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

 

[DZero]

Poor Intel, Apple just destroyed hard. Panther Lake should be a wonder if not, Apple would make a massive check mate for them.

 

[poke01]

Poor Intel, Apple just destroyed hard. Panther Lake should be a wonder if not, Apple would make a massive check mate for them.

The GPU in Panther lake is interesting at least

 

[S'renne]

The GPU in Panther lake is interesting at least

So is the GPU in M5

 

[jdubs03]

It’s odd based on this it makes it seem like the A19 Pro p-core, stronger relative to the M5 p-core.

But the geekbench scores tell the opposite story.

 

[poke01]

So is the GPU in M5

Yeah it is but gotta have something to look forward to for Panther lake lol

 

[Doug S]

It’s odd based on this it makes it seem like the A19 Pro p-core, stronger relative to the M5 p-core.

But the geekbench scores tell the opposite story.

Did he break out the individual SPEC tests? When I looked at GB6's LLVM test it confirmed to me the gain was "real", rather than being an artifact of a few benchmarks show larger gains (like a better SME unit or GB tests that overly benefit from memory bandwidth)

So it would be interesting to compare with SPEC's gcc test, which I'd expect would show a bigger gain than the overall SPEC results. If so my next question would be "OK so what test(s) were dragging down the overall SPEC number, and can we figure out why?" If not then I'd wonder what changed that caused a smaller gain for that type of code on M5?

One possible explanation is @name99's - that Apple had something new like a trace cache chicken bitted off in A19P but enabled in M5. It could help some stuff but have less benefit (or even hurt) others, but if they felt overall it benefited what they considered to be "more important" they'd enable it. It might be something that's not necessarily chicken bitted on/off but has been evolving/improving as the cores roll out. Who knows maybe the P cores in M5P/M5M are also different in what gets faster by how much...

 

[LightningZ71]

It might just be core feature and cache system tuning. There's nothing that says that the XTORS on the A19 cores must be exactly the same as the ones in the M series since they are different dies. Just like AMD did with Strix/Kraken point with having a few AVX512 data paths as compared to desktop CCD implementations of the Zen 5 cores, Apple MAY have changed certain structures between the two series to better fit their targets. A19 could be less aggressive in speculative execution or prefetching data to save power which gives artifacts in the benchmark data.

 

[mvprod123]

 

[mvprod123]

 

[mvprod123]

 

[mvprod123]

Massive improvements in Blender

 

[The Hardcard]

View attachment 132413
It’s odd based on this it makes it seem like the A19 Pro p-core, stronger relative to the M5 p-core.

But the geekbench scores tell the opposite story.

Given that SPEC takes hours to run, doesn’t thermal throttling make more of an impact? You are comparing a sustained benchmark versus a peak frequency number that wasn’t there for most of the run.

The difference between the iPhone heat dissipations is greater - with the added vapor chamber against aluminum - than the M5 devices. What were the clockspeeds near the end of the subtests?

Unless you can get a stable extended run frequency for each device and then use those frequency numbers, I don’t see how you use those numbers for IPC.

In addition to heat dissipation you also have to take core swapping into account, another thermal protection. iPhone cores have less time to cool down with only 2 versus the iPad and Macs 4 P cores, not to mention that M-series P cores are on both sides of the cache so it is possible to swap to a core less affected by the current core’s heat.

 

[mvprod123]

https://twitter.com/x/status/1980680483676582384

 

[jpiniero]

Safari is snappier?

 

[mvprod123]

Safari is snappier?

https://twitter.com/x/status/1981014969501331611

 

[Eug]

China Times is claiming initial TSMC N2 chips will cost 50% more than current N3.

https://www.chinatimes.com/newspapers/20250922000510-260110

-----

Somebody has posted iPad Pro M5 vs M4 benches at room temp and on ice.

Got my iPad Pro M5 (4P+6e)

Unlike the M4 model, the Geekbench 6 single thread score doesn't increase much with additional cooling. Too early to make any judgements, but it seems like M5 can hit near peak single-thread performance at normal temperatures unlike M4. Curious how it will hold up when it heats up.

M5 (4P+6e):
Idle/room temperature 4117 / 15845 https://browser.geekbench.com/v6/cpu/14596883
On ice 🥶 4179 (101.5%) / 16876 (106.5%) https://browser.geekbench.com/v6/cpu/14600126

M4 (4P+6e):
Idle/room temperature 3698 / 14739 https://browser.geekbench.com/v6/cpu/14600365
On ice 🥶 3975 (107.5%) / 15039 (102.0%) https://browser.geekbench.com/v6/cpu/14600606

 

[mvprod123]

Has Apple updated the media engines in the M5?

 

[Eug]

Has Apple updated the media engines in the M5?

View attachment 132442

Did they specify in the review what the source material for the export was? If it's a complex edit, there may be a fair amount of CPU and GPU involved.

 

[name99]

https://twitter.com/x/status/1981014969501331611

That's interesting because it appears to be so much larger than the M4 - 18 to 35% improvement, depending on which random reddit post you believe; the most recent comparable value I can find is
(25 days ago)"
Not sure.
Running Speedometer 3.1 I get 44.1(-+1.7) on an M4 MacBook Pro Safari Version 26.0 on Tahoe.
"

This would suggest that my trace cache theory is correct (at least in the sense that it is present in the M5). Might be interesting to get some Speedometer 3.1 scores for A19 Pro vs A18 Pro...

Both SPEC and GB6 do very little to stress the front end of an Apple core, while browsers do a somewhat better job of this.

 

[name99]

China Times is claiming initial TSMC N2 chips will cost 50% more than current N3.

https://www.chinatimes.com/newspapers/20250922000510-260110

They've been posting this sort of thing since at least 16nm. I'm not sure what the point is any more. At this stage it feels like a stock manipulation scheme more than any sort of attempt to actually inform the public about something important...

 

[smalM]

Given that SPEC takes hours to run.

A14 Firestorm 33:36
A14 Icestorm 1:57:28
A15 Avalanche 29:42
A15 Blizzard 1:28:58
Andrei used a cooler when running spec on an iPhone.
M5 P-core should be at ~¼h and M5 E-Core at ¾h.

 

[The Hardcard]

A14 Firestorm 33:36
A14 Icestorm 1:57:28
A15 Avalanche 29:42
A15 Blizzard 1:28:58
Andrei used a cooler when running spec on an iPhone.
M5 P-core should be at ~¼h and M5 E-Core at ¾h.

Well I do remember reading that many years ago. Cores have come a long way since Phenom. Still, 15 minutes is enough a core to get hot and for the greater difference in iPhone cooling to give an appearance of greater IPC increase measured against a stationary peak frequency.

 

[LightningZ71]

Do Air 1t numbers suffer as compared to the pro max? If they are broadly similar, I don't think that thermal saturation is hurting 1t IPC calculations any.

 

[Eug]

They've been posting this sort of thing since at least 16nm. I'm not sure what the point is any more. At this stage it feels like a stock manipulation scheme more than any sort of attempt to actually inform the public about something important...

Hmmm... By most accounts, N3B was indeed quite expensive. Are you trying to deny this was the case?

I mean, a common refrain here was that a big part of M3's short life was due to N3B's cost.

 

[pj-]

It's funny to read this thread and then get an m5 ipad and it's essentially indistinguishable from my m4 ipad (handing down to a family member, I'm not an insane person upgrading for no reason)

The wifi is much better, and the ram bump means apps will be swapped out of memory less often, but the cpu/gpu/npu improvements are basically meaningless, other than if the battery life is better from increased efficiency. Not sure yet if that'll be an appreciable difference