Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

leman

macrumors Core
Oct 14, 2008
19,522
19,679
SMT is not in Apple path far more so because they aren't trying to be a "everything for everyboad" CPU implementation. Apple is quite content to detour around some workloads that they don't consider to be important.

Or, Apple doesn’t use SMT because it has negative impact on power-efficiency (see Intel droppping SMT from Atom for this reason) and it requires doubling of certain hardware resources which is less feasible given Apples already humongous register file and reorder buffers.
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
hyperthreqdy exists in intel CPUs to paper over a poor design.

Hyperthreads exists because Intel marketing made up the name. Intel didn't invent SMT. It already existed when Intel adopted it. All the big iron RISC processors have it.

Apple doesn't have it for more so because Apple is far more focus on doing "small iron", single user, single drive systems then on multiple concurrent active users , multiple drives, multiple petabytes systems.

Intel has a dual edge sword of trying to use the same microarchitecture for both of those for most of their CPU offerings. Upside is that one baseline design to implement. Two have issues at the edges of workload differences.

P.S. Apple is also more highly dependent upon bleeding edge fab process to "hige" more bigger cores as "smaller" when implemented on the far leading edge. The have used increased transistor budget to add more cores over longer time. They aren't being highly driven by "core count" or "thread count" feature wars. THey are taking 'wins' as they come; not really forcing the issue.
 
Last edited:

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
Or, Apple doesn’t use SMT because it has negative impact on power-efficiency (see Intel droppping SMT from Atom for this reason) and it requires doubling of certain hardware resources which is less feasible given Apples already humongous register file and reorder buffers.

Sigh....

it doesn't require doubling of all the resources. You just primarily need more rename registers and some more logic in the dispatch and retirement areas. The number of the primary function units ( adders , mathops , loaders , etc.) stays exactly the same.

Apple's humongous "re-order" buffer puts very similar pressure on bulked up rename registers. Power gating stuff not using is a bigger issue why Apple's sysetm 'works' along with narrowing the workload metrics.


Intel didn't drop SMT from Atom as opposed to stop trying to recycle mainstream design under an Atom label. Atom didn't start off with SMT. It got it more so because Intel was being cheap rather than optimizing a specific Atom market design. The current ones don't have it because Intel went back to trying to build Atom architecture as a funded , distinct design.


Nobody in the core enterprise product groups at Oracle, IBM DB2 group , SQL server is having nighmare sleep about or "fear of missing out" drama over not having a port to M1. Apple not having SMT is a design choice with ramifications. If their "bread and butter" systems are phones that run on relatively small batteries with relatively small ( sub 1TB data sets ). That's a reasonable trade off to make. Not going to help run better over random access PB sized data sets though.


P.S. there was also some overlap with Atom development with Xeon D where Intel had server like duties, but wanted a "cheaper" core ( to both customers and themselves ). When coupled up to some higher end NAS/SAN/Network I/O data driven devices then SMT does make more sense.
 
Last edited:

CWallace

macrumors G5
Aug 17, 2007
12,528
11,543
Seattle, WA
Does anyone think the longer Apple waits to introduce any new MBP 14/16, the higher the odds are that the system will be based on new (A15) microarchitecture as opposed to existing (Firestorm/Icestorm)?

It seems that the delay is due to component shortages (MiniLED displays and perhaps other items) and not because Apple wants to use a new-generation SoC. All rumors so far point to that going into the 2022 MacBook Air (and perhaps the 2022 "iMac Pro", "Mac mini Pro" and Mac Pro 8.1).
 
  • Sad
Reactions: dustSafa

cmaier

Suspended
Original poster
Jul 25, 2007
25,405
33,474
California
It seems that the delay is due to component shortages (MiniLED displays and perhaps other items) and not because Apple wants to use a new-generation SoC. All rumors so far point to that going into the 2022 MacBook Air (and perhaps the 2022 "iMac Pro", "Mac mini Pro" and Mac Pro 8.1).

No, not “all rumors.”
 

Jorbanead

macrumors 65816
Aug 31, 2018
1,209
1,438
They could use a double-sided motherboard, which would allow PCIe slots on both sides.
There wouldn’t be enough width in the current chassis to have PCIe slots on both sides. So you’d have a really wide body which defeats the purpose of shrinking it.
 

theorist9

macrumors 68040
May 28, 2015
3,882
3,061
There wouldn’t be enough width in the current chassis to have PCIe slots on both sides. So you’d have a really wide body which defeats the purpose of shrinking it.
I just realized the Mac Pro already has a dual-sided logic board, with processor, graphics, and PCIe expansion on one side, and storage and RAM on the other. So they're already making use of both sides.
 
Last edited:

leman

macrumors Core
Oct 14, 2008
19,522
19,679
it doesn't require doubling of all the resources. You just primarily need more rename registers and some more logic in the dispatch and retirement areas. The number of the primary function units ( adders , mathops , loaders , etc.) stays exactly the same.

How is what you are saying different from what I am saying? Of course the backend stays the same. But you need hardware to track the multiple threads, which you yourself admit.

