Discussion Apple Silicon SoC thread

[Eug]

M1
5 nm
Unified memory architecture - LP-DDR4
16 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 12 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache
(Apple claims the 4 high-effiency cores alone perform like a dual-core Intel MacBook Air)

8-core iGPU (but there is a 7-core variant, likely with one inactive core)
128 execution units
Up to 24576 concurrent threads
2.6 Teraflops
82 Gigatexels/s
41 gigapixels/s

16-core neural engine
Secure Enclave
USB 4

Products:
$999 ($899 edu) 13" MacBook Air (fanless) - 18 hour video playback battery life
$699 Mac mini (with fan)
$1299 ($1199 edu) 13" MacBook Pro (with fan) - 20 hour video playback battery life

Memory options 8 GB and 16 GB. No 32 GB option (unless you go Intel).

It should be noted that the M1 chip in these three Macs is the same (aside from GPU core number). Basically, Apple is taking the same approach which these chips as they do the iPhones and iPads. Just one SKU (excluding the X variants), which is the same across all iDevices (aside from maybe slight clock speed differences occasionally).

EDIT:



M1 Pro 8-core CPU (6+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 16-core GPU
M1 Max 10-core CPU (8+2), 24-core GPU
M1 Max 10-core CPU (8+2), 32-core GPU

M1 Pro and M1 Max discussion here:

Page 78 - Discussion - Apple Silicon SoC thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M1 Ultra discussion here:

Page 109 - Discussion - Apple Silicon SoC thread

Page 109 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M2 discussion here:

Page 127 - Discussion - Apple Silicon SoC thread

Page 127 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Second Generation 5 nm
Unified memory architecture - LPDDR5, up to 24 GB and 100 GB/s
20 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 16 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache

10-core iGPU (but there is an 8-core variant)
3.6 Teraflops

16-core neural engine
Secure Enclave
USB 4

Hardware acceleration for 8K h.264, h.264, ProRes

M3 Family discussion here:

Page 215 - Discussion - Apple Silicon SoC thread

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M4 Family discussion here:

Page 263 - Discussion - Apple Silicon SoC thread

Page 263 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M5 Family discussion here:

Page 431 - Discussion - Apple Silicon SoC thread

Page 431 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Eug]

I also think if there is a v2 of this MacBook, it will get the A19 Pro and sport 12GB of RAM, that’s the one to hold out for. It also depends if they refresh this Mac every year or every 2 years.

This is what I'm worried about. Apple has a habit of crippling its v1 products, only to (mostly) fix the issues in v2. It would suck if Apple ships a crap-spec'd machine (8 GB RAM and single USB-C port) for that v1 crippled version and for the race to the price bottom. It's already enough of a downgrade to ship the MacBook with smaller 12.9" screen, no True Tone, lower screen brightness, slower SSD, no backlit keyboard, no MagSafe, and no fast charging. With all those downgrades, 12 GB RAM and dual-USB-C should be included, which should be easily doable if the retail price is $749-$799. That price tier would be a good incentive for some to upgrade to the 16 GB $999 MacBook Air instead too, a nice price ladder as Apple likes.

In our case, if they refresh this every two years, v2 will be too late for our initial kid laptop purchase. I guess we'd just have to suck it up and get the MBA instead.

 

[MerryCherry]

No back-lit keyboard is crazy. What was the last Macbook that didn't have it?

 

[johnsonwax]

Agreed. It seems like total bullsh!t to me. Nonetheless, some Apple rumour sites are reporting on this. 🙄 Lame.


I'm looking to get an A18 Pro MacBook or else an M5 MacBook Air for my kid soon. All that would really be needed in such a machine is 2 USB-C ports (MagSafe connector and headphone jack optional), 12/16 GB RAM, and the option to upgrade to 512 GB storage.

The kid's current machine is a 2015 13" 2.7 GHz Broadwell i5-5257U purchased mint with new Apple battery and top case in 2021. I can't believe it's already been almost 5 years with that used machine. Geekbench 6 of that i5 is about 1000/2000, but A18 Pro is about 3500/9000 which is a ginormous upgrade, much faster than M1. Having hardware h.265 HEVC encode/decode that the A18 Pro supports has now become a basic requirement too, and having hardware AV1 decode and Apple Intelligence support is also a nice bonus. However, it would also be nice if it could support 6K monitors, as I already own one.

