Apple's A5 processor could pave the way for larger chip trend

2456

Comments

  • swiftswift Posts: 436member
    Quote:
    Originally Posted by extremeskater View Post


    Apple didn't build the A5 chip, Samsung built the A5.



    Apple DESIGNED the A4 and A5. Who makes the chip? Pick the ones with the best fab.
  • shrikeshrike Posts: 494member
    Quote:
    Originally Posted by EgisCodr View Post


    I guess you guys don't realize that the Tegra 2 is last year's chip. You are comparing the A5 to a chip from 2010. The next chip will be out in products in June. Code-named Kal-El, it is the first quad-core processor on the market. Five times faster than the Tegra 2.



    That is THIS YEAR's chip. Cannot wait to see how the A5 stacks up.



    Apple would be better served buying the chips from Nvidia. By 2014, Nvidia will have chips 100 times faster than Tegra 2. This is an arena Apple will not be able to keep up in.



    That's just a little bit of exaggeration there, man. Tegra 3 is 5x times faster than Tegra 2? And June? Nvidia ARM SoCs will be 100x faster in 2014?



    How about Tegra 3 will be out in 2H 2011. The rumors are saying August, not June. There are no known design wins for it yet. They better announce one soon if a product is going to ship soon. In system throughput, a 1 GHz Tegra 3 has the potential to be 2x faster than a 1 GHz Tegra 2, and has potential to be 5x faster in GPU than Tegra 2.



    In 2014, 3 years away, ARM SoCs will be at 18nm to 20nm process nodes, representing 4x to 5x the transistor count from Tegra 2. This will translate to about 2x to 3x the single-threaded performance (not many low hanging fruit for single threaded performance after the A5) and 5x to 10x the multi-threaded performance (quad-core with 2-way SMT). GPU will be much more power consumption limited than desktop hardware GPU are. And I think 5x to 10x is the most one could expect as well. You likely can have 100x GPU performance, desktop SLI rigs today are probably already there, but I doubt they are going to fit in handheld products.



    Nvidia's press release roadmap is great and all, but it's a press release roadmap. You should inject some realism instead of propagandizing it.



    Tegra 3 is going to be built on the same 40 nm process as Tegra 2. The thing has about 50% more transistors then Tegra 2. It'll need to be operating at ~25% lower voltages and be power-gated up the wazoo to fit inside the same power envelope (~0.5 Watts) as the Tegra 2 in order to fit inside handhelds. People should be skeptical that they could do this, and quad-core SoCs probably won't fit until the 28nm to 32nm nodes. And that means 2012.
  • shrikeshrike Posts: 494member
    Quote:
    Originally Posted by Swift View Post


    Apple DESIGNED the A4 and A5. Who makes the chip? Pick the ones with the best fab.



    Well, it's really something like pick the fab that offers the best price that meets the specified requirements (X millions of chips/month, Y dollars/chip). That doesn't translate to the best fab.



    The best fabs belong to Intel. And obviously, Intel doesn't fab any ARM chips.
  • elliots11elliots11 Posts: 270member
    Quote:
    Originally Posted by EgisCodr View Post


    I guess you guys don't realize that the Tegra 2 is last year's chip. You are comparing the A5 to a chip from 2010. The next chip will be out in products in June. Code-named Kal-El, it is the first quad-core processor on the market. Five times faster than the Tegra 2.



    That is THIS YEAR's chip. Cannot wait to see how the A5 stacks up.



    Apple would be better served buying the chips from Nvidia. By 2014, Nvidia will have chips 100 times faster than Tegra 2. This is an arena Apple will not be able to keep up in.



    Well seeing as the chip isn't even released yet, you want to enlighten us as to how Apple is supposed to put it in its iPad? You understand that Apple's "Time Machine" is just back up software with a spiffy name, right?



    In all seriousness, if this "kal-el" chip (which is nerdy as hell if its not an April Fools joke) is worth a damn, Apple will improve upon it and stick it in the iPad 3. I'm sure there'll be a number of spec list wonderbombs with slapdash software and this chip on the market that will have failed by the time Apple comes out with iPad 3.
  • swiftswift Posts: 436member
    Quote:
    Originally Posted by EgisCodr View Post


    I guess you guys don't realize that the Tegra 2 is last year's chip. You are comparing the A5 to a chip from 2010. The next chip will be out in products in June. Code-named Kal-El, it is the first quad-core processor on the market. Five times faster than the Tegra 2.



