Apple A10 Geekbench 4 Score

[Eug]

It's TSMC 7nm.

I understand, but I still don't believe it... unless TSMC is playing even looser with their definition of 7 nm than they are with 10 nm.

To put it in clearer terms, if TSMC 7 nm represents a true full generation die shrink from TSMC 10 nm, then I would be completely shocked to see it in 2018 for the A12 in the iPhone 2018.

 

[witeken]

It's TSMC 7nm.

Your words are eternalized now in my signature. Let's wait and see.

 

[witeken]

So this time TSMC won't come away with it freely without noticing if they start juggling with their roadmap, products and release dates.

 

[Arachnotronic]

I understand, but I still don't believe it... unless TSMC is playing even looser with their definition of 7 nm than they are with 10 nm.

To put it in clearer terms, if TSMC 7 nm represents a true full generation die shrink from TSMC 10 nm, then I would be completely shocked to see it in 2018 for the A12 in the iPhone 2018.

It's not a full generation die shrink, TSMC says it's a 1.6x improvement in logic density from 10nm, IIRC. Full generation would be ~2x. 10nm is supposed to be a full generation shrink from 16FF+, though.

Think 32nm -> 28nm transition.

 

[witeken]

It's not a full generation die shrink, TSMC says it's a 1.6x improvement in logic density from 10nm, IIRC. Full generation would be ~2x. 10nm is supposed to be a full generation shrink from 16FF+, though.

I have only just made my new signature and you're already spinning the roadmaps. TSMC has said 7nm will be a 40-45% shrink, see sig. 40% is 1.67x.

Are you happy with how my sig is (which I have interleaved with some predictions about Intel for comparison), or do you want any changes?

 

[Arachnotronic]

I have only just made my new signature and you're already spinning the roadmaps. TSMC has said 7nm will be a 40-45% shrink, see sig. 40% is 1.67x.

Are you happy with how my sig is (which I have interleaved with some predictions about Intel for comparison), or do you want any changes?

It's your sig, put whatever you want in it

The node naming on TSMC's part is kind of BS, I would say Intel 10nm and TSMC 7nm should be roughly equal in terms of xtor density. The differentiating factors will be xtor performance, something that's probably too early to call though I would probably say Intel will be superior here.

 

[Nothingness]

witeken, I guess that if one would quote your history of already proven wrong Intel predictions, there'd be no room in the signature. But that'd be childish, wouldn't it?

Back to the subject, looking at the various results in Geekbench 4, it looks like frequency is ~2.34 GHz for the larger core and ~.40 GHz for the smaller one.

 

[Arachnotronic]

Back to the subject, looking at the various results in Geekbench 4, it looks like frequency is ~2.34 GHz for the larger core and ~.40 GHz for the smaller one.

I wonder if the smaller core is a custom job or if it's an off the shelf ARM core? My guess is that it's the same as the big core but the physical implementation is done to optimize for very low power (pretty much what Qualcomm is doing with Kryo and what Qualcomm tried with Tegra 3's 4+1).

 

[Eug]

witeken, I guess that if one would quote your history of already proven wrong Intel predictions, there'd be no room in the signature. But that'd be childish, wouldn't it?

Back to the subject, looking at the various results in Geekbench 4, it looks like frequency is ~2.34 GHz for the larger core and ~.40 GHz for the smaller one.

There are also a fair number of 1.05 GHz benches, more than the ones at 0.4 GHz. Interestingly, the single-core performance of some of those is coming at around 1800.

There are also some at 1.6 GHz, as well as some at 1.9 GHz.

 

[witeken]

witeken, I guess that if one would quote your history of already proven wrong Intel predictions, there'd be no room in the signature. But that'd be childish, wouldn't it?

Well, this forum seems to have a tradition of people quoting other people's comments and predictions in their sigs . I haven't made it personal, though.

 

[Arachnotronic]

Well, this forum seems to have a tradition of people quoting other people's comments and predictions in their sigs . I haven't made it personal, though.

I don't take it personally.

 

[Nothingness]

Well, this forum seems to have a tradition of people quoting other people's comments and predictions in their sigs . I haven't made it personal, though.

My bad, I took it as a personal attack against Arachnotronic. We are indeed all doing predictions, and I personally find it funny to look back at what I predicted and how wrong I was

 

[Arachnotronic]

My bad, I took it as a personal attack against Arachnotronic. We are indeed all doing predictions, and I personally find it funny to look back at what I predicted and how wrong I was

I've also made my share of bad (even really bad) predictions.

"It's difficult to make predictions, especially about the future" -- Yogi Berra.

 

[Eug]

It's not a full generation die shrink, TSMC says it's a 1.6x improvement in logic density from 10nm, IIRC. Full generation would be ~2x. 10nm is supposed to be a full generation shrink from 16FF+, though.

Think 32nm -> 28nm transition.

