Which tests?
http://www.anandtech.com/show/6112/...agon-s4-apq8064adreno-320-performance-preview
http://www.anandtech.com/show/5779/htc-one-x-for-att-review/4
I'm not seeing the "magical" 1.5x performance benefit claimed for S4 over A9. Also to note is that most S4 chips are clocked very high (1.4GHz or above) and they are most often compared to quad-core A9 SoCs. That's hardly comparable because quad-core scaling is obviously not as linear as dual-core (evident with the quad-core S4 Pro). You can safely say that S4 is 1.5x faster per core compared to quad-core A9 as a whole SoC, but I don't think you can safely say that S4 is 1.5x faster per core against any A9 SoC on average.
Subyman
Moderator <br> VC&G Forum
- Mar 18, 2005
- 7,876
- 32
- 86
The iPad 3 is plenty fast as is, I was surprised to see them bump the performance even more. Its great for developers. I'm glad they dropped support for iPad 1 for iOS6, having to be compatible with the original iPad was a hassle. That thing is very, very slow.
I think it was clear an A6 bump would be coming after the iPhone 5. I just didn't expect it to come THIS soon. Nothing's really challenging the iPad 3 yet, and I would've thought this could wait another 4 months.
Are you blind? Your very own links, outside of the Javascript benchmarks (which I've shown are more dependent on software) and 3D benchmarks (which obviously don't have anything to do with the CPU), you can see in the pages you posted yourself where the S4 monkeystomps the A9.
From User Experience benchmarks like Vellamo to synthetic but native CPU benchmarks like Linpack. Even in Basemark, a dual-core S4 exceeds the dual-core A9 more than twofold.
After you adjust for the higher clock speeds, it's pretty damn close to the 1.5X I'm talking about.
Uh...
Anand said:
And...
Anand said:
I'm not blind. I just know which benchmarks I need to take with a pinch of salt.
And I don't see anything in any of those charts that tells me Krait (S4) is 1.5x faster than A9 other than Linpack. Every other benchmark has S4 only slightly ahead or even on par with A9 parts.
The problem, as Anand highlighted, is that most tasks nowadays are integer-bound.
Personally, I'm a developer so I know Anand is just throwing a blanket statement (we can agree that Anand is wrong there), but for what it's worth... I still don't see any indication (even from my own development work) pointing to S4 having a 1.5x advantage over A9.
Last edited:
So you are blind. Since you missed out the other non-JS and 3D benchmarks like Basemark and Flashmark.
I'm being harsh here but seriously, point out some graphs from your links that aren't javascript or 3D. You keep saying that you don't see it but you've not actually shown that.
Of course, there's still the Microsoft Surface with a Sunspider score of 969ms you're ignoring. You've only focused on Sunspider when it benefited the iPhone5 but how do you explain the Tegra3 at 1.3GHz (same as the A6!) getting nearly the same score despite being an A9.
Last edited:
I don't mean to get in the way of your debate, but wouldn't the obvious explanation be core count?
Sunspider is single-threaded so that's not it either =/
Because of modern JIT javascript engines, Sunspider and other JS-based benchmarks are only useful for comparison the exact same software on the same platform. Even going between the Snapdragon S3 and the Snapdragon S4 skews things to a ridiculous degree (Chrome for Android running on Android 4.1.2 Cyanogenmod 10 scores within 10% per clock when comparing the S4 and the S3!)
Last edited:
Let's see...
S4 Pro: 1.5GHz quad-core S4 = 1139
HTC One X: 1.5GHz quad-core Tegra 3 = 936
Performance advantage: 1 + (1139 - 936)/936 = 1.22x
Where is said 1.5x performance advantage?
Or do you mean to say use the dual-core part's score? But hey... how come quad-core S4 scores 1139 while dual-core scores 1123? Holy ****! Dual-core S4 is almost 100% faster per core than quad-core S4!?
^ And that's why I don't think Basemark scores take into account all cores at once.
And I never used Sunspider to make the iPhone 5 look "better". jpeyton mentioned it because he wanted to let me know A15 may beat A6, and my reasoning is still that it's unclear if A15 can really beat A6 based on Sunspider.
