If we judge by Blender results, Nvidia is roughly 4x faster by product category, and that with a much higher shader core budget. E.g. 4070 RTX is 3x faster than M2 Max, and that GPU has 20% more shader clusters and is clocked almost 80% higher.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
If we judge by Blender results, Nvidia is roughly 4x faster by product category, and that with a much higher shader core budget. E.g. 4070 RTX is 3x faster than M2 Max, and that GPU has 20% more shader clusters and is clocked almost 80% higher.
But all new iphones have same battery life stated like last year.
Strange
Not strange, we don't have enough info yet. Any savings in the SoC could be spent on the screen, U2, Wifi, 5G, more background processing (especially NPU), etc. For that matter we don't know how big the batteries are yet! Perhaps they think lighter weight is a higher priority than longer battery life.
Yes, I have some faint hope that they clocked down, and we'll see better gains in the M3.
20-30% over the M2, while good, is not what I expected or hoped for, and it was reasonable to think we'd do better. There was room for that much on IPC alone, before clocks.
It's also possible that they've focused more on multicore scaling over the past year. We won't know until the M3 ships (or, probably, the M3 Pro/Max).
BTW, my guess is that the "Pro" in "A17 Pro" is so they can later ship a non-Pro A17 on N3E.
Already 20% would make an entry-level passively cooled Apple laptop as fast as than enthusiast-class x86 desktop. I don’t think this is disappointing at all.
Fine by me, the power you get *unplugged* in the MBP series is still unrivaled given the competition drops like a rock when put on battery power…
I am not sure I understood your logic. If both M2 and M3 are clocked 15% faster than their A counterparts it means same 10% boost for M3 over M2, is not it?
They are comparing A17 Pro to A16, which is already as fast as M2 (talking about single core performance). So I’m adding the 15% clock on top of the 10% announced difference.
I’m not smart enough to debate architectural improvements, but seeing the pattern of posts hoping that the A-series chips improve efficiency and the M-series improves performance usually doesn’t work out.
It seems like if the A-series gets a 10% improvement you can bet on the M-series getting the same.
Raytracing is interesting because we have no frame of reference for it on any Apple platform.
So A17 Pro now has more transistors and is faster than the M1 SoC that released less than 3 years ago that runs macOS. That’s pretty wild!
I’m confused as well. If A16 is as fast as M2 including the higher clock speed on M2, is that then because M2 is not based on A16 but A15? I have lost track, sorry.
M2 is based on A15, and there is good reason to believe that that M3 will be based on A17, skipping A16. This is why I expect M3 to bring a more substantial improvement over M2 (already A17 is going to be faster than M2 Max in SC).
I agree with you casting doubt on the idea that A-series are efficiency focused. It’s not a general pattern. But A16 was generally more efficiency focused than A15, it not only brought a reasonable (albeit not impressive) performance delta over A15, but also offered significant energy savings per unit of work done. Anandtech has a very good article about it. I believe that A17 is somewhat similar. These cores are already ridiculously fast. It makes sense to clock them conservatively and leave their power reserves to the desktop. At any rate, the way how A17 is clocked seems to be more of an “in your face” directed at Intel. If 10% improvement is accurate, that would put A17 exactly at the level of Intels enthusiast-level desktop CPU that burns over 50 watts to reach the same SC performance.
We kind of have, since raytracing (obviously without specialized hardware acceleration) has been available in Metal for a couple of years now. This makes it easy to study how good Apples hardware RT is, since you can use the same code on different devices.
Yes, precisely.
They usually release the documents a few days after the hardware becomes available. Should pop up over the next few weeks hopefully.
A few things:
1. We probably are at a point where CPU speed takes a backseat to the Neural Engine and GPU in transistor budget and design.
2. Perhaps N3B isn’t that great of a process in clock efficiency.
3. Perhaps there is some truth to the loss of chip design talent.
4. This is probably the first CPU designed primarily during WFH and in the middle of Covid. I doubt SoC engineering can be done as well at home, like software engineering can.
Maybe, Maybe not. I did get some really good ASIC designs done for me during that time from people working at home. I understand it is not the exactly the same but it is more than just software development.
Unfortunately, A17 Pro got only 10~20% compared to A16 especially since it got a new 3nm. The performance gain is quite disappointing so far and M3 might get less improvements over M2 because of how A17 Pro came out. Probably need to wait for benchmark but if it's true that CPU got only 10%, GPU got only 20%, that would be very disappointing.
Why the M3 might get less improvements over M2? Since also leman said that M3 may skip the A16...so M3 over M2 are 2 generations difference
My thoughts are that M3 will get more gains over M2 than the A17 got over A16
Yes, this was my takeaway assumption also. ASSUMING they DO in fact skip the A16 core design and utilise the A17 cores.
One would assume the A17 and its core architecture has been in design for some time, in parallel with the Mac version counterparts.
OR, we could see some hybrid where it's a 3nm process, A16 CPU cores, and A17 GPU cores with Ray Tracing! Who knows?
hopefully not, people and tech journalist will probably makes this very public with the first M3 mac that will be released
Because A16 is just another A15. If you compare with A15, then you just admitting that M3 won't get performance gain way more than M2 did especially since M3 is 3nm based, not 5nm based. A whole new CPU/GPU architecture, high clock speed, and 3nm supposed to give more performance boost.
I'm not a hardware designer. I presume that hardware designers need to work together more and have better coordination and have to use specialized testing tools that are harder to get WFH.
I'm a software engineer. The software quality and communication definitely suffered during the WFH era for me and my company.
It's way too really for this kind of pessimism. The crucial detail will be the clocks. If A17 runs at 3.6Ghz to achieve this 10% uptick, then it's indeed not a good sign. If instead it runs at 3.4Ghz, that would be an impressive IPC improvement. Shouldn't be too long until we have these details. My personal guess is that the architecture is capable of much higher clocks, they just wanted to limit the peak power consumption on the phone because it's already years ahead of closest competition.
Don't get infected by the likes of sunnyboy the fifth, that dude would still complain about "disappointing Apple" even if they released an Apple Watch that can take on an Nvidia GPU...
From a marketing standpoint, it looks bad. Apple focused more on improving performance than improving efficiency at the keynote. If Apple had designed the A17 to improve efficiency, Apple should have focused on that during the keynote.
It seems that TSMC N3B increases performance by 10~15% with the same power compared to TSMC N5.
| N3E vs N5 | N3 vs N5 | |
| Speed Improvement @ Same Power | +18% | +10% ~ 15% |
| Power Reduction @ Same Speed | -34% | -25% ~ -30% |
| Logic Density | 1.7x | 1.6x |
| HVM Start | Q2/Q3 2023 | H2 2022 |
TSMC Outlines 3nm Roadmap: Five FinFlex Nodes
TSMC's N3 node looks to have a big and bright future with FinFlex.