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

 

[smalM]

Blender score (62% better than M4)


View attachment 132185

That's the added scores of the RX 5700 and the RX 5700 XT I have in my two old Mac Pros

 

[CouncilorIrissa]

Is the cache larger or the memory bandwidth higher for the M5? If the underlying cores are identical then the obvious place to look first is the memory subsystem.

I think your assessment for #2 makes sense and is the most likely explanation.

Memory bandwidth probably is a factor. With M4 (with Pro/Max at the very least), a single cluster is able to pull ~220 GB/s from memory (and a single core around ~92 GB/s). For M5 Pro, this would mean that a single core can pull more read b/w than the single-channel setup of A19P can provide (9600 MT/s config results in 75 GB/s of memory bandwidth).

https://twitter.com/x/status/1860317455429828936

https://twitter.com/x/status/1860319640259559444

 

[Eug]

At this point, the performance is already more than I need. I've actually been more interested in I/O support in the last little while.

M1 Mac mini
- Thunderbolt 3
- DisplayPort 1.4
- HDMI 2.0

M2 Mac mini
- Thunderbolt 4
- DisplayPort 1.4
- HDMI 2.0

M4 Mac mini
- Thunderbolt 4
- DisplayPort 1.4
- HDMI 2.1

Thunderbolt 3 - 40 Gbps data but can't support true Thunderbolt hubs
Thunderbolt 4 - Allows the use of true Thunderbolt hubs

DisplayPort 1.4 - Can support ultra high resolution displays but needs DSC
DisplayPort 2.1 - Supports ultra high resolution displays without DSC

HDMI 2.0 - Maxes out at 4K 4:4:4 HDR at 30 Hz or 4K 4:4:4 SDR at 60 Hz
HDMI 2.1 - Supports ultra high resolution HDR displays

The M4 Mac mini was a great release, with cheaper pricing, more base RAM, nicer form factor, an increase to three Thunderbolt 4 ports, 2 additional front USB-C ports, and HDMI 2.1.

I was predicting we wouldn't get a meaningful upgrade in I/O features beyond what's in the M4 until maybe M7 or something. (I'm ignoring the Mx Pros.)

?? M7 Mac mini
- Thunderbolt 5
- DisplayPort 2.1
- HDMI 2.1

An M7 Mac mini in say 2029 with Thunderbolt 5 and DisplayPort 2.1 would be an absolute monster of a machine. I'm thinking it will be even longer until the base Mx Mac mini gets HDMI 2.2, but HDMI 2.2 is not high up on my wishlist for the foreseeable future.

 

[mvprod123]

At this point, the performance is already more than I need. I've actually been more interested in I/O support in the last little while.

M1 Mac mini
- Thunderbolt 3
- DisplayPort 1.4
- HDMI 2.0

M2 Mac mini
- Thunderbolt 4
- DisplayPort 1.4
- HDMI 2.0

M4 Mac mini
- Thunderbolt 4
- DisplayPort 1.4
- HDMI 2.1

Thunderbolt 4 - Allows the use of Thunderbolt hubs
HDMI 2.1 - Supports ultra high resolution HDR displays

The M4 Mac mini was a great release, with cheaper pricing, more base RAM, nicer form factor, an increase to three Thunderbolt 4 ports, 2 additional front USB-C ports, and HDMI 2.1.

I was predicting we wouldn't get a meaningful upgrade in I/O features beyond what's in the M4 until maybe M7 or something. (I'm ignoring the Mx Pros.)

?? M7 Mac mini
- Thunderbolt 5
- DisplayPort 2.1
- HDMI 2.1

An M7 Mac mini in say 2029 with Thunderbolt 5 and DisplayPort 2.1 would be an absolute monster of a machine. I'm thinking it will be even longer until the base Mx Mac mini gets HDMI 2.2, but HDMI 2.2 is not high up on my wishlist for the foreseeable future.