Exactly the same line of reasoning you are using here. It's unclear whether S4 is faster than A9 or not just based on Sunspider alone. We were never in disagreement. You are just too fixated on the fact that S4 MUST absolutely be 1.5x faster than A9. That's just not true.
Apple's A6 is indeed 1.5x faster than A9, though.
Please do your homework...
Anand said:
Last edited:
Let's see...
S4 Pro: 1.5GHz quad-core S4 = 1139
HTC One X: 1.5GHz quad-core Tegra 3 = 936
Performance advantage: 1 + (1139 - 936)/936 = 1.22x
Where is said 1.5x performance advantage?
Well well, we're talking about a benchmark that's actually somewhat scalable to quad cores.
I'll let you think about that. Look at the OneX (ATT) scores which is on a non-developer platform (i.e., optimized)
Hint: look at the Gnex which is also an A9 except dual core.
HTC One X: 1.5GHz dual-core S4 = 1092
HTC One X: 1.5GHz quad-core Tegra 3 = 936
Performance advantage: 2 x (1 + (1092 - 936)/936 ) = 2.33
So yeah, what about 1.5X performance? We're actually more like 2.0X performance.
Last edited:
Please explain why 1.5GHz quad-core S4 scored 1139 then.
Based on your reasoning, 1.5GHz dual-core S4 parts are 100% faster than 1.5GHz quad-core S4 per core.
Something is very off there.
In case you haven't noticed, top of that chart is the quad-core S4 (APQ8064).
Non-optimized developer platform. We can look at the S4Pro again when the LG Optimus G comes out.
But we have the dual-core A9 in the GNex to easily compare with the dual-core S4 OneX.
BTW, about the Sunspider and cores thing, I'm pretty sure Anand had simply forgotten that even on the desktop, all the current JIT engines are still single-threaded, much less on Android. Besides, a Sunspider run with OS Monitor enabled quickly verifies that the Android stock browser (actually Opera, Chrome and Firefox as well) is definitely still single-threaded while yielding the expected score.
Last edited:
Heh, I just copy and pasted while replacing numbers. Didn't stop to think about how to do it.
I don't think "non-optimized" means 2 cores turned off. Also:
Galaxy S3 dual-core S4 at 1.5GHz: 912
HTC One X dual-core S4 at 1.5GHz: 1092
Performance advantage: 1.20x
So... S4 beats itself at 1.5GHz by 20%? Something still seems off.
It looks to me like optimizations make more of a difference than hardware in that test.
https://developer.mozilla.org/en-US/docs/DOM/Using_web_workers
http://msdn.microsoft.com/en-us/hh549259.aspx
I think you need to do a lot more reading at this rate...
It's now possible to build multi-threaded Javascript benchmarks.
Yeah, that would make it pretty worthless for comparing performance for anything beyond two specific devices.
Doesn't really matter since Sunspider isn't multithreaded.
Besides, this is still a limited application of multi-threading capability and not applicable to many things.
You're only ever saying that when it's not in your favor (where's your big talk about the A6's 900ms Sunspider score?). At the very least, I'm talking about a 111% difference compared to the 20% difference you're nitpicking about.
I'm thinking I have a stronger case here.
Last edited:
You probably don't see it on Android in certain browsers. But some of the tests are actually multithreaded on systems that support it.
Really, it's possible. You're just grabbing at straws.
I'm sure Anand will agree with me:
http://www.anandtech.com/show/4144/...gra-2-review-the-first-dual-core-smartphone/8
I'm afraid Anand is simply wrong on this.
You can play with the core affinities all you like and you'll not find a difference in the scores of Sunspider. Or it's so small that it's not noticeable (sometimes trying to "multi-thread" something doesn't actually help).
Last edited:
Let's move on and look at something else, then. We're getting nowhere.