Apple's design focuses on maximizing single-threaded ILP, which they manage quite admirably, given the fact that their 3.2 ghz CPU offers competitive performance with 5Ghz state of the art designs from industry veterans. My understanding of hardware design is very limited, but it seems to me that adding SMT into the mix might interfere with their design goal. By adding SMT they will likely sacrifice the single-thread performance which is not what they wold want to do. Intel/AMD on the other kind benefit from SMT because it helps to offset the major frequency drop off associated with heavy multithreaded execution on their CPUs.

Power gating stuff not using is a bigger issue why Apple's sysetm 'works' along with narrowing the workload metrics.

That is a very good point indeed. Apple's expertise in power management is second to none and they fully incorporated it into their designs.

Intel didn't drop SMT from Atom as opposed to stop trying to recycle mainstream design under an Atom label. Atom didn't start off with SMT. It got it more so because Intel was being cheap rather than optimizing a specific Atom market design. The current ones don't have it because Intel went back to trying to build Atom architecture as a funded , distinct design.

Again, how is it different from what I am saying? If Intel decided that SMT is not suitable for an energy-optimized design, that's how it is, no?


Nobody in the core enterprise product groups at Oracle, IBM DB2 group , SQL server is having nighmare sleep about or "fear of missing out" drama over not having a port to M1.

Why wold they? Apple does not make large-scale server systems nor do they seem to be interested in the market. Their hardware targets the computational needs of individual users.

Apple not having SMT is a design choice with ramifications. If their "bread and butter" systems are phones that run on relatively small batteries with relatively small ( sub 1TB data sets ). That's a reasonable trade off to make. Not going to help run better over random access PB sized data sets though.

Given the fact that Appel users are not interested in petascale processing, seems like the design choice is reasonable.
 

theorist9

macrumors 68040
May 28, 2015
3,882
3,061
I for one have been thinking for a while that we won’t see firestorm/icestorm but something different. In fact, I expect a chip that’s quite unlike the mobile base that M1 is using.
I believe Jon Prosser predicts they'll be named snowstorm/s**tstorm.
 
  • Haha
Reactions: EntropyQ3

CWallace

macrumors G5
Aug 17, 2007
12,528
11,543
Seattle, WA
So MaxTech just released, IMO, a well-thought out video about the upcoming Macs and a new Apple Thunderbolt Display using the iMac's panel.

In this video, he says the next iMac will not be a pro model and will be priced identical to the current Intel iMac 5K at around $2000.

Ming-Chi Kuo (whose has strong contacts in the supply chain) did say in early 2020 Apple was sourcing a 27" 5K MiniLED panel for use in what he presumed would be a 2020 Intel iMac Pro refresh. I now wonder if this new iMac might have a 27" 5K MiniLED panel, which would be fantastic for me as a current iMac 5K owner. That being said, I think we could see a custom 28" 5.5K panel should Apple be able to source one cheap enough to meet the $2000 price point.

I believe it is pretty much a given that MacBook Pro 14/16 are coming either in late September or early October. And Apple will talk-up M1X a great deal to lay the groundwork for it's use in the next iMac and "Mac mini Pro" which I remain convinced will use M1X. I expect both to be launched in December 2021 or January 2022 via Press Release.

I remain convinced M2 will not make an appearance until the Spring of 2022 in the next MacBook Air.
 

theorist9

macrumors 68040
May 28, 2015
3,882
3,061
So MaxTech just released, IMO, a well-thought out video about the upcoming Macs and a new Apple Thunderbolt Display using the iMac's panel.

In this video, he says the next iMac will not be a pro model and will be priced identical to the current Intel iMac 5K at around $2000.

Ming-Chi Kuo (whose has strong contacts in the supply chain) did say in early 2020 Apple was sourcing a 27" 5K MiniLED panel for use in what he presumed would be a 2020 Intel iMac Pro refresh. I now wonder if this new iMac might have a 27" 5K MiniLED panel, which would be fantastic for me as a current iMac 5K owner. That being said, I think we could see a custom 28" 5.5K panel should Apple be able to source one cheap enough to meet the $2000 price point.

I believe it is pretty much a given that MacBook Pro 14/16 are coming either in late September or early October. And Apple will talk-up M1X a great deal to lay the groundwork for it's use in the next iMac and "Mac mini Pro" which I remain convinced will use M1X. I expect both to be launched in December 2021 or January 2022 via Press Release.

I remain convinced M2 will not make an appearance until the Spring of 2022 in the next MacBook Air.
I'm curious why you said 28" 5.5K, since that's very specific. Are there rumors about this size and resolution?

Since Apple hews pretty close to 220 ppi with its retina montitors (XDR, iMac and laptops), I'd expect if they did go to 5.5k, it would be 30". Though I'm not sure if they'd bother to change monitor size on an Intel iMac refresh.
 
Last edited:
  • Like
Reactions: reallynotnick

CWallace

macrumors G5
Aug 17, 2007
12,528
11,543
Seattle, WA
I'm curious why you said 28" 5.5K, since that's very specific. Are there rumors about this size and resolution?

Since Apple hews pretty close to 220 ppi with its retina montitors (XDR, iMac and laptops), I'd expect if they did go to 5.5k, it would be 30". Though I'm not sure if they'd bother to change monitor size on an Intel iMac refresh.

Nope, just guesswork based on seeing 4K displays at 28" so that panel size is in general production. The 30" displays all seem to use curved panels, which I don't see Apple doing.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.