I would be buying with education pricing, but even if the A18 Pro MacBook has all those features, the purchase decision would depend upon whether or not it qualifies for the Back-To-School promotion this summer. I'm pretty sure this A18 Pro MacBook would be part of the promotion though, since students would be a primary target demographic, but you never know given its rumoured low price point.

So, I have a lot of experience with Apple educational products, pricing and promotions. Those deals are usually best if you can buy them through an educational institution like a university bookstore. They'll typically offer better pricing than even Apple directly. And if Apple is focusing this toward education, there may even be special configurations you can't get outside of the educational channel. They don't always have these, but periodically a particularly educationally focused product will get this. I can't imagine it wouldn't qualify for BTS, but it might be something fairly small like some low-end Beats. I'd be surprised if the low price machine has BTO options. My guess is 2 fixed variants. Just making BTO available adds a lot to costs.

 

[johnsonwax]

No back-lit keyboard is crazy. What was the last Macbook that didn't have it?

I mean, if you're expecting a $699 or something laptop out of Apple, a lot of things are going to get cut, if nothing else to make the Air a viable upsell.

 

[Nothingness]

Ah well I was using Geekbench single-core as the basis.

The aggregated score of GB creates a distortion due to SME.

OTOH looking at this for instance: https://browser.geekbench.com/v6/cpu/compare/16766136?baseline=16768383
It shows 60% faster for clang (basically the workload I'm interested in). Somewhere in between my 40% and your 85% 🙂

 

[MS_AT]

Semi related side question, does the geekbench clang test consists of pure compilation of N files or is some linking / more complicated dependency tree involved? Does anyone know? That is also interesting from the integer throughput/latency perspective. As if it's pure N files, I would expect lower clocked wider cores to lead. Hmm, it seems I have started answering my own question, need to compare the MT/ST results, that should hint at the answer😉

 

[Nothingness]

Semi related side question, does the geekbench clang test consists of pure compilation of N files or is some linking / more complicated dependency tree involved? Does anyone know? That is also interesting from the integer throughput/latency perspective. As if it's pure N files, I would expect lower clocked wider cores to lead. Hmm, it seems I have started answering my own question, need to compare the MT/ST results, that should hint at the answer😉

https://www.geekbench.com/doc/geekbench6-benchmark-internals.pdf

The Clang workload uses the Clang compiler to compile the Lua interpreter, a popular opensource language interpreter. It models the use case of developers building their code and the just-in-time compiling that general users can encounter on their systems (such as JIT compilation for scripting Java and compilation for shading languages in GPU drivers).

This workload uses the musl libc as the C standard library for the compiled files.

The Clang workload compiles 8 files in single-core mode and 96 files in multi-core mode.

Not sure this fully answers your question.

 

[MS_AT]

Not sure this fully answers your question.

Thanks. It definitely brings me closer. But if I counted correctly lua itself has 31 TUs. What means they might be compiling musl libc too in MT test or might be compiling them multiple times. The M5 ratio between MT/ST approaching 6 would point to them compiling the files independently, but M1 Max ratio overtaking 7 seems to somewhat reinforce the idea but I would need to look at wider data set. [I have used the link you have posted previously for comparison].

The assumptions I have made: On Apple silicon the E-cores are not used in the benchmarks (this might be wrong, but I remember someone saying you have to try extra to spill work there, I don't have access to hw myself so unable to verify). And that full actual build will not scale linearly with the number of cores. So the final outcome actually depends on that if my 1st assumption is right or wrong 😉. The second assumption is also brittle on these samples as they are relatively low core cpus and I have no former experience with how well lua build can scale.

 

[Eug]

So, I have a lot of experience with Apple educational products, pricing and promotions. Those deals are usually best if you can buy them through an educational institution like a university bookstore. They'll typically offer better pricing than even Apple directly. And if Apple is focusing this toward education, there may even be special configurations you can't get outside of the educational channel. They don't always have these, but periodically a particularly educationally focused product will get this. I can't imagine it wouldn't qualify for BTS, but it might be something fairly small like some low-end Beats. I'd be surprised if the low price machine has BTO options. My guess is 2 fixed variants. Just making BTO available adds a lot to costs.

Yes, I’ve said before that the $599 base price guesstimate out there only makes sense to me if it is edu institutional pricing, not even edu individual pricing. This could represent an institutional discount alone or a combination of institutional discount plus a unique cut down spec configuration (eg. 8/128 GB).

