Qualcomm Roadmap (2011-2012)

[Arun]

http://www.timlab.it/Download/SESSION_2-TIM_LTE_Workshop.pdf
If that link is too slow (or no longer works) you can try using the Google cache: http://www.google.com/search?q=http://www.timlab.it/Download/SESSION_2-TIM_LTE_Workshop.pdf

Major revelations include:
- MSM8960's Krait core is running at 1.5-1.7GHz and it's only the APQ8064/MSM8974 that will reach 2.0-2.5GHz.
- 800MHz Cortex-A5 MSM7227A (& absurdly named 600MHz MSM7225A SKU) on 45nm for ultra-low-cost smartphones.
- MDM9615 (28nm shrink of the 100Mbps LTE MDM9600) is slated for Q2 2012 production which is earlier than expected.
- Unannounced MSM8974 is a combination of the APQ8064 application processor and the 150Mbps LTE MDM9625 baseband.
- MSM8230 will only support 50Mbps LTE/21Mbps HSPA+ rather than 100Mbps LTE/42Mbps HSPA+ like the MSM8960.
- Unconfirmed MSM8228 ("in planning"): 14.4Mbps HSPA+/2x1.0-1.2GHz Krait/512KB L2/720p video/Adreno 305.
- All future Qualcomm devices with LTE support TD-SCDMA for China Mobile.

Analysis:
- I'm willing to bet the iPhone 6 will use the MDM9615. Previously it was unclear if it would be ready on time, but if it's currently on schedule to be in mass production by Q2 2012 and Apple is the lead customer, it could be ready for an iPhone 6 in late September 2012. It's also the only solution that supports not only GSM/HSPA/LTE but also CDMA (for Verizon) and TD-SCDMA (for China Mobile).
- The 45nm MSM7227A is a very attractive solution for the ultra-low-cost market but it'll face tough competition from 40nm 1xCortex-A9 solutions from Broadcom and ST-Ericsson (which should be nearly as cheap since A9 isn't much bigger than A5 compared to a 3G baseband and LPDDR2 hopefully won't be noticeably more expensive than LPDDR1 in that timeframe). The package isn't clear from that presentation (11x11 is insane as you need 12x12 for PoP memory) but if it's fully compatible with the MSM7227 then that's a very big selling point. It also helps that Qualcomm has also indicated they are willing to sacrifice their gross margins for ultra-low-cost smartphones.
- The MSM8230 and MSM8228 are both very attractive chips for the mid-range and low-end smartphone markets in their target timeframe and it's nice to see that Qualcomm can apparently scale their LTE architecture all the way from 50Mbps to 150Mbps without paying more die size than necessary. It's not clear if the MSM8974 will be quite as impressive versus competitors but at least this presentation confirms it supports 1080p 60fps (->3D 1080p 30fps) and rather interestingly LPDDR3 (basically doubles data per raw MHz, up to 1600MHz effective versus 1066MHz effective for LPDDR2).

 

[Lazy8s]

I guess using their architectural license for an improved Cortex-A5 class core wasn't as practical as just licensing the A5 itself.

 

[Deleted member 13524]

Finally, an ARMv7 update to MSM7227. Low-end smartphones will finally be able to read websites with flash content, yay (at the expense of dirt-poor performance but whatever).

By the way, is there any way to know if the MSM7227A will have a FPU unit, like its predecessor?

They're also sticking to the old-as-hell Adreno 200, apparently with some ~35% higher clocks. I wonder if the Adreno 205 was that much larger..

 

[Erinyes]

I think Qualcomm have pretty much fixed any holes in their roadmap eh?

http://www.timlab.it/Download/SESSION_2-TIM_LTE_Workshop.pdfMajor revelations include:
- MSM8960's Krait core is running at 1.5-1.7GHz and it's only the APQ8064/MSM8974 that will reach 2.0-2.5GHz.
- 800MHz Cortex-A5 MSM7227A (& absurdly named 600MHz MSM7225A SKU) on 45nm for ultra-low-cost smartphones.
- MDM9615 (28nm shrink of the 100Mbps LTE MDM9600) is slated for Q2 2012 production which is earlier than expected.
- Unannounced MSM8974 is a combination of the APQ8064 application processor and the 150Mbps LTE MDM9625 baseband.
- MSM8230 will only support 50Mbps LTE/21Mbps HSPA+ rather than 100Mbps LTE/42Mbps HSPA+ like the MSM8960.
- Unconfirmed MSM8228 ("in planning"): 14.4Mbps HSPA+/2x1.0-1.2GHz Krait/512KB L2/720p video/Adreno 305.
- All future Qualcomm devices with LTE support TD-SCDMA for China Mobile.

