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

 

[kingsleyopara]

120GB/s (up from 100GB/s) memory bandwidth - so are we thinking LPDDR5X? Should we expect a similar uplift for the M4 Pro and Max?

 

[Mopetar]

If some models only have 3 performance cores, I'm curious if there will be cases where those are beat out by an older M2 model that has 4 performance cores.

It'll still have a bump from going from M2 cores to whatever the M4 is using (I'm thinking it's the same cores as M3 with some minor tweaks) and perhaps a clock speed bump as well, but I doubt that it makes up for the loss of a core. Sure it'll have an extra efficiency core to cover, but that's not going to be quite as good.

 

[gregoritsch]

with M3 being about 20% faster than M2 and now the claim that M4 is up to 1.5x the performance of M2, we are looking at 25% increase over the M3.
Personally I'd think, that most of those improvements are a combination of faster RAM, a small improvement due to the N3P node, as well as the 2 extra efficiency cores.

Will be interesting to see deeper analysis about clock speed and other changes, once the devices are out

 

[Eug]

So how big is this M4 chip now? 160-170 mm2?

If some models only have 3 performance cores, I'm curious if there will be cases where those are beat out by an older M2 model that has 4 performance cores.

It'll still have a bump from going from M2 cores to whatever the M4 is using (I'm thinking it's the same cores as M3 with some minor tweaks) and perhaps a clock speed bump as well, but I doubt that it makes up for the loss of a core. Sure it'll have an extra efficiency core to cover, but that's not going to be quite as good.

I betcha all the review units are 1 TB with 10-core CPU and 16 GB RAM. So all the leaked Geekbench results will reflect that.

 

[repoman27]

120GB/s (up from 100GB/s) memory bandwidth - so are we thinking LPDDR5X? Should we expect a similar uplift for the M4 Pro and Max?

LPDDR5X-7500, so they stopped short of the LPDDR5X-8533 that I had predicted. The M4 Pro and Max will certainly follow suit, but could potentially go higher than the M4.

If some models only have 3 performance cores, I'm curious if there will be cases where those are beat out by an older M2 model that has 4 performance cores.

It'll still have a bump from going from M2 cores to whatever the M4 is using (I'm thinking it's the same cores as M3 with some minor tweaks) and perhaps a clock speed bump as well, but I doubt that it makes up for the loss of a core. Sure it'll have an extra efficiency core to cover, but that's not going to be quite as good.

Disabling a p-core makes no difference to single-threaded performance, and two additional e-cores plus the power and performance benefits of N3E are going to more than make up for the difference in multi-threaded scenarios. I don't see how the binned M4 would ever lose to the M3. And if it's only ever in an iPad, there will never be a real comparison anyway.

 

[Mopetar]

with M3 being about 20% faster than M2 and now the claim that M4 is up to 1.5x the performance of M2, we are looking at 25% increase over the M3.
Personally I'd think, that most of those improvements are a combination of faster RAM, a small improvement due to the N3P node, as well as the 2 extra efficiency cores.

Will be interesting to see deeper analysis about clock speed and other changes, once the devices are out

I think you're making some improper assumptions. Apple has made no such claims. What they've claimed is that the new iPad is 50% faster than the previous generation iPad in a specific benchmark.

Some of that performance gain is coming from the two addition efficiency cores that the new M4 iPad has.

I expect the clock speeds to be higher, but Apple is probably binning more aggressively which is why they have a lot of 3-core M4 chips. Until further analysis or someone doing a deep dive on the M4, I'm going to assume they're using essentially the same core as for M3 and we get negligible IPC gains if any at all.

 

[Eug]

This was expected but I’m glad to see confirmation of hardware AV1 support. That was my main interest with M4 in the iPad. Otherwise most of the other benefits of M4 are lost on me. This is especially true since I’d get the 9-core variant of M4 anyway.

Edit: Oh yeah M3 already had it.

 

[repoman27]

with M3 being about 20% faster than M2 and now the claim that M4 is up to 1.5x the performance of M2, we are looking at 25% increase over the M3.
Personally I'd think, that most of those improvements are a combination of faster RAM, a small improvement due to the N3P node, as well as the 2 extra efficiency cores.

Will be interesting to see deeper analysis about clock speed and other changes, once the devices are out

The M4 is almost certainly based on the A17 Pro and uses the same Everest and Sawtooth CPU cores, which were also in the A16 Bionic and M3. The GPU core is probably the same as the A17 Pro and M3. It looks like that was one of the few things that got pulled forward for the M3 or bumped to the A17 Pro, depending on how you look at it. The M4 also gets the newer neural engine with INT8 support from the A17 Pro. That was always odd that the M3 NPU was only billed as 18 TOPS while the A17 Pro was at 35 TOPS. Now the M4 finally retakes the crown with 38 TOPS.

And I think you meant N3E, which provides +5% speed at ISO power or -7% power at ISO speed over base N3 (a.k.a. N3B).

