From what's posted on Apple's website, it looks like they didn't use a conventional periscope design to get the extended light path needed for the 120 mm focal length. Instead, they used a "folded glass structure" that internally reflects the incoming light four times to increase the path length. Essentially, they stacked four small periscopes instead of using a single large one.
Thread: iPhone 15 / Apple Watch 9 Event
- Thread starter exoticspice1
- Start date
- Posts
- 5,636
- Reaction score
- 9,255
Some good answers abpve./
As a CPU guy, one thing that comes to mind is engineering the instruction fetch unit in such a way that it can see further down into the instruction stream. For example, consider this sequence of instructions:
A=B+C
D=A+B
E=F+G
if you are looking for two instructions to execute, and you can see past the D= instruction, you will find that you can execute the E= instruction. Modern CPUs can already see a long way into the instruction stream, and perhaps Apple made it so this time it can see even further.
Another thing you can do is carefully choose what each ALU can do. For example, if you have code that does a lot of integer multiplications, but only one of your ALUs had an integer multiplier, you may get benefit from adding multipliers to other ALUs. (This is just an example: multiplications are comparatively rare).
You can also - and this is controversial - play games with hyperthreading. If you are simultaneously executing instruction streams from more than one thread, they are unlikely to have the sorts of collisions I’ve described that prevent instructions from being simultaneously executed. Apple has traditionally not needed to do that, because its done a good job of keeping the ALUs busy (they have the benefit of Arm, which is a lot easier to decode than x86, so that it is easier to look ahead and see more instructions to choose from).
- Posts
- 5,636
- Reaction score
- 9,255
I don’t know what it means when they say things like “N5 to N3 yields a roughly 15% performance increase.” That’s impossible to predict because every circuit is different. All you can accurately predict is that Ft, the toggle frequency of a minimum size unloaded transistor, will increase by a predictable amount.
Once you have real circuits, where transistors are never minimum W/L, and where you have to deal with wire loading, gate loading, etc., you can’t tell me any prediction and have me believe it.
Eric
Mama's lil stinker
Sounds like they have temperature sensors for the 8.
Track your nightly wrist temperature changes with Apple Watch - Apple Support
Apple Watch Series 8 or Apple Watch Ultra can gather wrist temperature data while you sleep to help give you insight into your overall well-being.
- Posts
- 5,636
- Reaction score
- 9,255
Yeah, they mostly focussed on it last year for ovulation tracking.Sounds like they have temperature sensors for the 8.
Track your nightly wrist temperature changes with Apple Watch - Apple Support
Apple Watch Series 8 or Apple Watch Ultra can gather wrist temperature data while you sleep to help give you insight into your overall well-being.support.apple.com
Thanks. I wasn’t aware of that.Sounds like they have temperature sensors for the 8.
Track your nightly wrist temperature changes with Apple Watch - Apple Support
Apple Watch Series 8 or Apple Watch Ultra can gather wrist temperature data while you sleep to help give you insight into your overall well-being.
I was going to post Jon Masters Twitter about possibly different architectures for M3 vs A17 but I think you saw it already
PS TB303-Devilfish on Twitter is an AMD troll which I think you’ve figured out
Last edited:
I've always had a hard time understanding this because I've never found anywhere it's clearly and rigrously explained. But from what I understand, the rate of photon flow into the camera is proportional to the square of the entrance diameter (D) of the lens, and inversely proportional to the square of the focal length (F) (because if you double the focal length you halve the subtended angle of what you're viewing, so the scene gives off 1/4 the light). Thus photon flow ∝ (D/F)^2, where D/F = 1/f, where f is the aperture ratio. And for a given image size and exposure time, its perceived brightness will in turn be proportional to the square root of the photon flow (because images are 2-dimensional), i.e. to 1/f.
So at least based on this, f is a measure of the light-gathering capability of the optical device, and it's independent of sensor size. Thus if you have a lens of a given f-value, you should have a constant amount of light falling onto the image circle, not a constant amount of light per unit area. But maybe I'm confused!
And thus what's done in a camera with a larger sensor is to spread out that light over a larger imaging area (which is why cameras with larger sensors [unless they do something fancy] tend to need more distance between the lens and the sensor, to provide the distance to allow that light cone to expand). Consequently, if your lens is designed for your sensor size (such that the crop factor is independent of sensor size, i.e., its corners go equivalently close to the edge of the imaging circle), the same amount of light should fall on the sensor, regardless of sensor size (for a given f-value, and assuming a fixed aspect ratio).
