The Intel Atom Thread

[Imouto]

How in the world will a 2 Ghz Turbo Silvermont have better single thread performance than a 2.4 Ghz one?

 

[Sweepr]

How in the world will a 2 Ghz Turbo Silvermont have better single thread performance than a 2.4 Ghz one?

If you look closely you will notice BT-T (up to 2.4GHz Turbo) does win 2 of those Javascript/HTML5 benchmarks by a respectable margin. A dual-core chip with 1.8-2.0GHz base clock and similar Turbo clock to Z3770 (2.2-2.4GHz) would be competitive (if that reflects overall performance). 🙂

 

[thunng8]

lol. The A7 wins exactly 1 benchmark and none of them are multi-threaded. Baytrail-T is a quad-core. I admit that single-threaded performance is more important but it's more or less a tie there while BT would destroy A7 in a multi-threaded benchmark since it has double the cores.

And we do not have any battery life numbers yet but due to OS these will also not be fully comparable.

Well according to geekbench 3 the a7 destroys bay trail in single threaded mode (977 vs 1414) and almost matches in multithread (2920 vs 2570)

We are also talking about a tablet platform that turbos to 2.4ghz vs 1.3 ghz for the a7.

 

[monstercameron]

Well according to geekbench 3 the a7 destroys bay trail in single threaded mode (977 vs 1414) and almost matches in multithread (2920 vs 2570)

We are also talking about a tablet platform that turbos to 2.4ghz vs 1.3 ghz for the a7.

geekbench cross uarch, cross os doesnt compare well between x86 and arm.

 

[StrangerGuy]

lol. The A7 wins exactly 1 benchmark and none of them are multi-threaded. Baytrail-T is a quad-core. I admit that single-threaded performance is more important but it's more or less a tie there while BT would destroy A7 in a multi-threaded benchmark since it has double the cores.

And we do not have any battery life numbers yet but due to OS these will also not be fully comparable.

Phone SoC in an actual soon to be bestselling phone versus an Intel designed reference tablet...

Feel free to ignore context next time, or GPU power for that matter.

 

[liahos1]

You're mixing architecture with CPU IP. You should read about ARM 😉

sorry to clarify they introduced a53/a57 in mid 2012. 18 months later we have a7 which is based on v8 ip. since then they introduced a12 (based on v7) and nothing past a53/a57 from a performance point of view.

 

[Nothingness]

sorry to clarify they introduced a53/a57 in mid 2012. 18 months later we have a7 which is based on v8 ip. since then they introduced a12 (based on v7) and nothing past a53/a57 from a performance point of view.

Cortex-A7 still is an ARMv7-A chip not an ARMv8 one.

 

[liahos1]

Cortex-A7 still is an ARMv7-A chip not an ARMv8 one.

a7 from apple, not cortex a7.

 

[Nothingness]

a7 from apple, not cortex a7.

Oh, that's the price to pay for mixing names 😛

Why are you comparing what ARM Ltd does against what Apple does? Apple designs chips for its own use only, while ARM Ltd designs CPU that must be sold to many partners with varying needs. Cyclone is a high-end chip, A12 is a middle range chip, not the same markets. In fact it's like saying Bay Trail stinks because Haswell is so much more powerful 😉

 

[thunng8]

geekbench cross uarch, cross os doesnt compare well between x86 and arm.

What is the use of geekbench if it you can't compare? They run through exactly the same tests whether it be Arm or x86

 

[liahos1]

Oh, that's the price to pay for mixing names 😛

Why are you comparing what ARM Ltd does against what Apple does? Apple designs chips for its own use only, while ARM Ltd designs CPU that must be sold to many partners with varying needs. Cyclone is a high-end chip, A12 is a middle range chip, not the same markets. In fact it's like saying Bay Trail stinks because Haswell is so much more powerful 😉

sorry my post is a bit confusing. What im basically saying is this

- A7 (apple) using a custom architecture built off of v8 performs on par with bay trail

- Qualcomm likely wont have v8 based product until 2h of next year

Assuming the apple a7 performs better than vanilla v8 cores (a53/a57) intel should still be able to compete against v8 based offerings with bay trail. But that comp is not the right comp based on projected timing. the right comp is V8 cores vs Airmont cores (which we havent seen anything of yet).