 

[repoman27]

This was expected but I’m glad to see confirmation of hardware AV1 support. That was my main interest with M4. Otherwise most of the other benefits of M4 are lost on me.

M3 had AV1 decode as well, I think. Although it was never in an iPad.

 

[repoman27]

Oh, and iPad Pro orders immediately after event, available next week on Wed 5/15. So this was a proper launch for the M4.

 

[SpudLobby]

True, it looks a clock speed increase again. That’s not bad cause they are using a different node.

At this point I will stop expecting IPC improvements from Apple and when they come they will come.

I don’t even know if they have a big clockspeed increase for the big core. N3E over N3B shouldn’t provide much either (iso-power). E cores I could see

 

[Glo.]

Wait, so Apple did NOT move to LPDDR6(X)?

Who could have predicted that?!

 

[repoman27]

I think you're making some improper assumptions. Apple has made no such claims. What they've claimed is that the new iPad is 50% faster than the previous generation iPad in a specific benchmark.

Some of that performance gain is coming from the two addition efficiency cores that the new M4 iPad has.

I expect the clock speeds to be higher, but Apple is probably binning more aggressively which is why they have a lot of 3-core M4 chips. Until further analysis or someone doing a deep dive on the M4, I'm going to assume they're using essentially the same core as for M3 and we get negligible IPC gains if any at all.

Just a quick note on binning. The M3 generation saw the heaviest binning to date by Apple, which was completely to be expected given the situation with the N3 node. The binned version of the M3 had 2 out of 10 (1/5) of the GPU cores disabled, which is a lot more die area than a single performance CPU core.

 

[mikegg]

The M4 is almost certainly based on the A17 Pro and uses the same Everest and Sawtooth CPU cores, which were also in the A16 Bionic and M3. The GPU core is probably the same as the A17 Pro and M3. It looks like that was one of the few things that got pulled forward for the M3 or bumped to the A17 Pro, depending on how you look at it. The M4 also gets the newer neural engine with INT8 support from the A17 Pro. That was always odd that the M3 NPU was only billed as 18 TOPS while the A17 Pro was at 35 TOPS. Now the M4 finally retakes the crown with 38 TOPS.

And I think you meant N3E, which provides +5% speed at ISO power or -7% power at ISO speed over base N3 (a.k.a. N3B).

What do you make of the claim that M4 uses 50% of the power of M2 at the same speed as M2?

 

[Eug]

Oh, and iPad Pro orders immediately after event, available next week on Wed 5/15. So this was a proper launch for the M4.

It seems like they have plenty of stock too. I chose a couple of different configurations, and they are still delivering for May 15. This is much different from a typical iPhone launch.

I won't be ordering until June though. Usually each year both the iPad Pro and iPad Air are included in the education promotion for say US$100 back as a gift card.

What do you make of the claim that M4 uses 50% of the power of M2 at the same speed as M2?

That was Apple's claim, although we don't actually know what that means.

 

[Mopetar]

I don’t even know if they have a big clockspeed increase for the big core. N3E over N3B shouldn’t provide much either (iso-power). E cores I could see

It's not the node that does it, but more aggressive binning. Yields are not so bad that Apple has to disable a performance core for the base models. Rather they have more aggressive performance targets and they disable the weakest core if all four can't hit the mark.

Most people won't buy the models that have all 4 cores enabled. Maybe it's 20% of their sales. That lets them be a lot stingier with their performance targets. They can pick a number that 95% (or more) chips can hit with 3 cores that only the top 20% can hit on all four cores.

 

[Eug]

It's not the node that does it, but more aggressive binning. Yields are not so bad that Apple has to disable a performance core for the base models. Rather they have more aggressive performance targets and they disable the weakest core if all four can't hit the mark.

Most people won't buy the models that have all 4 cores enabled. Maybe it's 20% of their sales. That lets them be a lot stingier with their performance targets. They can pick a number that 95% (or more) chips can hit with 3 cores that only the top 20% can hit on all four cores.

Yes, it's interesting that likely >2/3rds of the iPad Pros out there will have the binned 3P variant.

I wonder what this means for an MacBook Air M4. Will the MacBook Air's M4 be binned, and if so, will it be at higher clocks? Or will the MacBook Air skip the M4 entirely? From a technical standpoint, it could make sense for the MBA to skip the M4 refresh, but then again, the MBA 13" and 15" are Apple's best selling laptops, so a refresh could be beneficial from the marketecture point of view.

 

[Muadib]

So are you guys saying that the new iPad Pro is a buy? I saw the thing today, and was quite tempted, because the iPad I currently have is from 2018, and I could use an upgrade.

 

[SpudLobby]

It's not the node that does it, but more aggressive binning.

No dude N3E has better yields and electrics. I don’t even know what you’re on. I’m talking about small (if present) power/performance gains for each core.