    That is THIS YEAR's chip. Cannot wait to see how the A5 stacks up.



    Apple would be better served buying the chips from Nvidia. By 2014, Nvidia will have chips 100 times faster than Tegra 2. This is an arena Apple will not be able to keep up in.



    It showed at last year's CES. There are a couple of half-assed tablets on the market now with that chip. Not before this year, though they've told us all multiple times that it was coming. You can tell the difference between "shipping" and "in the design stages," can't you?



    A quad chip? Will it finally run Flash?
  • wizard69wizard69 Posts: 11,714member
    There are several things the article didn't consider.



    1.

    The article didn't take into account Apples FAST logic that trades space for lower power and high performance.



    2.

    The CPU is assembled in a module that stacks RAM on top of it. The chips need to line up properly to accomplish this.



    3.

    No body has a sound idea as to what is actually included on A5. Chipworks can guess at possible functionality but they won't get 100% of the functionality. So the question is how much of this comparison is one to one.



    4.

    I will take a trade off that gives me better performance and lower power any day. Customer satisfaction means more than chip size!!!!
  • shrikeshrike Posts: 494member
    Remember, the reference articles are about a NVDA (Nvidia) stock analyst defending his rating on NVDA. The stock has lost 30% of its value since February, and other analysts attribute it to the iPad 2 basically crushing the Xoom, or that the Xoom came out too early, whatever.



    But that's just nonsensical. Apple's A5 doesn't compete with Nvidia's Tegra. If people haven't noticed, Apple doesn't sell the A5 and the A5 can't take any design wins away from Tegra.



    Apple's iPad 2 can take customer wins away from Moto Xoom or LG G-Slate or Samsung Galaxy Tab 8.9/10.1. So there is an indirect competition. But Tegra's real competition is TI OMAP 44xx, Qualcomm Snapdragon, and Samsung Exynos (and others), the chips that power other Android handheld devices.



    Circumstantial evidence seems to say that NVDA got a nice ramp up because: Google is baselining Tegra 2 for Honeycomb, Tegra 2 has a lot of tablet and cell phone design wins, and this all occured during trade show season from December to February. NVDA started falling when: it was apparent Honeycomb wasn't ready, Tegra 2 phones so far have be uninspiring (Moto Atrix and LG Optimus 2x), Tegra 2 tablets have been uninspiring, and the Verizon Driod Bionic (a CDMA Moto Atrix) is rumored to be late. In the meanwhile, Apple released the iPad 2 which seems to be taking the wind out of the sails of competitor tablets.



    So all this is just a stock analyst defending has reputation or defending his stock.
  • granmastakgranmastak Posts: 298member
    Tostitos with a hint of lime! Mmmmm

    What's funny to me is that all these knuckleheads are so green with Apple envy they will find just about any straws to grasp on.



    I got news for all these clowns: what matters is the entire product and that somehow eludes them. Despite of blockbuster sales of the ipad2 and incredible demand, the knuckleheads keep pontificating and littering the web with these mindless rants. From Dell to Microsoft to this chip "expert". I say keep up the comedy
  • solipsismsolipsism Posts: 25,726member
    Quote:
    Originally Posted by wizard69 View Post


    There are several things the article didn't consider.



    1.

    The article didn't take into account Apples FAST logic that trades space for lower power and high performance.



    2.

    The CPU is assembled in a module that stacks RAM on top of it. The chips need to line up properly to accomplish this.



    3.

    No body has a sound idea as to what is actually included on A5. Chipworks can guess at possible functionality but they won't get 100% of the functionality. So the question is how much of this comparison is one to one.



    4.

    I will take a trade off that gives me better performance and lower power any day. Customer satisfaction means more than chip size!!!!



    Good points.
  • shrikeshrike Posts: 494member
    Quote:
    Originally Posted by ksec View Post


    It isn't so much about die size at all. 45nm is roughly 25% larger then 40% i best case scenario. @ 91.5 mm2 A5 is still double the size of Tegra 2.