There has been nothing out of ARM itself (yet) pointing to revisions to v8 past a53/a57 so i dont know what 2015 looks like.

As far as im concerned for the android camp, the best game in town is still baytrail until v8 based soc's come out. and even then they will be at par with baytrail.

 

[raghu78]

This. The fact that the A7 comes close to BT-T performance @ ST is impressive, but BT-T delivers that kind of ST with twice the number of cores, I doubt even a slightly higher-clocked A7X would come close to it in MT tasks. We need more benchmarks to compare both but I'm guessing a >2GHz dual-core Silvermont would still offer better single-thread performance (overall) than the A7, which means Merrifield (2C Silvermont + Rogue GPU) would be competitive in both CPU and GPU fronts.

BT-T is running at max turbo clocks of 2.4 Ghz in most apps which load only the CPU. only when you run a 3D game which taxes both CPU and GPU does Baytrail-T run at 1.46 Ghz. even when running Cinebench R11.5 multi thread Baytrail runs at close to max Turbo. you can derive that from the scaling from 1 to 4 threads.

http://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested/2

Cyclone is a faster core than Silvermont. at 1.3 Ghz its competing with Baytrail at 2.4 Ghz. Cyclone has a full 128 bit FPU. Silvermont has a 64 bit FPU. Baytrail looks impressive because of aggressive turbo.

Merrifield is a dual core chip. So if it runs at 2.4 Ghz max Turbo and is running close to max turbo under workloads which do not tax the GPU then it can keep up with a A7 running at 1.3 Ghz. otherwise Merrifield will be slower.

 

[trivik12]

if qualcomm can upgrade their core to bring in same level of improvements then intel is definitely in a for a big challenge. May be Intel will be aggressive in getting Airmont/Goldmont out at the earliest. Problem for intel is if atom performance gets too good then it impacts the sales of core series which I am sure they dont want.

 

[Nothingness]

sorry my post is a bit confusing. What im basically saying is this

This post was definitely clearer, thanks 😉

 

[Sweepr]

Actual Bay Trail-T score @ SunSpider 1.0: 329.6ms

http://www.youtube.com/watch?v=UXCobb134Vo

That makes (up to 2.4GHz) Bay Trail-T >20% faster than 1.3GHz A7 in the 3 Javascript/HTML5 single-thread benchmarks tested by Anand. If they can run Merrifield above 2GHz coupled with a similar PowerVR Rogue GPU (+competitive power levels) then Intel will offer an interesting A7 alternative to Android phones.

 

[dahorns]

Well according to geekbench 3 the a7 destroys bay trail in single threaded mode (977 vs 1414) and almost matches in multithread (2920 vs 2570)

We are also talking about a tablet platform that turbos to 2.4ghz vs 1.3 ghz for the a7.

1. You are comparing the 64bit single thread score of a7 vs 32bit single thread score of BT. The 32bit numbers are pretty much dead even. The question is: once Microsoft corrects Windows 8 in 1st half of next year, will BT see a similar jump in 64bit performance?

2. No doubt the a7 is very impressive, especially if it runs max at 1.3 ghz. However, we don't know that to be the case yet do we?

3. We don't know power consumption numbers.

4. It certainly possible that a Dual Core Silvermont design running 64bit could equal (not surpass) the a7. In fact, the numbers seem to indicate that it would, especially if the same jump happens at 64bit.

At any rate, it is obviously a great chip and a class leader SOC for phones and probably tablets.

 

[Khato]

Before declaring defeat for Intel I'll wait to see what A7 power consumption under load is like. I'd hope that the Haswell macbook air demonstrating superior battery life to the iPad 4 has taught the lesson of battery life under extremely light load having no correlation to full load power consumption.

