Thread: iPhone 15 / Apple Watch 9 Event
- Thread starter exoticspice1
- Start date
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
W
Wonderful!
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 365
Let's not forget from A14 to A17 the improvements are mainly from clock increases. While the charts are impressive they would be more impressive if the clocks stayed the same.
@leman and @Andropov forgot that the clocks increased that's HOW you got those increases.
Clock the A17 to A14 levels there would be barely IPC improvements. Apple needs a new design soon they cannot do this forever otherwise they will end up like Intel in the CPU space
@leman and @Andropov forgot that the clocks increased that's HOW you got those increases.
Clock the A17 to A14 levels there would be barely IPC improvements. Apple needs a new design soon they cannot do this forever otherwise they will end up like Intel in the CPU space
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 365
Now if the P cores are more efficient that's a different story.
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
I don't believe they forgot. I think their point is twofold: firstly percentages are misleading. Think in absolute point increase terms. Secondly, does it matter where the increases come from if they keep coming? They have been consistent for 10 years. We can worry that it’s coming to an end and perhaps it will, but it’s not ending now and this year is showing one of the biggest increases yet.
Andropov
Site Champ
- Joined
- Nov 8, 2021
- Posts
- 712
The A14 Bionic had a max clock of 3.1GHz, while the A17 Pro has a maximum clock (if the leaked benchmarks are true) of 3.78GHz (a 21.9% increase in clock speed), but it's 46.7% faster.
Still, every year they have outdone the clock speed increases by a significant margin:
And single core is not the only story here: Apple has some ridiculously awesome efficiency cores. I think we just talk less about them because their performance is harder to measure, as we can't do a single core run on an E core. The years Apple focuses on the E cores, the single core score is not going to see the same increase in performance as the years the P cores get a redesign, but it doesn't mean the Apple Silicon team is standing still.
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
Anyone know more about AMD’s latest. Im being questioned by fanboys declaring that AMD’s Zen+ and later hit 20% increase. I’m guessing this is multi-core?
In general AMD’s Zen cores have seen impressive IPC gains generation over generation but they started from a pretty bad spot. Zen 3 is when they finally overtook Intel (though Intel has been improving lately too). I don’t believe they can match Apple at Apple’s clock speed (a reminder that IPC is dependent on clock). We’ll see for Zen 5 which will be out next year.
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
Thanks.
Andropov
Site Champ
- Joined
- Nov 8, 2021
- Posts
- 712
I've no idea about AMD, but I'm interested to know how Intel will do this year. I was under the impression that they put everything they had in Alder Lake, I'm curious to see if they manage to release a substantial upgrade with the 14th gen chips.
We didn’t forget about it. But your expectations are hardly realistic. Apple already delivers higher IPC than anyone else, by a large margin. At some point you’ll be hit by diminishing returns, hard. And we do see some pretty significant IPC improvements in selected tests. Actually, I’m surprised they were able to improve the branch predictor, theirs was already best in class.
Well, yes, but they are already way ahead on the IPC game. I mean, this 3.7Ghz phone chip is competing in peak performance with Intel’s 5.7ghz parts. I don’t think they’ll get in trouble any time soon. But at some point, yes, one needs principally new approaches to keep getting faster. Probably more than CPU tech alone...
If I remember correctly, Zen4 improves the IpC around 10-15% over Zen3, which is great. But Zen3 single core pretty much sucked. It’s easier to get better improvements if you are below average.
Last edited:
3.7 GHz phone chip
Ehhh Zen 3 was okay, not as good as Apple of course, but that’s when Zen really became more than just the budget multicore chips. Zen 2 and earlier were indeed pretty bad in terms of single core performance both in absolute terms and IPC. However if Zen 5 again improves IPC as much as leaks have suggested (we’ll see), it should be a really nice bump.
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
Do we have any estimate how Zen5 might compare to Apples best? I had a look but could only find Cinebench R23.
3.7 GHz phone chip
Ups, you are right. Fixed it!
I meant comparing to Alder Lake.
Not really. There are only “leaks” which right now are bullish. But I don’t think spending too much time analyzing them is particularly fruitful - you can’t know what’s real or even if real what conditions they might be tested under. When Zen 5 comes out sometime in the next 6-8 months, we’ll see. Obviously leaks close to release, particularly in a review window can be analyzed, but we aren’t even close to that.
Fair but also Alder Lake was the first big jump in Intel’s IPC in awhile. And one could argue should be compared to Zen 4 which let AMD keep pace. Bottom line: Intel and AMD are now trading blows in single core which is very different from where either of them were a few years ago.
The way I see it, Intel and AMD both have a lot of IPC to catch up on while Apple has a lot clock speed head room left even without blowing their power budget (TSMC fabrication of course can be credited here too).
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,085
First Metal score for the 15 Pro Max = 27158
Not bad. A little less than an M1, but now with ray tracing.
Not bad. A little less than an M1, but now with ray tracing.
iPhone16,2 - Geekbench
Benchmark results for an iPhone16,2 with an ARM processor.
browser.geekbench.com
Geekbench GPU doesn’t test ray tracing right? So that’s pure compute/shader?
That’s correct, though it does test ray tracing in CPU. IMHO, blender will be the most representative benchmark for ray tracing, so I’m kind of glad geekbench left it out of their GPU compute tests.