 

[Nothingness]

Thanks. It definitely brings me closer. But if I counted correctly lua itself has 31 TUs. What means they might be compiling musl libc too in MT test or might be compiling them multiple times.

I count 34 .c files for LUA 5.4.8 🙂 My guess is that they compile a subset of LUA, and launch 12 parallel compilations for that subset. I guess we could reverse-engineer this by looking at (lack of) scaling of Geekbench MP results for various CPUs (but I won't get into this, it'd be time consuming).

The M5 ratio between MT/ST approaching 6 would point to them compiling the files independently, but M1 Max ratio overtaking 7 seems to somewhat reinforce the idea but I would need to look at wider data set. [I have used the link you have posted previously for comparison].

The assumptions I have made: On Apple silicon the E-cores are not used in the benchmarks (this might be wrong, but I remember someone saying you have to try extra to spill work there, I don't have access to hw myself so unable to verify). And that full actual build will not scale linearly with the number of cores. So the final outcome actually depends on that if my 1st assumption is right or wrong 😉. The second assumption is also brittle on these samples as they are relatively low core cpus and I have no former experience with how well lua build can scale.

I don't know if you can exclude any core in macOS. Even affinity setting seems impossible 🙁

 

[MS_AT]

I don't know if you can exclude any core in macOS. Even affinity setting seems impossible

I remember reading that macOS scheduler will be hesitant to use E-cores for anything not specifically marked as background task, but I would need to look around to find the source. Might be I am misremembering something. Unfortunately not having hw on hand makes it hard to verify such claims.

 

[Eug]

I think Apple had planned to release the M5 Pro/Max alongside the base M5 last October, but due to the transition to new SoIC packaging, supply chain delays, and the transition of its PCC servers to M5, it delayed the high-end M5 versions until March.

Still, it's a very optimistic view. You could be right, but I wouldn't want to plan my work plans around that hope. 🙂

However, I think there is a reason to consider waiting, if possible. That reason is not for M6 Max, but for a redesigned chassis along with OLED and touch screen. I don't expect M5 Max MBPs to get those features, but the rumour mill suggests it's a high possibility with the M6 series MBPs. I wonder if they might be priced a bit higher though.

 

[johnsonwax]

I remember reading that macOS scheduler will be hesitant to use E-cores for anything not specifically marked as background task, but I would need to look around to find the source. Might be I am misremembering something. Unfortunately not having hw on hand makes it hard to verify such claims.

Yeah, that's pretty much correct, at least according to Apple's documentation. But background will only ever run on E cores according to Apple, so it's more than hesitant - it shouldn't ever get promoted to P cores. If you spin off your compilation threads as utility instead of user initiated, they should shunt to the E cores if you saturate the P cores. User initiated and user interactive seem to rarely if ever go to E cores. Of course you risk being pre-empted by other foreground processes, so I think that's probably only useful if you are trying to use all cores in a testing environment. Do you want Safari getting priority over a compile?

 

[Doug S]

Why A19 Pro, when A20 Pro will be available?

Apple's repurposing of A series SoCs in non iPhone products like iPad, Apple TV, Cinema Display and so forth seems to serve multiple purposes:

1) it isn't worth doing a special SoC for products that sell in lower numbers, even after you account for wasted chip area (i.e. Apple TV doesn't have cameras so doesn't need the blocks that handle them)
2) they can potentially use chips that aren't acceptable for iPhone but are for different and/or less demanding roles
3) the more chips they run from each generation, the less fixed cost design/mask/etc. overhead is attributable to each product using it
4) they can "use up" overproduction that can't be avoided when you release products by calendar

That's probably why they appear to be using A18 Pro in the low end Macbook, rather than A19 Pro despite that being available and costing basically the same as A18 Pro.

 

[Doug S]

This is what I'm worried about. Apple has a habit of crippling its v1 products, only to (mostly) fix the issues in v2. It would suck if Apple ships a crap-spec'd machine (8 GB RAM and single USB-C port) for that v1 crippled version and for the race to the price bottom. It's already enough of a downgrade to ship the MacBook with smaller 12.9" screen, no True Tone, lower screen brightness, slower SSD, no backlit keyboard, no MagSafe, and no fast charging. With all those downgrades, 12 GB RAM and dual-USB-C should be included, which should be easily doable if the retail price is $749-$799. That price tier would be a good incentive for some to upgrade to the 16 GB $999 MacBook Air instead too, a nice price ladder as Apple likes.

