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
Yeah, I would have liked to see that too, but I assume that the difference from A16 scores would be in the realm of statistical uncertainty. A more interesting question would be the power draw running at that frequency. From Geekerwan's video, the per-core power draw with reduced frequency is essentially identical to that of A16. But I would be more interested in seeing the total SoC power draw.
That is what Mark Gurman said he found in developer logs or something and he’s very accurate usually.
There is also apparently 36GB and 48GB RAM options in testing
Attachments
M3 Max with 4 extra P cores? that should erased the M Max from the 14" Mbp right? Or apple should limit that SoC for the 14"
The 14" was also on its limit till now....but if this will come up true, am i wrong that Apple should put only the M3 Pro into the 14"?!
For better info about what he said across the M family
Well I hadn't seen this data points before (to be honest, I only skimmed through the video), arrows are mine:
Isn't that to say the A17 Pro can match the maximum performance of the A16 Bionic at a significantly reduced power draw?
Isn't that to say the A17 Pro can match the maximum performance of the A16 Bionic at a significantly reduced power draw?
Gurman isn’t a developer log kinda guy nor is he a reverse engineer, so I’ll maintain my skepticism there.
As far as RAM goes, anyone can look up what is available, so I concur with him I suppose 🤷♂️
Agree, but he is right, in general Mark is on point when it comes to these...he was accurate regarding the M2 family..so probably has some people that provide this to him for sure. But we will start to find out in spring next year probably
Yeah, I don’t understand that graph. It seems to show both single and multi core, but the data doesn’t make sense. For example, A15/16 results seem to be GB6 scores instead of GB5.
I assume that the extreme result of A17 there is multi core GB under aggressive cooling. Where are similar results for other chips? Weird data presentation.
It doesn't show both the single and multicore, it shows the multicore performance across different power levels. They say the reason they don't connect those dots and draw a single line like they do with the results from Android phones is that with Android phones they can set the processors to use a specific TDP in a more granular manner, but they can't do that on iOS without a jailbreak, something that's a lot harder to do on newer iOS versions (they were able to do it for A13 and A14). On iOS they can only know the performance of Low Power mode and whatever limitations that puts on the SoC in terms of power draw, and regular mode (and I guess the 3rd data point is when they strap Peltier cooling on to the display?).
I think those are just multicore results at different power levels. No idea how they got to test those power levels though.
It can't be just low power mode, there are 3 data points for the A17 Pro, at roughly 1.7, 9.5 and 14 watts. I guess the 9.5W result could be the low power mode result, but what about the 1.7W score? How did they measure that?
As I mentioned in the edit, I would guess the 1st point is low power mode, 2nd point is regular mode (perhaps when the phone is in their hand with no accessory other than a cable going to their computer), and 3rd point is the one in the picture attachment:
Attachments
I don’t know those reviewers, so they could credit Tim Cook and Johny Sroiji, but that doesn’t mean anything to me 🤷♂️
Possibly.
Regarding GPU performance in general, rasterization appears 20% faster than on the A16, but the A17 has 20% more CPU cores and consumes 20% more power. So it doesn't look that efficiency has increased at all. What I am missing?
I assumed you knew based on you saying "someone who's done proper research", my bad. I don't know most of the names either but the name that sticks out to me the most is Andrei Frumusanu, who was previously a writer for Anandtech doing iPhone reviews among other things, and someone who's highly knowledgeable and respected. Specifically was the top guy people would go to for deep dive analyses of new A series processors. He now works at Nuvia, the company bought by Qualcomm.
What are they thanking them for? It was implied earlier that they had been helped in a more proactive way. From the video it appears that they are crediting them as significant reviewers of Apple Silicon historically. That’s why @name99 was asked for confirmation. I would also be very surprised if Andrei had anything to do with this review.
Ah, thanks, that explains it! Still a bit confused about the low score of A17 in low-power mode. But if we disregard that one, it seems that graph shows two things:
- A17 is significantly more power efficient in MC than any other tested device
- Apple doesn’t frequency-throttle the P-cores if enough thermal headroom is available
I’d say this is an impressive result for Apple
I don't think that you are missing much. The new GPU is probably more about capability than efficiency improvements. Then again, there is this, which shows that the system is capable of efficient operation under demanding conditions
Well, I should have my phone by the end of the week and I'll try to run some basic benchmarks too
But in the test you show, the A17 only beats the A16 by a few percents (I suppose frame rate is limited to 60).
At the same level of performance, it is expected for the GPU with more cores to consume less, right (assuming an equivalent level of performance per core)?
Well, I should have my phone by the end of the week and I'll try to run some basic benchmarks too
Just remember to wait until after all the new phone background stuff is finished doing its thing...?
I do agree with the 1st point: it must be low power mode. Apparently on low power mode the iPhone turns off the P cores and uses only the E cores. The score roughly matches what we would expect from the aggregate of the E cores, so it's likely to be the low power mode. But if it is, does that mean that the A17 Pro is much slower when in low power mode than the A16 or A15? Both of those score >3500 Geekbench 5 points, but the 1st point of the A17 Pro doesn't even reach 2500 points.
For the 2nd point: I disagree. That can't be the "regular" score, it barely matches the A16, while we know that the A17 Pro beats the A16 by ~10-12% in "regular" benchmarks. I'm inclined to say that this is the "throttled" result, where the A17 Pro clocks down to 3.46GHz. But the same reviewer also said that the A17 Pro used the same power as the A16 at that frequency...
For the 3rd point: I think that one could be the "regular" score? It's the only score that beats the reported A16 score. Though it does use massive power to do so.
This is all quite confusing.
You are right, this is odd. I had a look at A17 GB5 scores. They seem to be around 5900-6000. Two results are around 6200, and two are around 7500(!!!), but those are weird since they use the wrong OS. It just doesn't add up.
Agree with all of your points, however as far as I know they don't turn off the P-cores in Low Power Mode, but rather limit them to a very low clock speed (~1500MHz). If they turned off the P-cores it wouldn't be a very meaningful data point in this chart since you're not taking all of the cores and restricting their total power draw to a specific wattage (you're not taking advantage of all of the hardware). Though, we don't know if this behavior is the same on the 15 Pros as it has always been on previous iPhones, i.e if the efficiency cores exceed a certain ideal performance threshold Apple might just choose to turn off the P-cores and use only the E-cores in the future for further power savings in that mode. Geekerwan didn't say anything about the low power mode in that regard so I'm assuming it's the same (he did say that he experienced worse performance in Genshin in low power mode on the 15 Pro Max than on the 14 Pro Max towards the end of the video).