Last edited:
Jimmyjames
Site Champ
- Posts
- 892
- Reaction score
- 1,023
I did indeed! Jon seems perplexed by it.I was going to post Jon Masters Twitter about possibly different architectures for M3 vs A17 but I think you saw it already
PS TB303-Devilfish on Twitter is an AMD troll which I think you’ve figured out
Lol TB303-Devilfish's trolling game is weak. The other person in that convo is Jordan. A really good, knowledgeable person. I should invite him here.
- Posts
- 5,636
- Reaction score
- 9,255
Remains to be seen, but the only conclusion I would draw is that the A17 clock rate did not increase by the scale factor.
These, along with a few other iPhone 15 results, were just posted here: https://browser.geekbench.com/ml/v0/inference
Ah, never mind--they list the same MB's as the iPhone 14, and so are probably spoofed.
Ah, never mind--they list the same MB's as the iPhone 14, and so are probably spoofed.
Last edited:
Natural titanium seemed good to me, too. Honestly, it’s like with cars, the color is the least important thing in making the decision. Besides, I’ll be putting it in a case, and probably getting a LifeProof this go-round since I seem to spend half my life in the water.
My thoughts on materials: It still seems that aluminum is the higher-performing frame material, especially if you use a case (and thus don't need to worry about scratches): Their relative stiffness:weight ratios (comparing A7075-T6 to Ti6Al4V) means you can achieve the same stiffness with aluminum at ≈75% of the weight (extrapolating from flat sheet mechanics).
Titanium's advantages are scratch-resistance, frame thinness (15% thinner at the same stiffness--not sure how much that matters) and, of course, marketing.
Aluminum's main practical downside is that, at the same weight, it has the same yield strength. Thus if you take advantage of its higher stiffness:weight ratio to make the frame 25% lighter, it will have 25% less yield strength, and thus be correspondingly more susceptible to corner damage.
At least that's the theory.
Titanium's advantages are scratch-resistance, frame thinness (15% thinner at the same stiffness--not sure how much that matters) and, of course, marketing.
Aluminum's main practical downside is that, at the same weight, it has the same yield strength. Thus if you take advantage of its higher stiffness:weight ratio to make the frame 25% lighter, it will have 25% less yield strength, and thus be correspondingly more susceptible to corner damage.
At least that's the theory.
Last edited:
Given N3 is that wacky FinFlex, how likely do you think it might be that the M3s will use slightly different gate configurations for performance gains over the A-series? Would that involve (have involved) too much reworking of the designs?
There’s a lot of overlap (time, apparent design rules, and physics) between N3B and N3E, that I’d hope no reworking is required because they’d be designed in tandem. I’ve got a lot of thoughts on this, but maybe @Cmaier has a copy of TSMC’s N3 design rules in his back pocket that he could share?
I know very little but my understanding is that this isn’t the correct interpretation of their mechanical properties:
Titanium vs Aluminum: Workhorse Metals for Machining, 3D Printing
Aluminum and titanium work great for 3D-printed and machined parts thanks to their strength, light weight, and tolerance to heat.
www.protolabs.com
I do believe you correct that Aluminum has higher stiffness but it isn’t as strong. That’s true but that means Aluminum is more susceptible to staying bent once bent - so more than just corner damage. When I looked it up the above quote recommending material for components appeared to be universally held: titanium is better if you can afford it - Aluminum is cheaper but not quite as good overall. The advantage of aluminum in dissipating heat is why Apple kept it on the inside. One hopes that the inside honeycomb being Aluminum and the outside being Titanium gets you the best of both …
Last edited:
Ultimate yield strength is basically just the force per volume to permanently deform a material, corner damage included. Corner damage is actually worse because it usually involves gouging or cracking, which introduces stress risers that further fracture with much less force. Juxtapose that to stainless steel (e.g. my iPhone 12 Pro)— it has high toughness which is partly due to the material’s yield strength and partly its malleability. My phone weighs something like 2g more than the 15 Pro because less thickness is required.
TSMC gave a range from 10-15% perf at the same power for N5->N3 however, technically, the A16 is at N4 not N5 which already (supposedly) offered a 5% perf increase over N5. I don't know what the N4->N3 performance increase at the same power was estimated to be.
I always use a case, too. The only reason I pay any attention to color is that I try to avoid ones that may make it harder to sell the phone when I'm done with it.
Similar threads