 

[Doug S]

Based on the 'best' scores in GB6 ST Apple gained about 15% from M2 to M3. Apple E cores tend to be worth about 1/3 of a P core, so an 8 core M2 had 5.3 cores vs 10 core M4 having 6 cores, accounting for 12%. From N3E TSMC says you should get 6-7%. If you multiply all that together (can't add it, since they are cumulative) you get 38%. They need to gain another 9% from other sources to achieve a 50% overall gain.

There is more memory bandwidth, that could help some MT tests that are memory bound but another mid single digit ST gain seems likely. That doesn't have to mean a new core, it could be essentially the same core with some minor reworking to improve cache size/latency, or some tweaking with FinFlex to use higher power/faster transistors in a few chosen places that are restricting the ability to up the clock a bit.

Another possibility is that the E cores saw some significant performance gains while the P cores remained pretty much flat, but it is harder to track down that information (it would be nice if Geekbench had a setting where it ran benchmarks pinned to a CPU's smaller cores...have to remember to suggest this next time John pops up on RWT)

The short turnaround between M3 and M4 makes it less likely we'd see any major changes. That becomes even more true if this "early" release of M4 was a one off and M5 is back to the fall. I think Apple has good reasons to want Apple Silicon to hit earlier in the year in general, but we'll have to see what happens next year to know.

 

[Mopetar]

No dude N3E has better yields and electrics. I don’t even know what you’re on. I’m talking about small (if present) power/performance gains for each core.

That's my point. N3E had better yields, hence Apple has no reason to sell you an iPad with only 3 performance cores when the amount of area they take up is such a small part of the overall SoC.

The only other way that you get to three cores outside of Apple fusing one off just to be evil, is that they require each of those cores to be able to hit a certain clock speed within a given voltage.

Suppose that each core independently has a 70% chance of hitting this targets. That means that there's only a 25% chance that all of them can do it. There's the binning target if Apple thinks 25% of sales will be for that variant. Meanwhile the base product gets the next bin that the vast majority of silicon can hit.

 

[SpudLobby]

Anyways I think it’s possible they get some 5% ST gain from N3E frequency but I’d be more impressed if they had an Arch IPC tweak for 5% or so and if power dropped down too.

Do we think it’s wider than M3, or are they comparing against the M2? That’s really my question here. Did they just mention “wider” again and it’s because they’re comparing against M2 iPads?

Seems *very* likely.

I think it’s funny how often we keep talking about the phantom big Apple core upgrade ever since A14/M1. I am hardly partial to AMD/Intel camps that underestimate how far Apple is ahead when it comes to ST performance/W curves — it’s still massive — and AMD/Intel can’t even match the X Elite (they get blown out, and yes Phoenix on N4 too) — not to mention the E core advantages.

However I find it grating the Apple camp really believed that by the time of M4 they’d have the same margin (or close ish) of IPC and performance/power gains that they did 4 years ago.

That hasn’t happened at all really. It shows no sign of coming and it’s kinda ridiculous to believe they’re going to pull it out of their hat, I think they’ve lost too much talent for right now.

 

[Doug S]

No dude N3E has better yields and electrics. I don’t even know what you’re on. I’m talking about small (if present) power/performance gains for each core.

Yeah the node that made binning complex for Apple was N3B because it is so bad. N3E is back to normal TSMC yields, where Apple's "bin on defects in large structures" strategy to gain another few percent of usable dies makes sense.

It is interesting that they switched from binning on defective GPU cores to defective P cores this time around, since the GPU is overall a larger structure and thus more likely to have a defect.

 

[SpudLobby]

That's my point. N3E had better yields, hence Apple has no reason to sell you an iPad with only 3 performance cores when the amount of area they take up is such a small part of the overall SoC.

The only other way that you get to three cores outside of Apple fusing one off just to be evil, is that they require each of those cores to be able to hit a certain clock speed within a given voltage.

Suppose that each core independently has a 70% chance of hitting this targets. That means that there's only a 25% chance that all of them can do it. There's the binning target if Apple thinks 25% of sales will be for that variant. Meanwhile the base product gets the next bin that the vast majority of silicon can hit.

They never did sell you one with 3 performance cores.

I’m aware they bin but a lot of that is actually just segmentation — Apple doesn’t clockspeed bin for the whole CPU. They bin by one or two extra gpu cores or one extra cpu cores now. That’s more artificial.

 

[SpudLobby]

Yeah the node that made binning complex for Apple was N3B because it is so bad. N3E is back to normal TSMC yields, where Apple's "bin on defects in large structures" strategy to gain another few percent of usable dies makes sense.

It is interesting that they switched from binning on defective GPU cores to defective P cores this time around, since the GPU is overall a larger structure and thus more likely to have a defect.

Yeah that makes me think it’s basically artificial and more about segmentation here. The odds the one E core is that big of a deal yield wise is really low I think.