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.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
Intel has been doing it for dual channel memory support since 2004 called Flex Mode
AMD also has similar mode with less performance penalty.
As listed in my table, Apple has two configs need to test thoroughly, namely 4+8 and 36GB mode. 12/24/48 memory sizes are multiply of 4+8 specs, that's why you see M3 Max is being tested with 48GB.
36GB seems odd at first but if you count in increment of 12GB from 24GB, which is similar from 12GB to 24GB with M3 SoC, then it makes perfect sense for Apple to go for it.
Last edited:
With the rumored 36 GB and 48 GB RAM models, I've been speculating too whether that could indicate that the lower models will be 12GB and 24GB isntead of 8 and 16. That would be a big improvement for everyone, and I wonder too if it could save costs for apple to use 12GB RAM modules all around rather than having to produce additional 8GB modules. It would hopefully raise RAM standards all around too and could make Windows computers start to produce laptops with 12 and 24GB instead of 8 and 16. I'm not getting my hopes up, but it would be really cool to see
It'd be a game changer for the kinds of users who are just on the cusp of 8GB not being enough. There's a decent chunk of the market in terms of folks who tend to have quite a lot of windows open at once, but might not necessarily be in the market to build-to-order a 16GB system either.
Yes, on low-end, low cost computers. I just don’t see Apple going that route as they never compromised on these things. Whatever configurations we will see, I’m fairly sure they will be symmetrical.
it would be sad that a bigger imac will come still just with M3 and not also with M3 Pro/Max
This won’t happen, but I think it’s an interesting speculation.
How about an M3 variant that is optimize for power and size? The application would be for a new Mac variant entirely. Let’s just call it a “Mac micro”. A small package (external SSD size) with a single thunderbolt 4 port. it would be powered by the thunderbolt 4, and its intended use, would be to be plugged into a monitor that would supply power and act as a hub for additional ports. Basically, it would allow you to turn any monitor (Pro Display , studio display, etc.) with sufficient interfaces, into a pseudo- iMac.
It is just wild unlikely speculation, but an interesting thought experiment
How about an M3 variant that is optimize for power and size? The application would be for a new Mac variant entirely. Let’s just call it a “Mac micro”. A small package (external SSD size) with a single thunderbolt 4 port. it would be powered by the thunderbolt 4, and its intended use, would be to be plugged into a monitor that would supply power and act as a hub for additional ports. Basically, it would allow you to turn any monitor (Pro Display , studio display, etc.) with sufficient interfaces, into a pseudo- iMac.
It is just wild unlikely speculation, but an interesting thought experiment
I think Apple would use an A-series chip for that kind of product, just as they have done for many iPads. There is already a convention of using an HDMI port these these kinds of things as it provides both power and video out. Such a device would probably only connect wirelessly to other hardware.
This was rumored many years ago, and frankly, I still really don't see the point in such a thing. A portable “stationary” like that does not make sense in our connected world. Especially not when Apple wants large margins on expensive products.
Just for argument sake…
The idea would be to streamline Apple’s low end desktop offerings. Trying to combine the iMac and Mac mini lines into a single product. Basically reduce/eliminate some low end/ low margin products. Provide additional motivation for the studio display and pro display products, which are higher end/large margin products. Kill off the Mac mini pro, to eliminate the overlap with the entry level Mac Studio.
I think it makes business sense for Apple but it is just pretty awful for Mac users.,
A replaceable computer in a screen is an excellent idea, considering that the biggest drawback of the iMac is that you often will through away a perfectly functional screen when the computers is slow after a few years. Apple will score some sustainability point in marketing with such a solution.
Note: many iMac user wants a clean desk without cables.
Combining the Mac Mini and iMac into one product line will never happen. A lot of the Mac Minis being sold are not going into an office environment, but in server farms. Additionally, the Mac Mini fills an important spot in the desktop lineup, which is the sub-$1000 market. The era of the four-quadrant model Steve Jobs implemented when he returned to Apple is long gone, because it is no longer needed.
Someone tell IBM? They have SMT-4 processors. ( a level that Intel and AMD never went to). May not see a another 4 , but sliding back to 2 would not be walking away. Some workloads it works well ; typically top fuel drag racing workloads, it really doesn't . 10 homogenous threads doing the same thing present differently than 10 completely heterogenous threads each doing something different. Whether you properly implement VM security isn't really a matter for SMT or not; it is just not letting security override trying to hit higher 'hot rod' numbers.
