Next-Gen iPhone & iPhone Nano Speculation

[Ailuros]

Hondo uses rather simple x86 cores at 1GHz, a 64-bit memory system, and run the 80 "radeon cores" at undisclosed (presumably quite low) clock speeds in order to achieve its 4,5W TDP. Essentially, it is a very cut down PC part.
There is every reason to believe that a SoC like the A6x will walk all over it in terms of performance.

Frequencies between the Hondo and A6X GPU block might even be comparable. Assuming that GPU frequencies aren't too far away you have 80 DX11 SPs on one side and 128 DX9L3 SPs on the other, and even that is most likely an oversimplification, but not a factor I'd ignore. It'll get probably even more interesting if someone would compare TMU count amongst others.

Then again, Hondo doesn't compete directly against the ARM SoCs, it exists as a low cost alternative to Intels offering in the x86/Windows space. In that respect it's a godsend for the hardware manufacturers in that segment, since otherwise they would be completely at the mercy of Intel. Remains to be seen how many are interested in running x86 legacy code on tablets - to me it seems like the netbook crippleware experience all over again - but Hondo is the lowest cost ticket into that market you can find.

True.

 

[ams]

Hondo uses rather simple x86 cores at 1GHz, a 64-bit memory system, and run the 80 "radeon cores" at undisclosed (presumably quite low) clock speeds in order to achieve its 4,5W TDP. Essentially, it is a very cut down PC part.
There is every reason to believe that a SoC like the A6x will walk all over it in terms of performance.

The difference in framerate between A6X and Hondo in the GLBenchmark 2.5 HD onscreen test is "only" 16%, so the difference in GPU performance seems to be pretty minor at best (at least based on this particular bench test). The disadvantages for Hondo are twofold: 1) Use of a 40nm fabrication process (vs. a 32nm fabrication process for A6X), and 2) Use of a separate Fusion Hub Controller chip which adds somewhere between 0.5-0.75w to the power consumption of the device. Note that the Radeon GPU used in Hondo reportedly has an operating frequency of ~ 276MHz, which is very similar to what is reported for the SGX 554MP4 GPU used in A6X. Even though the power consumption of Hondo is not ideal for a thin tablet form factor, do note that the power consumption of the ipad 3 (and presumably the ipad 4 too) is not stellar either. According to Intel internal testing, with HD video playback, the ipad 3 has up to 90% higher power consumption than the Asus Transformer Pad Infinity (which happens to be the leakiest and least power efficient Tegra 3 variant), and the ipad 3 has up to 136% higher power consumption than the ipad 2 (of which the more power hungry higher resolution screen used in the ipad 3 accounts for only 72% out of this 136% total difference in power consumption). So if Intel and AMD's power consumption data is reliable and comparable to each other, then Hondo only has slightly higher power consumption compared to the ipad 3 with respect to web browsing and HD video playback (normalized for differences in idle power consumption between the 1080p screen used in Hondo and the more power hungry higher resolution screen used in the ipad 3), with the main difference in power consumption explained by the separate FHC chip used in Hondo.

 

[Ailuros]

The difference in framerate between A6X and Hondo in the GLBenchmark 2.5 HD onscreen test is "only" 16%, so the difference in GPU performance seems to be pretty minor at best (at least based on this particular bench test).

Why is the Hondo running at 2048*1536? If onscreen for the Hondo and the 36 fps was for 1080p the difference is rather at 44% or relatively close to the difference in SP amount.

The disadvantages for Hondo are twofold: 1) Use of a 40nm fabrication process (vs. a 32nm fabrication process for A6X), and 2) Use of a separate Fusion Hub Controller chip which adds somewhere between 0.5-0.75w to the power consumption of the device. Note that the Radeon GPU used in Hondo reportedly has an operating frequency of ~ 276MHz, which is very similar to what is reported for the SGX 554MP4 GPU used in A6X. Even though the power consumption of Hondo is not ideal for a thin tablet form factor, do note that the power consumption of the ipad 3 (and presumably the ipad 4 too) is not stellar either. According to Intel internal testing, with HD video playback, the ipad 3 has up to 90% higher power consumption than the Asus Transformer Pad Infinity (which happens to be the leakiest and least power efficient Tegra 3 variant), and the ipad 3 has up to 136% higher power consumption than the ipad 2 (of which the more power hungry higher resolution screen used in the ipad 3 accounts for only 72% out of this 136% total difference in power consumption). So if Intel and AMD's power consumption data is reliable and comparable to each other, then Hondo only has slightly higher power consumption compared to the ipad 3 with respect to web browsing and HD video playback (normalized for differences in idle power consumption between the 1080p screen used in Hondo and the more power hungry higher resolution screen used in the ipad 3), with the main difference in power consumption explained by the separate FHC chip used in Hondo.

For which video playback power consumption has little or nothing to do with the GPU itself.

 

[Entropy]

The difference in framerate between A6X and Hondo in the GLBenchmark 2.5 HD onscreen test is "only" 16%, so the difference in GPU performance seems to be pretty minor at best (at least based on this particular bench test). The disadvantages for Hondo are twofold: 1) Use of a 40nm fabrication process (vs. a 32nm fabrication process for A6X), and 2) Use of a separate Fusion Hub Controller chip which adds somewhere between 0.5-0.75w to the power consumption of the device. Note that the Radeon GPU used in Hondo reportedly has an operating frequency of ~ 276MHz, which is very similar to what is reported for the SGX 554MP4 GPU used in A6X. Even though the power consumption of Hondo is not ideal for a thin tablet form factor, do note that the power consumption of the ipad 3 (and presumably the ipad 4 too) is not stellar either. According to Intel internal testing, with HD video playback, the ipad 3 has up to 90% higher power consumption than the Asus Transformer Pad Infinity (which happens to be the leakiest and least power efficient Tegra 3 variant), and the ipad 3 has up to 136% higher power consumption than the ipad 2 (of which the more power hungry higher resolution screen used in the ipad 3 accounts for only 72% out of this 136% total difference in power consumption). So if Intel and AMD's power consumption data is reliable and comparable to each other, then Hondo only has slightly higher power consumption compared to the ipad 3 with respect to web browsing and HD video playback (normalized for differences in idle power consumption between the 1080p screen used in Hondo and the more power hungry higher resolution screen used in the ipad 3), with the main difference in power consumption explained by the separate FHC chip used in Hondo.

Ailuros already addressed the graphics benchmark.

The power draw of the very cut down Hudson M2T FCH isn't included in the 4,5W TDP, on the other hand it is less than a Watt, so it is not as disastrous as the previous Hudson M1 that consumed 4W+. That AMD chooses to use 40nm processing is a business decision that helps them achieve a low price point, and helps them bring the device to market at all.

Also, these numbers do not include the RAM or anything else necessary for operation, which is worth remembering when you compare to real world devices. The iPad3 is a last generation device, no longer sold. I'm using an iPad4 with an A6x processor right now. When I can go out and buy a Hondo equipped tablet, then one might be able to make a relevant comparison. For all the use that would be, Hondo is only going to be found in x86-Windows tablets, and the A6x only in the iPad4, so there is only an academic interest in such an exercise, they don't compete for the same customers.

That said, it would be interesting to know what kind of memory bandwidth utilization advantage ImgTech tilers enjoy under real-world circumstances, but I'm not aware of any benchmark that allows good comparisons. GLBench is pretty much as good as it gets. But I assume that Apple didn't go to the additional expense of an 128-bit bus for no good reason. And I would (less confidently) assume that AMDs GPU is not going to more frugal with bandwidth than the ImgTech tilers. If both of these are true, the A6x would have a very substantial bandwidth advantage. Hondo has 8 TMUs and 4 ROPs, and unless I misremember, the 554MP4 (and 543MP4) would be 8 TMUs/ROPs.Together, this points to a significant real world fill-rate advantage for the A6x.

 

[zed]

Um OK the ipad4 is twice the speed of an ipad3
but if u take most games/test (unless differntly written) they will not perform better on the ipad4.
u target a FPS rate on all hardware eg me Im hoping my game runs at 30fps on iphone 3GS as well as ipad 4, i.e. I dont want it to run at 5000fps and overheat the battery (and burn it out in an hour)

 

[Pressure]

Um OK the ipad4 is twice the speed of an ipad3
but if u take most games/test (unless differntly written) they will not perform better on the ipad4.
u target a FPS rate on all hardware eg me Im hoping my game runs at 30fps on iphone 3GS as well as ipad 4, i.e. I dont want it to run at 5000fps and overheat the battery (and burn it out in an hour)

Well, iOS vsyncs to 60 FPS anyway, so it will never run faster than that.