    There are lots of good reasons for the A5 to be larger than Tegra 2:



    1. PowerVR SGX 543MP2 has 4x the performance as Tegra 2 ULP GeForce. TANSAAFL. That probably means 4x the transistors.



    2. Apple's A5 appears to implement a pipelined FPU while the Tegra 2 doesn't. That's more transistors.



    3. Apple has likely contracted Samsung to fab the A5 with less metal layers than Nvidia has asked TSMC. This can result in larger die sizes for the same transistor count.



    You're absolutely right though. Cost-wise we have no idea. There are good reasons for the A5 to be cheaper.



    1. Samsungs 45 nm fab is mature and has much better yield than TSMC's 40 nm.



    2. Larger die size may imply less metal layers which imply better yield than what Nvidia has asked TSMC to do with 40 nm.



    3. Older fab may mean cheaper product!



    Quote:

    It is about the pricing and economy. Both Nvidia and Apple's Chips costs in $2x range. Not what the article state Tegra 2 are in $15 range, as far as i know even high volume clients and in the very low end $20 per chip.



    Actually, we have no idea how much an A5 costs Apple. Absolutely none. Apple themselves may not even be able to separate an actual real live cost out of it. Apple basically has billions of dollars in contracts with Samsung for CPU-GPU chips, flash chips, RAM chips, LCDs, and who knows what else.



    They may have a specific contract for the A5 SoC (which includes the MCM packaging of the RAM chips) so they could have the precise number, but then again, Apple may have amortized the cost of Samsung components over multiple components! Who knows.
  • ssampathssampath Posts: 10member
    Quote:
    Originally Posted by Shrike View Post


    Remember, the reference articles are about a NVDA (Nvidia) stock analyst defending his rating on NVDA. The stock has lost 30% of its value since February, and other analysts attribute it to the iPad 2 basically crushing the Xoom, or that the Xoom came out too early, whatever.



    But that's just nonsensical. Apple's A5 doesn't compete with Nvidia's Tegra. If people haven't noticed, Apple doesn't sell the A5 and the A5 can't take any design wins away from Tegra.



    I think the question is what value add does Nvidia provide in the Arm food chain if an upstart (in the chip business) like Apple can outdo them with the A5. Also there is at least 4 other companies doing Arm chips - Marvel, Samsung, Ti and Qualcomm. It is not clear to me if this is going to be a great business for Nvidia.
  • solipsismsolipsism Posts: 25,726member
    Quote:
    Originally Posted by Shrike View Post


    3. Apple has likely contracted Samsung to fab the A5 with less metal layers than Nvidia has asked TSMC. This can result in larger die sizes for the same transistor count.



    Excellent points, but could you explain this one in more detail. I’m wondering why less metal layers would create a larger chip.
  • dagamer34dagamer34 Posts: 494member
    Quote:
    Originally Posted by Ssampath View Post


    I think the question is what value add does Nvidia provide in the Arm food chain if an upstart (in the chip business) like Apple can outdo them with the A5. Also there is at least 4 other companies doing Arm chips - Marvel, Samsung, Ti and Qualcomm. It is not clear to me if this is going to be a great business for Nvidia.



    With a billion cell phones sold each year, getting 10% of that market would be 100 million devices. I think 100 million of anything is quite a number.
  • alandailalandail Posts: 603member
    Quote:
    Originally Posted by tipoo View Post


    Yep. Remember, to get the surface area of a transistor, figures like 45nm are squared, so a difference of 5nm is larger than it seems. Plus, the A5 uses the MP2 version of the 543 GPU, which obviously would take more room, seeing as its a whole 'nother graphics core. Not really surprising that its bigger, nor that it does better in raw graphics benchmarks.



    I know the 543 was built from the ground up to be scalable (up to 16, I think, and the PSP2/NGP will use the 4 core version), I wonder if Nvidia's equivalent GPU is like that? If so, they could catch up soon.



    not at the same price.
  • ssampathssampath Posts: 10member
    Quote:
    Originally Posted by dagamer34 View Post


    With a billion cell phones sold each year, getting 10% of that market would be 100 million devices. I think 100 million of anything is quite a number.