-Yea i remember reading something to that effect by Metafor in one of the threads. Its lower than expected but IPC is pretty much double that of Snapdragon right? APQ8064/MSM8974 are probably only going to be used in tablets so the higher clock makes sense there.
-How does the Cortex A5 stack up against the current CPU in the MSM7227/7225? (im not sure what they have)
-Hopefully that should reduce power consumption significantly and we'll see LTE phones with more than half a day of usable battery life
-You mean the baseband is integrated or is it a two chip combination?
-I think you mean MSM8930 right? And it makes sense i guess. Its a single core so likely destined for the mid-range market (im not even sure how many of those phones will even need LTE)
-Now MSM8228 sounds very nice for a mid-range smartphone. Krait will bring a big boost in CPU power over current dual cores, and should be more power efficient as well. Not everyone needs LTE and this should also result in cost and power savings.

Analysis:
- I'm willing to bet the iPhone 6 will use the MDM9615. Previously it was unclear if it would be ready on time, but if it's currently on schedule to be in mass production by Q2 2012 and Apple is the lead customer, it could be ready for an iPhone 6 in late September 2012. It's also the only solution that supports not only GSM/HSPA/LTE but also CDMA (for Verizon) and TD-SCDMA (for China Mobile).
- The 45nm MSM7227A is a very attractive solution for the ultra-low-cost market but it'll face tough competition from 40nm 1xCortex-A9 solutions from Broadcom and ST-Ericsson (which should be nearly as cheap since A9 isn't much bigger than A5 compared to a 3G baseband and LPDDR2 hopefully won't be noticeably more expensive than LPDDR1 in that timeframe). The package isn't clear from that presentation (11x11 is insane as you need 12x12 for PoP memory) but if it's fully compatible with the MSM7227 then that's a very big selling point. It also helps that Qualcomm has also indicated they are willing to sacrifice their gross margins for ultra-low-cost smartphones.

-Yea that makes sense. That would allow them to have a single model for the entire world and would simplify manufacturing by having only one SKU.
-If it is compatible with MSM7227 i can see it doing very well. The number of devices which are shipping with MSM7227 right now are staggering. But even if it isnt it should drop costs even further for entry level smartphones. But the Broadcom and ST-E designs with their 800 mhz Cortex A9 class CPU's are still going to be a class ahead so Qualcomm will probably be relegated to the lower end of the low-cost market!

Finally, an ARMv7 update to MSM7227. Low-end smartphones will finally be able to read websites with flash content, yay (at the expense of dirt-poor performance but whatever).

By the way, is there any way to know if the MSM7227A will have a FPU unit, like its predecessor?

They're also sticking to the old-as-hell Adreno 200, apparently with some ~35% higher clocks. I wonder if the Adreno 205 was that much larger..

Yea the Adreno 200 will be downright ancient by the time these SoC's are available. Granted they are typically going to be used on devices with QVGA/WQVGA/HVGA resolutions, but a bit more graphics horsepower would have been nice. Maybe not Adreno 205 level but something in between would have been nice

 

[Wishmaster]

On the matter of pure specs of those SoC's I'm not impressed, of course I'm talking about high-end.
- No quad-core CPU until 2013, too late IMO(at least from marketing POV)
- Clock speed's seem low, although without informations on IPC of Krait it's hard to tell what kind of performance we can expect. Definitely faster than A9, but how will it fare against A15?
- GPU numbers and dual channel memory support should mean quite a boost in performance with 225. Will it be enough in 2012? Compared to 220(which is actually leading in many GLBenchmark subtests) this could give us something on the ipad2 level, although DX 9.3 is worrying considering windows 8 support, but this could still change before the release.
- Adreno 305, from the numbers it looks almost like 220. So with improved drivers it should be more than enough for higher-mid/lower-high end devices.
- Adreno 320, almost 2x faster than 225 with dual-channel LPDDR3. Sound impressive, but still not on 'Rouge' level.

More interesting things that I would like to know:
- Wi-Fi Display support coming with recently announced connectivity chips. Will it be used by those new SoC's?
- Will adreno 225, 305, 320 support OpenGL ES '3.0'?

 

[convergedw]

- No quad-core CPU until 2013, too late IMO(at least from marketing POV)
- Clock speed's seem low, although without informations on IPC of Krait it's hard to tell what kind of performance we can expect. Definitely faster than A9, but how will it fare against A15?

FWIW, some datapoints from another Qualcomm presentation on Krait.

Dual core next gen CPU micro-architecture outperforms a quad core A91
•23% more headroom than A15 expected performance
•47% lower power consumption than A15

 

[metafor]

FWIW, some datapoints from another Qualcomm presentation on Krait.

Dual core next gen CPU micro-architecture outperforms a quad core A91
•23% more headroom than A15 expected performance
•47% lower power consumption than A15

*very loud cough*

 

[Wishmaster]

FWIW, some datapoints from another Qualcomm presentation on Krait.

Dual core next gen CPU micro-architecture outperforms a quad core A91
•23% more headroom than A15 expected performance
•47% lower power consumption than A15

Yeah, I've already seen this presentation. Let's wait and see what Krait will really offer. Maybe they really managed to develop better CPU than A15, although that is rather unlikely.

 

[Arun]

Its lower than expected but IPC is pretty much double that of Snapdragon right?

No way - even A15 isn't double Snapdragon. I think it's clear that the A15 will have the performance crown this generation (higher IPC *and* higher clocks) but maybe Krait will be more power efficient.

APQ8064/MSM8974 are probably only going to be used in tablets so the higher clock makes sense there.

According to Qualcomm, APQ8064 is tablet-centric, but MSM8974 supposedly targets smartphones as well.

How does the Cortex A5 stack up against the current CPU in the MSM7227/7225? (im not sure what they have)

ARM11 is 1.25DMIPS/MHz, Cortex-A5 is 1.57DMIPS/MHz, Snapdragon is 2.1DMIPS/MHz, Cortex-A9 is 2.5DMIPS/MHz. Once again, DMIPS doesn't mean that much so don't take these numbers too seriously.

-You mean the baseband is integrated or is it a two chip combination?

Integrated, although just like the MSM8960 which was moved from 40nm to 28nm this looks like a likely contender to be moved to 20nm if it gets delayed too much.

I think you mean MSM8930 right? And it makes sense i guess. Its a single core so likely destined for the mid-range market (im not even sure how many of those phones will even need LTE)

MSM8930 and MSM8230 are two SKUs of the same chip according to that roadmap, and both are dual-core. Also LTE support is good as it costs less per bit for operators, plus Qualcomm doesn't really have a choice here - their 21Mbps baseband architecture in the MDM8200(A) is very inefficient, so LTE basically comes for free if they use they scale down their LTE architecture instead.

-Now MSM8228 sounds very nice for a mid-range smartphone. Krait will bring a big boost in CPU power over current dual cores, and should be more power efficient as well. Not everyone needs LTE and this should also result in cost and power savings.

Absolutely, I don't really understand how it can be slated for production at the same time as MSM8930 despite still being in planning though...

Yea that makes sense. That would allow them to have a single model for the entire world and would simplify manufacturing by having only one SKU.

Ha! You wish. The problem is power amplifiers: for a true world phone you'd need (at least) 5 bands for HSPA, 2 bands for CDMA, and one band for TD-SCDMA. Quite expensive but still doable. But then comes the real problem: LTE bands are not at all harmonised over the world. You'd need one band for AT&T 700MHz, one band for Verizon 700MHz, at least three more bands for the other spectrum they plan to reuse for LTE (depending on your PA arch you *might* be able to share those with the 3G bands). Then you need a bunch of bands for Europe, a bunch of FDD bands for Asia, one TD-SCDMA and one TD-LTE band for China, and so on...

I think going from 3 to 5 bands today to 15-20 bands isn't very realistic. I expect they'll have *at least* one model for the USA (5xHSPA/2xCDMA/USA-LTE), one model for Europe plus some of Asia (5xHSPA/Non-USA-LTE), and one model mostly for China (5xHSPA/TD-SCDMA/TD-LTE). The only way they'll get around this problem is with something like Nujira's Coolteq-l and a wide-band power amplifier (or rather two, one for the bottom bands and one of the top bands). It's possible but not very likely.