OK fair enough, but still, it seems awfully optimistic. BTW, 32->28 is a half node shrink, and 10 nm -> 7 nm is more than that. I thought I read somewhere that no foundry has ever been able to do two successive full node shrinks in two successive years. 16 nm to 7 nm isn't 2 full nodes, but it's getting relatively close. (Forgive me if I am not using the terminology correctly.)

However, if it happens, perhaps I'll buy a 7nm A12 iPhone in 2018, skipping 10nm entirely.

P.S. It's amazing what we've come to expect for our phones these days. I just realized that my primary Windows desktop (Athlon II X3 435), my primary Mac desktop (Core i7 870), my primary Windows laptop (Pentium SU4100), and my primary Mac laptop (Core 2 Duo P8400) are all still all 45 nm parts.

 

[Arachnotronic]

OK fair enough, but still, it seems awfully optimistic. BTW, 32->28 is a half node shrink, and 10 nm -> 7 nm is more than that. I thought I read somewhere that no foundry has ever been able to do two successive full node shrinks in two successive years. 16 nm to 7 nm isn't 2 full nodes, but it's getting relatively close. (Forgive me if I am not using the terminology correctly.)

However, if it happens, perhaps I'll buy a 7nm A12 iPhone in 2018, skipping 10nm entirely.

P.S. It's amazing what we've come to expect for our phones these days. I just realized that my primary Windows desktop (Athlon II X3 435), my primary Mac desktop (Core i7 870), my primary Windows laptop (Pentium SU4100), and my primary Mac laptop (Core 2 Duo P8400) are all still all 45 nm parts.

Ignore the naming, it's 100% pure marketing BS

 

[krumme]

When even machines like a single 3400 is like 150M its abvious a matter of raw r&d budget. Highend Smartphones sell in huge numbers and still do. The cheaper ones take the rest of the depreciation. If it continues - but frankly i dont know why a A10 is needed in a phone (sans eg better fixed function in dsp) - then Intel will lose its process advantage. Its natural and they are adapting fine.

Gf is just bound to get merged into a bigger partner as its just a pain. What is the meaning ? Seriously it has to stop now. What a waste of ressources.

Mubadala will lose this game to Samsung who have "the galaxy profit" and actual competences to support the cash. And the strategic purpose like apple to stay up front.

 

[mikk]

Well, actually, no it's not that greater challenge, and as Intel correctly stated (Yet Anandtech somehow incorrectly stated), it's a 4.5-45w window, which is nothing special at all (despite their marketing spin) , and even then, it encompasses a large range of IGP spec's eating into that TDP range.

Wrong, it's a 4.5W - 91W window. Check out i7-6700k, same Die as a H-series mobile 45W SKU, only the package is different. Furthermore 4.5W and 91W SKUs are combined with the same iGPU called GT2. Your posting is full of nonsense.

 

[majord]

Wrong, it's a 4.5W - 91W window. Check out i7-6700k, same Die as a H-series mobile 45W SKU, only the package is different. Furthermore 4.5W and 91W SKUs are combined with the same iGPU called GT2. Your posting is full of nonsense.

Last time I checked , Core M was a Dual core, and 6700K was a Quad core. Intel were sensible enough to contain there statement to SKU's with an equal core count,

To be completely accurate one should isolate core power, but cores are not rated on their own.

So please explain further how my post is full of nonsense.

 

[stingerman]

I don't understand why Intel even comes up as a competitor, since it doesn't even compete in this space. It lost it. If the A10 Fusion is being compared to Intel processors on larger form factors that's not good for Intel in any way.

 

[nvgpu]

http://arstechnica.com/apple/2016/0...at-annual-upgrades-with-one-major-catch/5/#h1

Apple utterly dominates & crushes Qualcomm & Samsung in CPU performance. Whatever you think about Apple as a company, you have to admit their CPU design team keeps knocking it out of the park when it comes to performance & power efficiency.

 

[scannall]

What's really surprising, is the single core benchmark gets about the same as my OC'd 5660. At a much lower clock speed, and a tiny fraction of its power consumption.

 

[deasd]

What is funny is that 'Fusion' branding using in A10, which is already ditched by AMD since several years ago. XD

 

[stingerman]

Mubadala will lose this game to Samsung who have "the galaxy profit" and actual competences to support the cash. And the strategic purpose like apple to stay up front.

I understand that "the galaxy profit" is exploding these days, one can even say they're on fire

 

[stingerman]

What's really surprising, is the single core benchmark gets about the same as my OC'd 5660. At a much lower clock speed, and a tiny fraction of its power consumption.

That's some scary stuff, maybe Apple developed a time machine and is bringing back technology from the future

 

[witeken]

What's really surprising, is the single core benchmark gets about the same as my OC'd 5660. At a much lower clock speed, and a tiny fraction of its power consumption.

Just shows how unreliable cross OS comparisons are. For all we know Apple could optimize their CPUs for benchmarks and no one would ever know.

I haven't seen other benchmarks yet, but I will take a look when AT publishes their review. For sure I want to know what they have done with the architecture.