Here's the Peacekeeper score for that new Samsung Chromebook with the Exynos 5250:
Source: https://plus.google.com/u/0/+JoeWilcox/posts/8LrBK9CKJG4
Looking at those numbers, suddenly the Exynos 5250 doesn't really have that big an advantage over the iPhone 5 (shown right beneath). 971 over roughly 900 is not that big a jump in score, and we're talking ARM Cortex A15 @ 1.7GHz vs Apple's Swift core at 1.3GHz there (both dual-core).
And that's exactly what I was trying to show from the very beginning. A15 just doesn't have that much of an advantage over Apple's Swift... if at all.
Here's the Peacekeeper score for that new Samsung Chromebook with the Exynos 5250:
Source: https://plus.google.com/u/0/+JoeWilcox/posts/8LrBK9CKJG4
Looking at those numbers, suddenly the Exynos 5250 doesn't really have that big an advantage over the iPhone 5 (shown right beneath). 971 over roughly 900 is not that big a jump in score, and we're talking ARM Cortex A15 @ 1.7GHz vs Apple's Swift core at 1.3GHz there (both dual-core).
And that's exactly what I was trying to show from the very beginning. A15 just doesn't have that much of an advantage over Apple's Swift... if at all.
Peacekeeper is a browser benchmark. I thought you already said that we can't use that kind of benchmark since we threw out Vellamo.
And it's actually a worse test than Vellamo since it also tests things like Video Playback in your browser which is completely independent of the CPU with mobile platforms since that's entirely done by the dedicated decoding hardware.
And it's actually a worse test than Vellamo since it also tests things like Video Playback in your browser which is completely independent of the CPU with mobile platforms since that's entirely done by the dedicated decoding hardware.
Last edited:
And I said the same for Basemark. Yet you insist that S4 must absolutely be 1.5x faster than A9 and that Basemark must be true because it showed that.
But that's that. This is this. There is no other way to compare to Chrome OS. The whole thing is basically a glorified web browser. What else can you do to show performance comparison between Exynos 5250 and A6?
Or shall we resume this discussion next year when Exynos 5250 finally hits phones and tablets?
And I refuted the Basemark claim. The S4 comparison between the One X and S3 is 20% while the difference between those and the A9 chips is over 100%. There's an obvious conclusion here.
Why are we talking about the Chromebook? We have perfectly good S4 regular platforms, operating within 20% of each other, to compare to.
Just as a review, the S4, the A6 and the A15 all represent a huge jump over the A9. The S4 is supposed to be slower than the A15 per clock but not by a huge margin. I'm claiming that the A6 appears to be like the S4 in terms of performing much better than the A9 but short of the A15.
Last edited:
Because the original thing that sparked all of this was a pure mention that the Chromebook is faster than the iPhone 5 at Sunspider.
And the rest followed.
I think here is what sums things up the best:
1) A15, S4, and A6 are all faster than A9, but only under certain circumstances. Hence the "up to" preposition. I'm sure you can break things down in Geekbench and find certain tests where A6 and S4 don't really show any advantage over A9 at all (other than clockspeed). It remains to be seen how A15 fares.
2) Since A6 is only available on iOS, A15 is only available in one device that's basically running a glorified web browser, and S4 is only available on Android, God knows how we can compare them in a more comprehensible manner. So the 3 chips are inherently not comparable for now.
3) ARM has made it clear that A15 is a high performing part without much regard to power consumption, so it remains to be seen whether OEMs will be able to "downsize" A15 to fit into phones and tablets without resorting to companion A7 cores. In that regard, A15 performance will likely suffer, or power consumption will suffer. It's likely we won't see much of a difference when A15 hits. I think we are making A15 look too... magical.
4) It's clear Apple's A6 is at least leading in power consumption compared to other SoCs. This is as a whole for the CPU, GPU and RAM modules among other things. It may or may not have to do with how Apple implemented their Swift core compared to Krait or A15. Basically, I'm saying the power saving benefits of A6 may not come from the CPU alone. It remains to be seen whether the A6X will follow suit, because it's evident A6X will feature increased clock speed over A6 at least for the GPU.
And point 5 brings us back to A6X. Now... disregarding how fast A6 is compared to A15 or S4, how fast do we think A6X would be compared to A6?
Last edited:
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter