Apple A9X Geekbench

[teejee]

I hope you mean turning point in the good way. I dunno, they should announce something revolutionary.

Like, "we have a real, silicon quantum computer, not the limited quantum annealing thing from DWave" sort of revolutionary. Or some kind of NVM with unlimited endurance and speed like SRAM and NAND density and cost. Or on chip silicon photonics. Or III-V and Ge with Quantum Wells and with secret EUV ("haha we were actually joking about EUV for 7nm, we actually DID something with the 5B we invested in ASML") and 450mm ("yeah, that's why we were spending so many billions of Capex for many years"). Or a new architecture, like that hyped RISC thing ("you thought we delayed 10nm, but that was actually to pull in EUV and 450mm, but as compensation we've got something totally uninteresting: Kabylake with twice the IPC, and double the frequency at 10nm"). Or heck, even near threshold voltage computing. Or a neural chip like IBMs TrueNorth. Or 3D logic.

Guess reality won't be that spectacular (cf. Larrabee) and the lack of revenue growth will just let other companies like TSMC catch up while they reap the benefits of the mobile industry's money.

Well, I meant the turning point where Intels dominance is going away.
Not because I dislike Intel but because of the reasons you mention. I want something new to happen. There are so many possibilities other than silicon semiconductors.
But Intel today is too comfortable just cashing in...

 

[Arachnotronic]

I hope you mean turning point in the good way. I dunno, they should announce something revolutionary.

Like, "we have a real, silicon quantum computer, not the limited quantum annealing thing from DWave" sort of revolutionary. Or some kind of NVM with unlimited endurance and speed like SRAM and NAND density and cost. Or on chip silicon photonics. Or III-V and Ge with Quantum Wells and with secret EUV ("haha we were actually joking about EUV for 7nm, we actually DID something with the 5B we invested in ASML") and 450mm ("yeah, that's why we were spending so many billions of Capex for many years"). Or a new architecture, like that hyped RISC thing ("you thought we delayed 10nm, but that was actually to pull in EUV and 450mm, but as compensation we've got something totally uninteresting: Kabylake with twice the IPC, and double the frequency at 10nm"). Or heck, even near threshold voltage computing. Or a neural chip like IBMs TrueNorth. Or 3D logic.

Guess reality won't be that spectacular (cf. Larrabee) and the lack of revenue growth will just let other companies like TSMC catch up while they reap the benefits of the mobile industry's money.

FYI, even though 14nm has been kind of a bust for Intel and 10nm is looking in rough shape too atm, Intel has some very, very cool stuff coming at 7nm. Just be patient.

 

[jpiniero]

What I was pleasantly surprised to see was that the A9X seems to be drawing similar to the Air 2 as opposed to Core M. I am still a little surprised to see dual core though given the split screen mode.

 

[witeken]

Intel has some very, very cool stuff coming at 7nm. Just be patient.

Couldn't be vaguer. And the future always looks "very, very cool". Remember when back at the beginning of 2013, it seemed Silvermont would obliterate everything that that was smartphone or tablet from ARM. And it would be 64-bit. Oops, didn't see Cyclone coming.

Or at the investor meeting 2014. Intel had good year and thought 2015 would be just as well. Of course there are always thing you couldn't have seen coming.

And when will 7nm be released? Before 2020? The Kabylake delay to end-of-2017-but-actually-just-start-of-2018 doesn't predict good stuff.

 

[Arachnotronic]

Couldn't be vaguer. And the future always looks "very, very cool". Remember when back at the beginning of 2013, it seemed Silvermont would obliterate everything that that was smartphone or tablet from ARM. And it would be 64-bit. Oops, didn't see Cyclone coming.

Cyclone was awesome but I think if Intel had developed more competitive SoCs in the 22nm generation it could have gotten the ball rolling on some share gains against MediaTek and maybe even Qualcomm. Merrifield was just a loser and Moorefield was really late.