That said, the A7 is definitely a nice design, basically a refinement of the groundwork that they laid with the A6. Which makes sense on such a short development cycle, just make iterative improvements on the design you already have so long as it's a good foundation. Will be quite interesting to see where they go from here - my guess is that the next ~2x CPU performance is just going to be from doubling the number of cores.

 

[krumme]

If the results is representive its just a slaughter. Hands down. No need to wrap it into anything else. No where did i see that comming thats for sure.

What is interesting is then where did this performance come from? What is attributed to eg new compiler gains from v8 whats is xx?

 

[shyam334]

Hello!
I'd probably wait for a finished device aka shipping product for the scores 64bit baytrail. We don't have enough public data on how the turbo works in BT against power consumption.

And after the antutu fiasco, I don't really trust these benchmarks to compare across uarch.

 

[Shivansps]

personally i dont care about A7 because ill never buy anything from Apple, thats the botton line for me.
But is an idea on how A57 arm chips will perform in the future.

 

[Blandge]

I think it's also important to keep in mind that Apple has a really tight software stack, so it's difficult to compare the two SoCs. Especially with something as software sensitive as Sunspider

 

[Sweepr]

Bay Trail-T score @ WebXPRT: 574 +/- 8

Anand iPhone 5S review:

 

[Shivansps]

Again it does really matter if it beat the A7 or not? it will be more important when A57 ARM chips are around, and by that time, Airmont will be close. People that buy Apple will be keeping buying Apple and for the people that dont will be take more than that.

 

[Khato]

Again it does really matter if it beat the A7 or not? it will be more important when A57 ARM chips are around, and by that time, Airmont will be close.

Quite true. Right now there are still too many unknowns beyond the performance benchmarks we do have to get too excited. Both Apple's A7 and Intel's Baytrail-T look quite promising though. Hopefully the picture will flesh out in the next few weeks.

 

[Enigmoid]

Bay Trail-T score @ WebXPRT: 574 +/- 8

Anand iPhone 5S review:

Actual Bay Trail-T score @ SunSpider 1.0: 329.6ms

http://www.youtube.com/watch?v=UXCobb134Vo

That makes (up to 2.4GHz) Bay Trail-T >20% faster than 1.3GHz A7 in the 3 Javascript/HTML5 single-thread benchmarks tested by Anand. If they can run Merrifield above 2GHz coupled with a similar PowerVR Rogue GPU (+competitive power levels) then Intel will offer an interesting A7 alternative to Android phones.

Easily see that intel still has some work to do with software. Nowhere is this more obvious that in the disproportionate scaling between baytrail and ivy bridge HD 4000 where adding 4x the EU and increasing frequency (~20-40%) only increases performance about 2.75-3.2x.

Geekbench numbers are flawed unless you are comparing very similar architectures.

Before declaring defeat for Intel I'll wait to see what A7 power consumption under load is like. I'd hope that the Haswell macbook air demonstrating superior battery life to the iPad 4 has taught the lesson of battery life under extremely light load having no correlation to full load power consumption.

That said, the A7 is definitely a nice design, basically a refinement of the groundwork that they laid with the A6. Which makes sense on such a short development cycle, just make iterative improvements on the design you already have so long as it's a good foundation. Will be quite interesting to see where they go from here - my guess is that the next ~2x CPU performance is just going to be from doubling the number of cores.

Look at the AT battery life tests. There is almost no difference when battery capacities are normalized between the 5 and the 5s.

5c is more efficient and lasts longer despite a smaller battery.

Not really more efficient than 5c.

Quite a bit worse than 5c.

Similar to 5c, 4% larger battery, lasts 13% longer on a more efficient GPU.

Does not really look like the 5s is more efficient than the 5c at low level CPU tasks. GPU is obviously more efficient but the CPU seems to be consuming a bit more power. Bit disappointing considering the move to 64 bit, 28 nm from 32 nm, and a more advanced CPU architecture.

The rest of the phone excluding RAM is very similar.