 

[ltcommander.data]

http://www.anandtech.com/show/6429/ipad-mini-review

Anand's iPad Mini review is finally up. Performance is unsurprisingly the same as the iPad 2. Battery life is iPad 3 class or worse than the 45nm iPad 2 so process improvements in the SoC and other components couldn't make up for the smaller battery compared to the iPad 2. Anand isn't optimistic for a Retina iPad Mini for 2013. An "analyst" was pushing the idea today that a Retina iPad Mini would come in March 2013, which didn't make sense anyways given the extremely short 5 months life-cycle even worse than the iPad 3.

Also included were iPad 4 benchmarks. The A6X vs A6 CPU results are as expected for a 1.4GHz vs 1.3Ghz comparison. Anand revealed the GPU results previously. Notably, battery life is improved over the iPad 3 so process improvements helped here. The display also seems better than the iPad 4 although that might be individual device variation rather than silent technical improvements.

 

[ltcommander.data]

http://www.digitimes.com/news/a20121128PD204.html

Another Digitimes report that Apple is likely to switch to TSMC in 2013. Apple will need 200 million SoC a year requiring at least 200,000 12-inch wafers. TSMC's other customers like nVidia are reportedly worried that Apple will eat into their production capacity.

 

[wco81]

http://www.digitimes.com/news/a20121128PD204.html

Another Digitimes report that Apple is likely to switch to TSMC in 2013. Apple will need 200 million SoC a year requiring at least 200,000 12-inch wafers. TSMC's other customers like nVidia are reportedly worried that Apple will eat into their production capacity.

Yeah another story along the same vein:

http://arstechnica.com/apple/2012/1...o-move-a-series-mobile-chips-to-28nm-process/

So they have a new fab coming online to support 20nm and Apple is trying to get them to dedicate it to Apple SOCs. Can they really turn down that kind of volume?

Or is it true as the TMSC guy quoted says that if the fab is used for one type of product, it may be unusable for anything else once the product life of that one product is complete?


BTW, the story says producing SOCs on more efficient process is the only way to slim down the full iPad design? Wouldn't IGZO do more to reduce power consumption and the need for battery capacity?

Only thing about IGZO is whether Sharp is going to survive to produce it?

 

[ltcommander.data]

Or is it true as the TMSC guy quoted says that if the fab is used for one type of product, it may be unusable for anything else once the product life of that one product is complete?

Well processes used to make memory aren't necessarily the same as that used to make SoC. But in this case, they're probably referring to over-capacity of an obsolete process once Apple moves away from it rather than making a custom process for Apple different from the process used by competing SoC.

BTW, the story says producing SOCs on more efficient process is the only way to slim down the full iPad design? Wouldn't IGZO do more to reduce power consumption and the need for battery capacity?

Well the screen should be the biggest power consumer, but that's not to say the SoC isn't getting very power hungry.

http://www.anandtech.com/show/6330/the-iphone-5-review/12

Anandtech found that peak SoC power consumption has been steadily rising in successive generations from the A4 to A6. Average battery life may be kept in check by lower idle power consumption and finishing tasks faster, but there are still concerns about thermal dissipation under load especially as devices get thinner. Continuous tasks like gaming that don't allow idling are also showing decreased battery life in successive generations. Anand found this in the iPhone 5 over the iPhone 4S and previously the iPhone 4S was worse than the iPhone 4. Cutting down SoC power consumption under load would certainly be helpful to reduce the need for a large battery and chassis while maintaining similar battery life under active use.

 

[dagamer]

Yeah another story along the same vein:

http://arstechnica.com/apple/2012/1...o-move-a-series-mobile-chips-to-28nm-process/

So they have a new fab coming online to support 20nm and Apple is trying to get them to dedicate it to Apple SOCs. Can they really turn down that kind of volume?

Or is it true as the TMSC guy quoted says that if the fab is used for one type of product, it may be unusable for anything else once the product life of that one product is complete?