    But this is the story of many dwarfs competing for the same market just like all the Android phone manufacturers right. They all drive the prices down and no one makes any profits. Sure there are millions of Android phones being sold but Apple and RIM still make most of the money!
  • alandailalandail Posts: 603member
    Quote:
    Originally Posted by EgisCodr View Post


    I guess you guys don't realize that the Tegra 2 is last year's chip. You are comparing the A5 to a chip from 2010. The next chip will be out in products in June. Code-named Kal-El, it is the first quad-core processor on the market. Five times faster than the Tegra 2.



    That is THIS YEAR's chip. Cannot wait to see how the A5 stacks up.



    Apple would be better served buying the chips from Nvidia. By 2014, Nvidia will have chips 100 times faster than Tegra 2. This is an arena Apple will not be able to keep up in.



    Apple has proven they can turn around a new chip design in 12 months, add substantial performance, and have it shipping in high volumes. WHere is the evidence they can't compete when they are currently ahead?



    And even nVidea says the Kal-El won't be in systems in June.



    And 5x is graphics, not CPU. Is it even clear that a quoted 5x increase, which isn't going to be uniform, even pulls ahead of the A5?







  • realisticrealistic Posts: 1,111member
    Quote:
    Originally Posted by extremeskater View Post


    Apple didn't build the A5 chip, Samsung built the A5.



    Yes, we already know Samsung makes the A5 under contract for Apple. Apple doesn't manufacture anything. Apple designed the A5 and Samsung can only sell the A5 to Apple. So what is your point?
  • thepixeldocthepixeldoc Posts: 2,257member
    Quote:
    Originally Posted by Shrike View Post


    ...

    (clip)



    Tegra 3 is going to be built on the same 40 nm process as Tegra 2. The thing has about 50% more transistors then Tegra 2. It'll need to be operating at ~25% lower voltages and be power-gated up the wazoo to fit inside the same power envelope (~0.5 Watts) as the Tegra 2 in order to fit inside handhelds. People should be skeptical that they could do this, and quad-core SoCs probably won't fit until the 28nm to 32nm nodes. And that means 2012.



    But seriously dude... any Fandroid worth his propeller, will be able to get past the "gates" by tweaking the BIOS, water-cooling, and over-clocking the beast... then it will be truly "awesome", or?
  • solipsismsolipsism Posts: 25,726member
    LOL I’d love to see a water-cooled smartphone.
  • ksecksec Posts: 1,177member
    Quote:
    Originally Posted by Shrike View Post


    There are lots of good reasons for the A5 to be larger than Tegra 2:



    1. PowerVR SGX 543MP2 has 4x the performance as Tegra 2 ULP GeForce. TANSAAFL. That probably means 4x the transistors.



    2. Apple's A5 appears to implement a pipelined FPU while the Tegra 2 doesn't. That's more transistors.



    3. Apple has likely contracted Samsung to fab the A5 with less metal layers than Nvidia has asked TSMC. This can result in larger die sizes for the same transistor count.



    You're absolutely right though. Cost-wise we have no idea. There are good reasons for the A5 to be cheaper.



    1. Samsungs 45 nm fab is mature and has much better yield than TSMC's 40 nm.



    2. Larger die size may imply less metal layers which imply better yield than what Nvidia has asked TSMC to do with 40 nm.



    3. Older fab may mean cheaper product!







    Actually, we have no idea how much an A5 costs Apple. Absolutely none. Apple themselves may not even be able to separate an actual real live cost out of it. Apple basically has billions of dollars in contracts with Samsung for CPU-GPU chips, flash chips, RAM chips, LCDs, and who knows what else.



    They may have a specific contract for the A5 SoC (which includes the MCM packaging of the RAM chips) so they could have the precise number, but then again, Apple may have amortized the cost of Samsung components over multiple components! Who knows.



    Well, die size i was referring to Apple doesn't care about die size as much as Nvidia due to cost issues.



    TSMC 40nm LP is actually much better then what other people think. But Samsung Holds all the liscense needed to create Apple's SoC.



    Th cost of A5 comes from some other analyst and isuppli. Which predict from somewhere 20 - 30 Range. So i took the middle of the estimate as $25.
Sign In or Register to comment.