In our case, if they refresh this every two years, v2 will be too late for our initial kid laptop purchase. I guess we'd just have to suck it up and get the MBA instead.

The reason you "cripple" the lower end products is to not only cost less but create a perception of value difference in customers to justify a decision to go to the more expensive option. If they addressed all your complaints and the only difference was the A series SoC, the BOM savings could hardly deliver even $100 lower retail price - and interest in the product would suffer when it wasn't much of a savings / didn't cost much to step up to the MBA (depending on from which way you look at it)

Some of the things you complain about are based entirely on the differences in the A series SoCs versus M series. There is only one USB-C port supported in the A series, so your choices are ship a Macbook containing it with one USB-C port, or add a bridge chip to split it into two ports. Then people would complain the bandwidth is shared between the two, or they'd think the second port should be USB-A not USB-C (that's what I would vote for if they were doing this, YMMV) They probably figure it makes sense that the minority of customers who want to simultaneously plug multiple USB devices into an entry level Mac use a hub, rather than add BOM for the bridge chip and second physical port for everyone. That's not a lot but the cumulative effect of little decisions like that move up into a higher price band.

Same with slower SSD. That's an A series limitation, it has half the bandwidth to the NAND off the SoC versus M series. It isn't "crippled" SSD speed wise anymore than a Macbook Pro with M4 Pro is "crippled" when you compare its memory bandwidth to a Macbook Pro with M4 Max. Just differences in how they design each SoC.

 

[johnsonwax]

Apple's repurposing of A series SoCs in non iPhone products like iPad, Apple TV, Cinema Display and so forth seems to serve multiple purposes:

1) it isn't worth doing a special SoC for products that sell in lower numbers, even after you account for wasted chip area (i.e. Apple TV doesn't have cameras so doesn't need the blocks that handle them)
2) they can potentially use chips that aren't acceptable for iPhone but are for different and/or less demanding roles
3) the more chips they run from each generation, the less fixed cost design/mask/etc. overhead is attributable to each product using it
4) they can "use up" overproduction that can't be avoided when you release products by calendar

That's probably why they appear to be using A18 Pro in the low end Macbook, rather than A19 Pro despite that being available and costing basically the same as A18 Pro.

I mean, if this product was intended to be released some time ago and was delayed, then it wasn't designed for the A19 and they produced A18s for this product. I think that's why the AVP launched with an M2 instead of the newer released processor - it was designed for an M2 and got delayed.

I think all of your points are correct in how Apple does things. The AppleTV choice of processor seems more expedience than technical need, for example. A series are manufactured in such large volume that these smaller product lines only really incur the marginal costs - the large fixed costs were long ago paid for by the iPhone. But in this case I think it was just late and they didn't want to redesign an already delayed product for a new processor.

 

[johnsonwax]

The reason you "cripple" the lower end products is to not only cost less but create a perception of value difference in customers to justify a decision to go to the more expensive option. If they addressed all your complaints and the only difference was the A series SoC, the BOM savings could hardly deliver even $100 lower retail price - and interest in the product would suffer when it wasn't much of a savings / didn't cost much to step up to the MBA (depending on from which way you look at it)

Some of the things you complain about are based entirely on the differences in the A series SoCs versus M series. There is only one USB-C port supported in the A series, so your choices are ship a Macbook containing it with one USB-C port, or add a bridge chip to split it into two ports. Then people would complain the bandwidth is shared between the two, or they'd think the second port should be USB-A not USB-C (that's what I would vote for if they were doing this, YMMV) They probably figure it makes sense that the minority of customers who want to simultaneously plug multiple USB devices into an entry level Mac use a hub, rather than add BOM for the bridge chip and second physical port for everyone. That's not a lot but the cumulative effect of little decisions like that move up into a higher price band.

Same with slower SSD. That's an A series limitation, it has half the bandwidth to the NAND off the SoC versus M series. It isn't "crippled" SSD speed wise anymore than a Macbook Pro with M4 Pro is "crippled" when you compare its memory bandwidth to a Macbook Pro with M4 Max. Just differences in how they design each SoC.