I believe that TB5 will already appear in M6, and by 2029, we will get TB6.

 

[Eug]

I believe that TB5 will already appear in M6, and by 2029, we will get TB6.

That seems very optimistic. The Mac mini first got Thunderbolt 3 in 2018, and then moved one generation to Thunderbolt 4 in 2023, five years later. You're expecting to move two generations to Thunderbolt 6 in six years.

Furthermore, Thunderbolt 4 didn't even increase the data rate over TB3.

 

[DZero]

Meanwhile in China...

iPhone Air Vanishes in Minutes: Sells Out Instantly in China - Gizmochina

iPhone Air sells out within minutes in China as Apple sparks massive launch frenzy and strengthens its market presence.

www.gizmochina.com

So, iPhone Air might survive at all, but might become limited then.

 

[name99]

So now I have two questions about M5:

1) why does it have a 5-6% higher IPC than A19P? What is the difference in the cores which accounts for that?
2) why is it running at a higher clock in Macbook Pro than iPad Pro?

For #2 they may be binning on frequency to put the faster ones in MBP, or binning on power and as well sacrificing a bit of top end clock to manage heat and battery life in the iPad Pro's restrictive form factor. Though I suppose they could do the clock sacrifice to meet power goals without needing to bin.

We'll probably never know or not know for a long time.
Assuming the IPC gap is real and not just a stats artifact, I don't think extra memory bandwidth or larger cache are relevant; they're nice but minor effects.

My best guess remains what I said earlier - one of the large changes for A19/M5 (my guess would be the trace cache) was *almost* ready, but not quite in place to make A19 schedule, so was disabled for A19, whereas with an extra month (maybe even two given the smaller volumes) there was time to fix that last bug before M5 shipped.

 

[name99]

Memory bandwidth probably is a factor. With M4 (with Pro/Max at the very least), a single cluster is able to pull ~220 GB/s from memory (and a single core around ~92 GB/s). For M5 Pro, this would mean that a single core can pull more read b/w than the single-channel setup of A19P can provide (9600 MT/s config results in 75 GB/s of memory bandwidth).

https://twitter.com/x/status/1860317455429828936

https://twitter.com/x/status/1860319640259559444

View attachment 132214

His theory for why cluster bandwidth went up is wrong.
https://x.com/handleym99/status/1979606689952477289

Seems like the bandwidth of the basic NoC link is 256GB/s (what matters is the connection to the SLC which will run at max speed, beyond that you are throttled by DRAM). So two of the get you 512.

I'm guessing the E-core uses a quarter width link, to save both power and a little area, hence its 128. Probably some of the slower IO blocks do the same?

 

[jdubs03]

Up to 4355 ST. I’ve seen as high as 18095 MT.

Mac17,2 - Geekbench

Benchmark results for a Mac17,2 with an Apple M5 processor.

browser.geekbench.com

 

[mvprod123]

M5 iPad Pro
SNL Score

 

[Eug]

iPad Pro 11" M4 vs 13" M5 benches

3DMark Wild Life Extreme



3D Mark Solar Bay Extreme (ray tracing)



DaVinci Resolve (4K 12 minute video export)



Draw Things (local AI image generation)



I didn't bother posting the several Geekbench results

 

[mvprod123]

iPad Pro 11" M4 vs 13" M5 benches

3DMark Wild Life Extreme

View attachment 132364

3D Mark Solar Bay Extreme (ray tracing)

View attachment 132365

DaVinci Resolve (4K 12 minute video export)

View attachment 132366

Draw Things (local AI image generation)

View attachment 132367

I didn't bother posting the several Geekbench results

Well, I'm waiting for the results of testing the M5 MacBook Pro. Passive cooling in the iPad does not reveal the full potential of the chip.

 

[mvprod123]

200 pt in Cinebench 2024 ST!

 

[mvprod123]

Cinebech 2024 GPU Score: 6143 (50% better more than M4)

 

[mvprod123]

https://www.tomsguide.com/computing/macbooks/macbook-pro-m5-benchmarks-are-in-heres-how-it-compares-to-m4-windows-laptops-and-more#section-performance-tests

"Last but far from least, gaming. Apple talked big about its new 10-core GPU in the M5, delivering up to 1.6x faster graphics performance and frame rates over the M4 chip. Plus, it's hardware-accelerated mesh shading, ray tracing and Dynamic Caching.

That appears to have all paid off. In our usual round-up of game tests, the M5 MacBook Pro could achieve around 60 frames per second (FPS) or higher, meaning Apple's base chip can really deliver the gaming goods. In fact, it surpassed the Dell 14 Premium with an RTX 4050 GPU in Total War: Warhammer 3 and Shadow of the Tomb Raider, and considering that a discrete graphics card as opposed to the M5 chip's integrated graphics, that's a superb feat."

 

[poke01]

Dave2D

M5 Cinebench ST - 201

 

[mvprod123]

Solar Bay Extreme (faster than 3050)

 

[DZero]

View attachment 132372
Dave2D

M5 Cinebench ST - 201

Holy Sh... thius jump is MASSIVE And seeing from M1 to M5, the ST jump is scary

 

[johnsonwax]

200 pt in Cinebench 2024 ST!

That doubling of SSD read/write is nothing to sneeze at. A lot of my data science work was I/O constrained.

 

[Doug S]

That doubling of SSD read/write is nothing to sneeze at. A lot of my data science work was I/O constrained.

Its SSD speed is now on par with M4 Pro. I remember Apple mentioning that was doubling. I guess they upgraded the PCIe links between the SoC and NAND bridge to the next rev?

 

[mvprod123]

View attachment 132372
Dave2D

M5 Cinebench ST - 201

Geekerwan: 208

MacBook Pro M5性能测试：新架构GPU能带来什么？_哔哩哔哩_bilibili

苹果推出了M5处理器的几款新品，今天我们用新款MacBook Pro 14，来看看M5的性能表现如何，新GPU在游戏、AI、创作领域的表现是否都有明显提升？续航表现又如何？一起来看看吧！, 视频播放量 96398、弹幕量 380、点赞数 5393、投硬币枚数 1215、收藏人数 1068、转发人数 604, 视频作者 极客湾Geekerwan, 作者简介 爱折腾、超硬核的科技数码频道！商务合作请联系：bd@geekerwan.com，相关视频：女子不知apple pay花的是真钱 破产后崩溃哭诉 这智商绝了，为什么iPhone 17 Pro Max的寿命只有七天？，iPhone...

www.bilibili.com

 

[CouncilorIrissa]

View attachment 132372
Dave2D

M5 Cinebench ST - 201

That's....wow. Hats off. Absolutely relentless pace of iteration.

 

[mvprod123]

M5 Apple Silicon: It’s All About the Cache And Tensors

Today is a big day, M5 Day! I say this because I believe M5 is a bigger deal than the products it's in. That's not to say these products are bad, quite the opposite: they are so good and have been so good for so many years that it's hard to think of anything meaningful

creativestrategies.com

 

[trivik12]

Amazing performance. I wonder how the battery life is. I am expecting 20+ hrs of web browsing.

 

[jdubs03]

Based on Geekerwan spec2017 data from the MacBook Pro…

• The M5 p-core has about a 3.5% improvement in IPC vs the M4 - 6% overall.
• But the A19 Pro p-core has about a 6% improvement in IPC vs the A18 Pro - 11.5% overall.

And based on CB and GB single-core results there and elsewhere, we see a 10-15% increase for the M5, including variance and accounting for the slight frequency increase.

I’m a bit surprised spec2017 shows lower.

And then also the M5 iPad Pro being roughly the same frequency as the M4 iPad Pro is showing 10% improvement in GB.

Seems like it would be the opposite.