CPUs for "every day" , mainstream software. probably will fade away. There are conflicts when take hyper-speculative execution and then try to mix that with SMT. They are coming at part of the problem from two different directions. It gets harder to keep "two trains" on the same track heading in opposite 'directions' from 'colliding'.
However, Intel's 'rentable units' ( which is still a bit fuzzy from material leaked so far ) do not initially look like a total retreat from SMT features to address under utilization when prediction isn't enough though. It sounds like it isn't as 'lightweight' addition to the transistor budget though. ( sounds like the processor internal manager will turn on/off the "simultaneous' part when its 'smarts' judgement allows. ). Still going to be the case that if 'sloppy' with implementing security there will be leaks. There isn't a completely 'free ride'.
The Mini and iMac aren not low margin products. Nothing in Mac line up is . Apple doesn't have a 'low margin' product problem. So they don't need to wander off into the swamp lookin for a solution for that.
The margin is the percentage above the costs. Look at Apple's RAM and SSD prices. Do those reflect bill-of-material costs ? No. So no margin problem.
'low margin' would be arning 1-5% on the sale of the Mini. Folks get their underware in a twist because 25% of $999 is smaller than 25% of $3,999. That really isn't material. What matters there is units sold. If Apple sells 10,000 $999 at 25% then they are probably happier then is sell 10 $3,999 devcies. The first is more profit.
Making a "Micro" isn't likely going to increase the unit sales over those going to loose if just can't sell someone a plug--and-play iMsc. Of the lost sales on folks who have a monitor ( not a docking station display ) , but need a Ethernet jack, etc.
I believe they've had SMT-8 since at least power 8, and still have it in power 10.
I admit to being a bit puzzled when I heard about this - naively it feels like the transistor cost of orchestrating all the parts of a core of such complexity would be better spent on more cores with less SMT, but apparently this is another place my gut is too stupid to be relied upon, since it seems to work well for them.
It's a very different market segment, so your usual logic might not apply. And I also remember reading that some software licenses in that market are per CPU core, and with their SMT model IBM is somehow exploiting this. Don't ask me for details though, I don't understand how it works.
I hadn’t heard this, is the thought behind it that SMT is more of a performance hit now that cpus can pack ridiculous amounts of cores?
There is not much information about how this new Intel approach, but from the few initial bits and pieces available here and there it seems that they plan to approach the problem from the other side, so to speak. SMT is about running multiple threads on a single core in hopes to better utilise the core execution resources (e.g. one thread can use an execution unit the other thread cannot because it has to wait for some other operation to finish). And rentable units seem to be about "fusing" multiple cores into a bigger/wider core on demand. There has been some claims that this second approach will allow Intel to achieve much higher IPC than SMT. But I suppose we will know more when they are ready to talk about it.
At the end of the day, it's all about engineering tradeoffs. SMT works well for some workloads, but doesn't help with single-threaded performance, and it also incurs costs in terms of power consumption and die area.
Having SMT doesn't necessarily mean that you have less cores than you would on a similar SMT-less design, the die-area size is very similar. SMT increases transistor counts by less than 5%, but gives you a 15-40% boost in performance for multithreaded workloads (generally the performance increase is greater than the power usage increase, so it helps for performance per watt also).
It's really just a way to maximize the resources that are already on the chip. It has, however, complicated things for security ever since Meltdown and Spectre though.
Another way to look at it is that SMT is a way to reduce the resource waste your architecture is already suffering from. If SMT gives you 40% higher performance in a multithreaded scenario, that kind of means that 30% of your execution resources that are essentially wasted in a single-threaded scenario.
SMT is a cheap way to get some of it back, but now when the race for highest IPC is back on, designers will try to improve the resource efficiency at the core itself (pun). It's not easy, but it can be done, as Apple very clearly illustrates with their 3.6Ghz CPUs reaching the same performance as Intel at 5Ghz. And if your core is already very good at utilising the resources, SMT becomes redundant.
Intel will simply absurdly widen the cores and their execution massively increasing IPC.
The reason for dropping SMT/HT on consumer hardware is efficiency, and unifying execution between two types of CPU cores: Performance and Efficiency cores.
So far P cores do have SMT/HT while E cores do not.
AMD will follow similar pattern, with dropping SMT from their consumer CPUs.