Apple though is very cognizant of upsell as you note with 'perception of value difference'. I know this from discussions with engineers who designed products for Apple when presenting value options to management, management would have very clear ideas about how these products should be positioned and would generally choose with that in mind. I don't think it's so much that they nerf the product, as they might put a value option in favor of reliability over a bullet feature that consumers might like. To the consumer looking for that bullet feature it's a nerf, but there's still a value created there.

 

[Eug]

The reason you "cripple" the lower end products is to not only cost less but create a perception of value difference in customers to justify a decision to go to the more expensive option. If they addressed all your complaints and the only difference was the A series SoC, the BOM savings could hardly deliver even $100 lower retail price - and interest in the product would suffer when it wasn't much of a savings / didn't cost much to step up to the MBA (depending on from which way you look at it)

Some of the things you complain about are based entirely on the differences in the A series SoCs versus M series. There is only one USB-C port supported in the A series, so your choices are ship a Macbook containing it with one USB-C port, or add a bridge chip to split it into two ports. Then people would complain the bandwidth is shared between the two, or they'd think the second port should be USB-A not USB-C (that's what I would vote for if they were doing this, YMMV) They probably figure it makes sense that the minority of customers who want to simultaneously plug multiple USB devices into an entry level Mac use a hub, rather than add BOM for the bridge chip and second physical port for everyone. That's not a lot but the cumulative effect of little decisions like that move up into a higher price band.

Same with slower SSD. That's an A series limitation, it has half the bandwidth to the NAND off the SoC versus M series. It isn't "crippled" SSD speed wise anymore than a Macbook Pro with M4 Pro is "crippled" when you compare its memory bandwidth to a Macbook Pro with M4 Max. Just differences in how they design each SoC.

? Maybe you misunderstood me, but I think it's perfectly acceptable for the lower end A series MacBook to get a smaller 12.9" screen, no True Tone, lower screen brightness, slower SSD, no backlit keyboard, no MagSafe, and no fast charging. That's already a huge downgrade from the MacBook Air, which would justify the cheaper pricing with Apple's tiered product laddering, and I'm fine with all of that.

However, I think any Mac laptop should have two USB-C ports, even if it's shared bandwidth. As previously mentioned, it's not really about multiple USB devices, it's about having a single USB device at the same time as charging the machine, given that it likely won't have MagSafe. And no it should not be USB-A, and it can't be USB-A anyway if it's going to be USB charging. I suppose it would be OK if Apple implemented a separate charging port from a single USB-C port, but somehow I doubt that. This is the number one beef I had with the 2015-2017 MacBooks. They actually had two ports, but "wasted" one of those ports on a headphone jack, leaving only a single USB-C port for both charging and peripherals, and it was a incredible PITA.

 

[Doug S]

? Maybe you misunderstood me, but I think it's perfectly acceptable for the lower end A series MacBook to get a smaller 12.9" screen, no True Tone, lower screen brightness, slower SSD, no backlit keyboard, no MagSafe, and no fast charging. That's already a huge downgrade from the MacBook Air, which would justify the cheaper pricing with Apple's tiered product laddering, and I'm fine with all of that.

However, I think any Mac laptop should have two USB-C ports, even if it's shared bandwidth. As previously mentioned, it's not really about multiple USB devices, it's about having a single USB device at the same time as charging the machine, given that it likely won't have MagSafe. And no it should not be USB-A, and it can't be USB-A anyway if it's going to be USB charging. I suppose it would be OK if Apple implemented a separate charging port from a single USB-C port, but somehow I doubt that. This is the number one beef I had with the 2015-2017 MacBooks. They actually had two ports, but "wasted" one of those ports on a headphone jack, leaving only a single USB-C port for both charging and peripherals, and it was a incredible PITA.

The battery life on MBA is good enough (and I would expect this to have similar battery life, not quite as long but well past "all day use" for non pathological workloads i.e. not Cinebench in a loop) that using the laptop while charging would be the exception rather than the rule.

What do you envision people buying an entry level Mac will plug in to it?

 

[Eug]

The battery life on MBA is good enough (and I would expect this to have similar battery life, not quite as long but well past "all day use" for non pathological workloads i.e. not Cinebench in a loop) that using the laptop while charging would be the exception rather than the rule.

What do you envision people buying an entry level Mac will plug in to it?