Or at the investor meeting 2014. Intel had good year and thought 2015 would be just as well. Of course there are always thing you couldn't have seen coming.

Eh, the PC market is kind of a really bad neighborhood to be in. During 2014 there weren't too many drive-by shootings but in 2015 there were several.

And when will 7nm be released? Before 2020? The Kabylake delay to end-of-2017-but-actually-just-start-of-2018 doesn't predict good stuff.

I expect the first 7nm silicon from Intel to arrive in 2020.

 

[videogames101]

Couldn't be vaguer. And the future always looks "very, very cool". Remember when back at the beginning of 2013, it seemed Silvermont would obliterate everything that that was smartphone or tablet from ARM. And it would be 64-bit. Oops, didn't see Cyclone coming.

Or at the investor meeting 2014. Intel had good year and thought 2015 would be just as well. Of course there are always thing you couldn't have seen coming.

And when will 7nm be released? Before 2020? The Kabylake delay to end-of-2017-but-actually-just-start-of-2018 doesn't predict good stuff.

We got FinFET in 2012, was that not "very very cool"? I would expect something similar, new device shape or material. As far as computer architecture, well even Cyclone wasn't that exciting. I don't expect more than 10% IPC every 2 years from architecture changes alone at a given TDP. (barring a major design breakthrough)

 

[witeken]

I expect the first 7nm silicon from Intel to arrive in 2020.

And why will that be very, very cool?

 

[Arachnotronic]

And why will that be very, very cool?

From the very limited information that I have been able to learn about Intel's 7nm plans, the materials changes they will be doing are going to be very exciting. I don't have specifics sadly.

 

[witeken]

From the very limited information that I have been able to learn about Intel's 7nm plans, the materials changes they will be doing are going to be very exciting. I don't have specifics sadly.

III-V? Spintronics? Vertical wire / CNT?

 

[Arachnotronic]

III-V? Spintronics? Vertical wire / CNT?

Not a clue.

 

[witeken]

Not a clue.

Okay. That looks cool indeed. Although "futuristic" is both ridiculously vague and a truism.

http://www.fool.com/investing/gener...massive-plans-for-its-7-nanometer-proces.aspx

I do like the R&D graph, though. It says exactly what I've been wanting to say earlier today, and it's very cool that throughout 2015, it has continued to rise. I really wonder where all those dollars go to. I hope they're not just the dollars from all the bad chips from the poor 14nm yields . But I guess a non-insignificant amount is for multiple patterning (and also for transistor and interconnect, I hope), which will also be necessary with EUV.

Just too bad this is all in the far future, and Intel isn't the only company with long-tern R&D, although the 3D XPoint seems to have a reasonable TTM advantage. Just like silicon photonics and the QPI thing. HMC is late, though.

 

[Thanatosis]

Okay. That looks cool indeed. Although "futuristic" is both ridiculously vague and a truism.

http://www.fool.com/investing/gener...massive-plans-for-its-7-nanometer-proces.aspx

I do like the R&D graph, though. It says exactly what I've been wanting to say earlier today, and it's very cool that throughout 2015, it has continued to rise. I really wonder where all those dollars go to. I hope they're not just the dollars from all the bad chips from the poor 14nm yields . But I guess a non-insignificant amount is for multiple patterning (and also for transistor and interconnect, I hope), which will also be necessary with EUV.

Just too bad this is all in the far future, and Intel isn't the only company with long-tern R&D, although the 3D XPoint seems to have a reasonable TTM advantage. Just like silicon photonics and the QPI thing. HMC is late, though.

R&D won't save you from poor design decisions or a board of executives that can't seem to understand how important mobile performance is.



Don't fret though, intel isn't going anywhere. They still have a big lead in datacenter performance and that sweet sweet captive x86 server market. Intel can afford to become irrelevant in the consumer space, yet still bring in profits, for a long while. Perhaps forever.