BTW, the story says producing SOCs on more efficient process is the only way to slim down the full iPad design? Wouldn't IGZO do more to reduce power consumption and the need for battery capacity?

Only thing about IGZO is whether Sharp is going to survive to produce it?

Apple may bring TSMC a lot of business but they do not want to become Apple's bitch. It is not worth pissing off all of their other customers just to have either beefy contracts to Apple or Qualcomm, which is why they turned down offers for VIP status. Because what happens when those customers leave and you need to mend relationships with all your other previous customers you shat on?

 

[tangey]

any significance in the fact that the latest Khronos opengles2.0 conformance listings for apple:-
http://www.khronos.org/conformance/adopters/conformant-products/#opengles

lists the new ipad cpu as "ARMv7s", but the iphone5 as "ARMv7" (along with the rest of the iphone/ipad/itouches), or could they just not be arsed doing a separate listing for iphone5+SGX543

 

[ltcommander.data]

http://www.asymco.com/2012/12/06/does-s-stand-for-spring/

There's interesting speculation here based on a claim from John Sculley that Apple might be moving to a 6 month product cycle for all their products. This could explain why everything (all iDevices and most Macs) were uncharacteristically bunched up in a fall release this year. Fall will be for major redesigns and spring for refreshes. Presumably that would mean a 6 month release cycle for incrementally improved SoC which seems aggressive considering Apple hasn't been in the SoC business that long. Coincidentally, it's now reported that Apple will be ramping at TSMC in spring 2013 instead of the fall with a SoC on the 28nm process. That might mean 28nm TSMC A6/A6X for Q2 iPhone/iPad refreshes.

http://www.taipeitimes.com/News/biz/archives/2012/12/10/2003549765

 

[ltcommander.data]

http://www.oregonlive.com/silicon-forest/index.ssf/2012/12/oregon_courts_mysterious_proje.html

Some chatter that officials in both Oregon and New York are courting a codename "Azalea" which is thought to be TSMC to build a 3.2-million-square-foot fab in their state. The thinking is that this could be a dedicated TSMC fab for Apple SoC production.

 

[ltcommander.data]

http://www.anandtech.com/show/6709/slightly-smaller-apple-tv-32-a1469-contains-a5x-soc-bcm4334-combo

So the AppleTV refresh that Apple prematurely disclosed by posting the IPSW uses an A5X, probably a shrink given the S5L8947 model number up from the S5L8945 of the Samsung 45nm A5X. Assuming Apple is serious about switching to TSMC this year, I think this is likely the 28nm (HPM?) TSMC test chip. The A6/A6X design may have been finalized too late to be the basis of the initial shrink test, while the A5X is still Apple's biggest chip to date so serves as an aggressive trial.

The A5X in the AppleTV raises questions again on the possibility of an Apple gaming console given it's overpowered to just serve media content. But you'd think they'd go with an A6X or even just the A6 if they were building a console. Unless they only put the A5X in the AppleTV because they needed a low volume way to ship the test chip and there is no larger plan to make use of the expanded GPU. I suppose they could be adding 4k support.

 

[wco81]

Supposedly the iPhone 5 was supply constrained so I don't think they'd divert the A6 from a lucrative product to a minimally profitable one.

If they were going to enable games and apps, you'd think they'd have to spend more to have local storage and some kind of controller.

 

[AlphaWolf]

Supposedly the iPhone 5 was supply constrained so I don't think they'd divert the A6 from a lucrative product to a minimally profitable one.

If they were going to enable games and apps, you'd think they'd have to spend more to have local storage and some kind of controller.

Where is it supply constrained? They sold way way less than expected. Are yields awful or something?

 

[wco81]

They claimed in the earnings conference that they sold every unit they made and could have sold more.

I guess the market didn't buy this claim.

In any event, I don't see them putting one of their newest SOCs in a $99 product, especially since it would be overkill.

 

[silent_guy]

Where is it supply constrained? They sold way way less than expected. Are yields awful or something?

They shipped 48 million iPhones. 'Analysts' expected 50 million. IOW they were off by 4%. Given that the history error margin of analyst predictions is way larger than 4%, way way less than expected is maybe a bit of a stretch?

 

[Arwin]

I bet many were just looking for an excuse to take their profits on that insanely high stock price.