Eg. Even just for students: HDMI dongle for presentations. HDMI or DisplayPort monitor in the dorm room. USB drive for external file storage. Etc. None of the these necessarily requires major CPU or GPU performance, just available ports. Are there workarounds? Yes, sure for most of the these things, but I can tell you from first hand experience it's a royal pain. All-day battery life is not all day if you've forgotten to charge it or you've been using it much of the day for last minute edits or reviews before a one hour presentation. This would be solved if there is a single USB-C port paired with a MagSafe port, but people are saying it likely won't get MagSafe. Anyhow, I'd be happy if it had MagSafe + USB-C or else two USB-C, the latter preferred. And minimum 12 GB RAM.

 

[johnsonwax]

The battery life on MBA is good enough (and I would expect this to have similar battery life, not quite as long but well past "all day use" for non pathological workloads i.e. not Cinebench in a loop) that using the laptop while charging would be the exception rather than the rule.

What do you envision people buying an entry level Mac will plug in to it?

So, I will say from an education usage standpoint, which Eug kind of suggests because it's for their kid, quite a lot depending on their major.

College students (the context in which I best understand this stuff as I created laptop programs at a university and worked with lab design, software licensing, curriculum development, etc.) probably will have a fair bit of need to plug in the laptop - as it's not uncommon for students to come to campus early and stay late before returning home. You have mice and whatnot that can be BT, but there's a lot of USB drive sharing with other students (if you want to print something, that's almost always from a USB drive, not a direct connection), plugging into monitors/projectors, in technical fields hooking up equipment, etc. Yes, even on Macs because a lot of education-focused data acquisition tools are USB based and have Mac support. Now, you start to run in the problem there where the more of that you're doing, the less suitable a low-end A series Mac or MBA will be but what we found was that a lot of students heading off to college don't really know what they're going to be doing there, go in with a cheap laptop, and after a year have a better sense of their needs and move up to something more suitable - or not, if their cheap laptop fills that need.

We would never put a single USB port laptop on our recommended list. And we did tell Apple this when they offered a single USB port laptop.

You kind of need to think about the institutional security challenge that is a high school or a university and consider it from that angle. There's a shit-ton of things you just can't do wirelessly either because of a security problem, or because of an addressing problem. You have a campus wide wireless network serving 40,000 people. There are probably 1,000 printers there. You can't have them advertising on that network, and you can't really ask users to shift their SSID for each room they might find themselves in - again because you will have hundreds or thousands of SSIDs. You can't plop 300 AppleTVs on there all broadcasting so Airplay is out of the question except in secure networks. A lot, a lot, a lot of stuff has to fall back to wired simply to address these problems.

 

[MerryCherry]

Caches are misleading in terms of transistor count and area. They are not only super-dense and error-resistant due to their extreme redundancy, they also lower the overall power of the SoC thus is worth adding versus logic, which isn't always the case. So this isn't necessarily accurate analysis either. There's a saying "before you expand your uarch, consider if it'll be worth adding equivalent amount in caches instead". When they used to breakdown area, transistor and power of the large server cores years ago in Hot Chips, L3 caches would take up 60% of the space but use 10% of the whole power or less. This is ignoring the elephant in the room where cache access would mean less access in DRAM, meaning overall you LOWER power. I don't know where people got the idea caches are power or area inefficient. It is a "dumb" way of adding performance.

When M1 debuted, it's huge caches were one of the factors attributed to it's superior performance and efficiency.

Fast forward to today, and the situation has flipped. Intel,AMD have more cache per core, bur are still getting creamed by Apple.

Seems Apple figured out 'smarter' ways to increase perf?

 

[poke01]

When M1 debuted, it's huge caches were one of the factors attributed to it's superior performance and efficiency.

Fast forward to today, and the situation has flipped. Intel,AMD have more cache per core, bur are still getting creamed by Apple.

Seems Apple figured out 'smarter' ways to increase perf?

because Apple's performance is not only from their cores being great but their uncore being industry leading.

 

[Gerard Williams]

because Apple's performance is not only from their cores being great but their uncore being industry leading.

Your memory controller needs to be very good.
Your latency needs to be very good.
Your bandwidth to memory excellent.
Your caching hierarchy very very good.
Outstanding misses must be large.
Etc etc. its not one thing but a million ones which add up.

 

[511]

Your memory controller needs to be very good.
Your latency needs to be very good.
Your bandwidth to memory excellent.
Your caching hierarchy very very good.
Outstanding misses must be large.
Etc etc. its not one thing but a million ones which add up.

basically if any single part screws up you are screwed i guess also are you the real GWIII?