 

[stingerman]

Intel chose to use doped silicon and jumped into FinFet with it, whereas others delayed FinFet until they finished R&Ding new materials which they thought would be necessary down to 7NM. Now Intel needs to spend the time to bring on newer, faster materials. While they have been able to go down to 14NM, they haven't benefitted from significant performance increases.

IBM's approach delayed their jump into FinFet, but now Samsung and Global Foundries are using IBM's technology which includes the faster SiGe versus Intel's cheaper doped silicon. So intel benefited earlier on, leapfrogged AMD but now is paying the price giving chance for others, including TSMC to catch up and possibly excel.

 

[stingerman]

It's amazing how good the A9 and A9X are. Finally, some real competition on the CPU side. Forget AMD, it will be Apple and (hopefully Samsung+Qualcomm etc) that push Intel to actually bother with performance.

Makes you wonder about the A10X. If it shows strong gains again...will they replace their 13" MB Pros and MB Air's with their own CPU's?

Intel must be sweating. If Apple can do it, I have faith that Samsung etc can do it as well. ARM must be loving this.

Intel isn't intentionally being sluggish, they have to move to newer materials. The decision they made to stick to doped Silicon for their 22NM FinFet process forward gave them a greater lead time over their competitors and initially allowed them to leapfrog AMD and thus Global Foundries into FinFet, a technology they did not develop but was invented by the current TSMC chair at Berkley who served as TSMC's CTO from 2001 - 2004. It allowed Intel to basically make relatively minor upgrades to their fabs. At 14NM they saw a much lower performance boost then they hoped and now are upping their investment in R&D to fix the problem they are having.

 

[Edrick]

And when will 7nm be released? Before 2020? The Kabylake delay to end-of-2017-but-actually-just-start-of-2018 doesn't predict good stuff.

Sorry, I missed that. When was that delay announced? I thought Cannonlake was pushed back to 2017-2018.

 

[witeken]

Sorry, I missed that. When was that delay announced? I thought Cannonlake was pushed back to 2017-2018.

Cannonlake was pushed back to H2'17 during Q2 earnings call.

According to Intel, there should be millions of 10nm in the second half of that year. However, rumor suggests that Kabylake, the second Tick between SKL and CNL, will only be released at the very end of December, so if you count one year from KBL to CNL, we're basically 2018...

The investor meeting is next week or so, so we should get more clarity then. According to the CFO commentary, 10nm spending will start this quarter, which would be a positive sign of CNL in H2'17.

 

[kimmel]

Sorry, I missed that. When was that delay announced? I thought Cannonlake was pushed back to 2017-2018.

It wasn't. People are just making things up (speculating) on the internet again.

 

[dahorns]

3dMark benches for the A9X:

Ice Storm Unlimited
A9X:
Overall - 33,467
Physics - 15,2161
Graphics - 50,795
http://www.futuremark.com/hardware/mobile/Apple+iPad+Pro/review

A8x:
Overall - 21,674
Physics - 10,353
Graphics - 31,524
http://www.futuremark.com/hardware/mobile/Apple+iPad+Air+2/review

Core M3 (I think this is SP4):
Overall - 40,995
Physics - 22,842
Graphics - 53,039
http://www.3dmark.com/3dm/9219807
https://www.youtube.com/watch?v=M4EmpbCQ5cY

Core M 5y71 (unclear on TDP setting, Yoga Pro 3):
Overall - 49,619
Physics - 31,473
Graphics - 59,405
http://www.anandtech.com/show/9061/lenovo-yoga-3-pro-review/5

 

[sm625]

3dMark benches for the A9X:

Ice Storm Unlimited
A9X:
Overall - 33,467
Physics - 15,2161
Graphics - 50,795
http://www.futuremark.com/hardware/mobile/Apple+iPad+Pro/review

A8x:
Overall - 21,674
Physics - 10,353
Graphics - 31,524
http://www.futuremark.com/hardware/mobile/Apple+iPad+Air+2/review

Core M3 (I think this is SP4):
Overall - 40,995
Physics - 22,842
Graphics - 53,039
http://www.3dmark.com/3dm/9219807
https://www.youtube.com/watch?v=M4EmpbCQ5cY

Core M 5y71 (unclear on TDP setting, Yoga Pro 3):
Overall - 49,619
Physics - 31,473
Graphics - 59,405
http://www.anandtech.com/show/9061/lenovo-yoga-3-pro-review/5

I wish they would do battery life testing by looping Ice Storm Unlimited.

 

[Nothingness]

I wish they would do battery life testing by looping Ice Storm Unlimited.

FWIW there's a WebGL battery life test in arc technica review. Battery life of the iPad Pro is 359 minutes vs 217 minutes for the Surface Pro 4.

 

[dahorns]

I wish they would do battery life testing by looping Ice Storm Unlimited.

You'd have to normalize for performance to have it mean much.

FWIW there's a WebGL battery life test in arc technica review. Battery life of the iPad Pro is 359 minutes vs 217 minutes for the Surface Pro 4.

I'm assuming that is the 6300u SP4, so the test shouldn't be that surprising given the higher wattage processor. Again, you'd need to normalize for performance to really understand what a test at load like that means. For instance, the Ipad Pro and the SP4 basically scored the same battery life in the wifi-browsing test.

May be interesting to take something like Ice Storm and frame cap it and then run it in a loop.

 

[Mondozei]

Reading this thread made me chuckle. I posted a thread earlier this year where I basically laid out the possibility of Apple abandoning Intel.

If you read the replies there today, even if it has only been a months in passing, you'll see the wall-to-wall denial that it could ever happen.

One of the few people who bought it then was raghu78, because he has seen the massive performance increases coming for a long time and never bought into the "Intel is unbeatable" hype.

Just as I didn't buy into the conventional wisdom that Intel would crush everyone in mobile because their "node leadership is unsurpassable". I guess it is.

Apple may or may not replace Intel in their laptops, but we're no longer having a discussion of whether or not their CPUs are powerful enough for it. They already are. The software side would take longer, but anyone who thinks such a hindrance is "insurmountable" hasn't learned anything these past few years.

The major question to me is if Apple continues with PowerVR or if their own in-house GPU efforts take fruit next generation with A10/A10X. So far, PowerVR has been surprisingly capable. Let's see if it can last.

 

[Arachnotronic]

Reading this thread made me chuckle. I posted a thread earlier this year where I basically laid out the possibility of Apple abandoning Intel.

If you read the replies there today, even if it has only been a months in passing, you'll see the wall-to-wall denial that it could ever happen.

One of the few people who bought it then was raghu78, because he has seen the massive performance increases coming for a long time and never bought into the "Intel is unbeatable" hype.

Just as I didn't buy into the conventional wisdom that Intel would crush everyone in mobile because their "node leadership is unsurpassable". I guess it is.

Apple may or may not replace Intel in their laptops, but we're no longer having a discussion of whether or not their CPUs are powerful enough for it. They already are. The software side would take longer, but anyone who thinks such a hindrance is "insurmountable" hasn't learned anything these past few years.

The major question to me is if Apple continues with PowerVR or if their own in-house GPU efforts take fruit next generation with A10/A10X. So far, PowerVR has been surprisingly capable. Let's see if it can last.

Cool, since the A9X is at this level of performance, the next MacBook should be A9X based right?

 

[witeken]

Mondozei brings up a good point.

Apple releases a new SOC every year. Even A8 had a pretty nice boost. However, Intel does the opposite thing. So while Apple will release A10(X) with whatever improvements, Intel will have... Kabylake. Which will simply introduce GT4e as far as I understand. And maybe 5% IPC boost.

 

[TechGod123]

Cool, since the A9X is at this level of performance, the next MacBook should be A9X based right?

Probably A10X if it happens.