
It amazes me everyone’s getting so butt hurt over this. This Mac Pro is exactly what my industry (film) asked for and we’re buying it in droves. I don’t know anyone that is complaining over potential “lost” computing power when Apple has given us 99% of what we need. None of us have any desire for the fastest possible workstation, just what gets the job done, reliably.
Because all you know for a comparison point is the Mac Pro 2013. Of course it's going to be faster. The new Mac Pro performs like a high end machine from 2017 while costing like a machine in 2025.
Wouldn't you rather it perform like a high end machine from 2019/2020 while costing reasonable?
This is why the Mac Pro will ultimately fail and have very poor sales numbers.
Wouldn't you rather it perform like a high end machine from 2019/2020 while costing reasonable?
This is why the Mac Pro will ultimately fail and have very poor sales numbers.
Wow, so say'ith the ambassador to the entire film industry. Thanks!
Of course none of the other professional (or *snork* enthusiast) industries matter of course. PishPosh Sniff Sniff.
That’s not true. A huge amount of us went with hacks and/or Windows based systems where we could while we waited for 7,1. I used both myself. Still have them. Hacks in our environment are unreliable. They were mostly there, but some of us had problems, and most of us couldn’t justify the risk of missing deadlines due to hack issues. Windows was also fine. All but a few pieces of software run on Windows. But most of us dislike it and are willing to sacrifice price and possibly some performance (our needs are already met) to continue using MacOS.
And yes, that would be amazing, but we're not getting that and knowing Apple's history, they weren't going to give it to us.
My point is that this is a tool. If it doesn't suit the needs of the user, there are other options. In this particular instance, it fits our needs for the most part. There are much more powerful options that will run Windows/Linux/Unix depending on what you need.
EDIT: Also, forgive me for saying this, but it really seems that Apple built this thing as a workstation for those in the film/tv/music industry. It's marketed that way. It *might* not be the computer for everyone.
[automerge]1579461791[/automerge]
Point taken. I definitely don't represent everyone, but I understand the basic needs of those I work around, and 7,1 is mostly there. Enough so that we're happy.
Mind if I ask what you were planning on using the Pro for? Is a Windows workstation not an option?
Last edited: