Today's Apple event

[Aaronage]

Yeah, I’m really trying to resist the urge to doom, and it’s certainly worth waiting to see actual benchmarks, but it does feel underwhelming to me.

Agreed. It’s not terrible, it’s great even, but Apple has set the bar so high.

 

[Jimmyjames]

Agreed. It’s not terrible, it’s great even, but Apple has set the bar so high.

They could really do with supplying actual benchmark numbers. I’m critical of Qualcomm, but at least they provide something. Manipulated as it may be.

 

[Jordanmarkstone]

They could really do with supplying actual benchmark numbers. I’m critical of Qualcomm, but at least they provide something. Manipulated as it may be.

At least we won’t have to wait 6-9 months to find out

 

[Aaronage]

Grumbling aside, I still ordered a 14” M3 Pro 12C/18C with 36GB of memory. Looking forward to trying it out.
Strangely, despite being CTO, I’ve got a delivery estimate of November 10. That would be quite a fast turnaround.

 

[Yoused]

Do the math. TSMC has been burning about 50 wafers a day since March. If they are getting an average of 20 good chips off each wafer, Apple would have over two hundred thousand A17s, M3s and watch chips in hand by now (or out the door). Good turnaround time is not all that surprising. This is not the turn of the century Apple.

 

[dada_dave]

The memory configurations are weird. All sorts of non-overlapping tiers.

M3 Pro: 18GB or 36GB
M3 Max 14/30: 36 GB or 96 GB
M3 Max 16/40: 48 GB or 64GB or 128GB

 

[exoticspice1]

I kinda expected more? I mesh shaders and RT have been in NV GPUs since 2018. The fact it took Apple 5 years to bring it Mac is honestly shocking.
They showed 0 Ray traced games.

No mention of Nvidia this time. But they mentioned Intel cause they are ahead in CPU.

These are just my thoughts.

 

[theorist9]

M3 is still limited to two displays (as expected). Though OWC does have a product that allows you to add another, and since it's OWC it should be a robust solution.

 

[theorist9]

Here's how the pricing compares for the M2 and M3 MBP's (models listed at bottom):

• Same starting price.
• Same upcharges for SSDs.
• Same upcharges for RAM, except the first two steps for the M3 are 18/36 instead of 16/32, and the M3 has added options of 48 and 128.
• Upsteps for CPU/GPU aren't comparable but: M3 has lower upcharge to go from base Pro to top Pro (+$200 for M3 vs. +$300 for M2), and higher upcharges to go from top Pro to base Max (+$400 vs. +$200) and base Max to top Max (+$300 vs. +$200). Total upcharge to go from base Pro to top Max in M3 and M2 are +$900 and +$700, respectively.

 

[theorist9]

Will be interested to hear what the memory buses are. The low-end Max on the M3 is degraded (300 GB/s; on the M2, both Max variants were 400 GB/s).

The top M3 Max's 400 GB/s probably means they're still using LPDDR5.

 

[mr_roboto]

The memory configurations are weird. All sorts of non-overlapping tiers.

M3 Pro: 18GB or 36GB
M3 Max 14/30: 36 GB or 96 GB
M3 Max 16/40: 48 GB or 64GB or 128GB

Based on the bandwidth numbers @theorist9 posted just above this, they must be populating 3/4 of the M3 Max memory channels on the 14cpu/30gpu chips. This makes sense: if 400 GB/s is adequate to feed the full 40-core GPU (which is the biggest bandwidth monster in the system), then 300 GB/s should be good enough for the 30-core.

 

[leman]

Will be interested to hear what the memory buses are. The low-end Max on the M3 is degraded (300 GB/s; on the M2, both Max variants were 400 GB/s).

The top M3 Max's 400 GB/s probably means they're still using LPDDR5.

View attachment 27042

128-bit on M3, 192-bit on M3 Pro, and the two chops of Max have 384 and 512-bit interfaces, respectively

 

[Andropov]

No, I think not. Looking at the die shots, with my untrained eye, the E-cores appear to be nearly half the size of the P-cores, when they were closer to a quarter on older designs. This might be partly a quirk of the N3 process, which does not down-scale SRAM very much, but I believe Apple is putting more juice into the E-cores so that they can make good use of them without having to resort to putting in mid-cores.

Oh, I was thinking of a 3rd tier below the E cores, not between the P and E cores. As you say, the E cores have been growing every year (in size and performance, power consumption is still stellar)… at which point would we call them middle cores?

I kinda expected more? I mesh shaders and RT have been in NV GPUs since 2018. The fact it took Apple 5 years to bring it Mac is honestly shocking.
They showed 0 Ray traced games.

No mention of Nvidia this time. But they mentioned Intel cause they are ahead in CPU.

These are just my thoughts.

At least on the GPU front, this update is quite big IMHO. Redshift is more than twice as fast.

 

[Andropov]

Memory configuration is so weird this time around. If you select M3 16 CPU cores + 40 GPU cores the base RAM is 48GB (good), and you can select 64GB or 128GB, but not 96GB, as that configuration is reserved for the 16 CPU + 30 GPU M3 Max

Getting the fastest CPU now requires getting the fastest GPU too, which is quite unfortunate. Getting the top CPU was like $2000 on the M1 generation vs almost $5000 on M3. And since the M3 Pro is clearly the weakest upgrade of the generation, it‘s not a great upgrade for people looking to spend $2-3k.

 

[leman]

No, I think not. Looking at the die shots, with my untrained eye, the E-cores appear to be nearly half the size of the P-cores, when they were closer to a quarter on older designs. This might be partly a quirk of the N3 process, which does not down-scale SRAM very much, but I believe Apple is putting more juice into the E-cores so that they can make good use of them without having to resort to putting in mid-cores.

I don't recognise absolutely anything in those die shots, except the GPU clusters and the memory controller

But an A17 analysis claimed that the E-core is still 1/4 of the P-core.

iPhone15 Pro用A17 Proのダイ写真確認〜構成をA15/A16と比較 - iPhone Mania

iPhone15 Proシリーズ用A17

iphone-mania.jp

 

[Andropov]

I found this surprising too: https://x.com/iancutress/status/1719176613445742855?s=46&t=AVo4Ae4rwcqD3xOi5WvXyg

Looks like the M3 doesn’t have the A17’s Neural Engine.

 

[dada_dave]

I found this surprising too: https://x.com/iancutress/status/1719176613445742855?s=46&t=AVo4Ae4rwcqD3xOi5WvXyg

Looks like the M3 doesn’t have the A17’s Neural Engine.

So maybe they are mixing and matching designs from across generations … but why?

 

[Jimmyjames]

I found this surprising too: https://x.com/iancutress/status/1719176613445742855?s=46&t=AVo4Ae4rwcqD3xOi5WvXyg

Looks like the M3 doesn’t have the A17’s Neural Engine.

This, 8gb starting memory and reduced Pro performance leaves a bad taste in my mouth. These are not cheap laptops.

 

[Yoused]

I don't recognise absolutely anything in those die shots, except the GPU clusters and the memory controller

But an A17 analysis claimed that the E-core is still 1/4 of the P-core.

I go by the cache blocks (big rectangles) and look at the structures around them. When you know how many cores there are, you can pick them out by counting. The E-cores are around the smallest cache block (a square-ish shape, usually), and to me they look larger in the M3-series – probably not half, but more than a quarter.

 

[leman]

Gnattu over at MR posted the following dia annotation demonstrating that Apple moved away from 4-core cluster organisation to 6-core clusters. That's a very significant departure from the previous design!