Maybe not Adreno 205 level but something in between would have been nice

I'm not sure what you think the difference between 200 and 205 is, but there's basically no possible in-between step. Adreno 200 is a single TMU with a single ALU pipeline, Adreno 205 is a single TMU with two ALU pipelines. The rest of the difference is just minor architectural tweaks and clock speed.

- No quad-core CPU until 2013, too late IMO(at least from marketing POV)

AFAIK APQ8064 isn't cancelled - it simply isn't on that roadmap because it isn't a MSM chip. Of course, maybe it really always was the same chip, just with the baseband disabled to speed up certification and be more attractive for tablet designs that want to use 3G/4G modules anyway...

- Adreno 320, almost 2x faster than 225 with dual-channel LPDDR3. Sound impressive, but still not on 'Rouge' level.

Actually I think performance will probably be better than the raw triangle specs would imply. But yes, it's still not on Rogue level.

- Wi-Fi Display support coming with recently announced connectivity chips. Will it be used by those new SoC's?

Qualcomm said Wi-Fi Display will be supported by all 28nm SoCs when paired with the WCN3660, so presumably all of them except for the MSM7227A. The WCN3660 is an intriguing little chip BTW - older presentations clearly say it's still on 65nm (and it'd be too early for 40nm anyway - even CSR and Broadcom are only just starting to sample their first chips this quarter) but it's only 15mm² which is about 40% smaller than the best competitor. My guess is they're actually integrating the MAC on the 28nm baseband itself and only keeping the PHY and RF on the WCN3660 (just like they've been putting the GPS logic entirely on the baseband and kept the RF separate on the 3G RF chip). If I'm right, that's a very good integration strategy.

- Will adreno 225, 305, 320 support OpenGL ES '3.0'?

AFAIK Adreno 305/320 will, 225 will not. More interestingly, this presentation implies 3xx won't support more than DX9.3, whereas both Rogue and Mali-T604 support up to DX11.

 

[Lazy8s]

Not to start rumors about that last bit, but that might not then be their longer-term solution.

 

[mczak]

Finally, an ARMv7 update to MSM7227. Low-end smartphones will finally be able to read websites with flash content, yay (at the expense of dirt-poor performance but whatever).

By the way, is there any way to know if the MSM7227A will have a FPU unit, like its predecessor?

They're also sticking to the old-as-hell Adreno 200, apparently with some ~35% higher clocks. I wonder if the Adreno 205 was that much larger..

Yeah, it's an interesting chip. Basically looks like a straight shrink of MSM7227 (same graphics core, same lpddr1 memory, same video support, seemingly same baseband capabilities). Except the cpu core where the arm11 got replaced with the cortex-a5.
Somehow I can't think that they'd skip the fpu (though maybe skip neon?). The a5 isn't a massive upgrade in any case, but it should be slightly faster, using even less die area than the arm11, and the whole chip more power efficient. Overall not too bad, though not really an exciting upgrade over the 7227 (but I'd guess even cheaper).
btw is this the first Cortex-A5 to appear? In any case looks like the A9 didn't quite make it into the cheap category for smartphones appearing end of 2011 / beginning of 2012.
The MSM8228 though will be a massive upgrade for low-end smartphones.

 

[Wishmaster]

Actually I think performance will probably be better than the raw triangle specs would imply. But yes, it's still not on Rogue level.

If it's even better then we should get GPU faster than what we have now in iPad2, but slightly too late.

AFAIK Adreno 305/320 will, 225 will not. More interestingly, this presentation implies 3xx won't support more than DX9.3, whereas both Rogue and Mali-T604 support up to DX11.

That's something I can't understand. Slower GPU will get 'Halti' support, faster GPU's won't support DX11, and not even DX10.1. But there's hope for OpenCL, I remember last year presentation where it was implied that those new GPU's will support OpenCL. Hopefully there will be some changes to the GPU architecture. BTW I wonder whether it will be partially based on z460 or maybe completely redesigned and new GPU made by the people that previously worked on Imageon.
Do you think it's safe to assume that 320 will be nothing more than four 305 cores?

 

[darkblu]

Finally, an ARMv7 update to MSM7227. Low-end smartphones will finally be able to read websites with flash content, yay (at the expense of dirt-poor performance but whatever).

If a sole A8 is all you have to rely on for flash, you're dead in the water. At the same time, an A8 with some 'accelerator' support can be quite adequate for flash player 10.1 through .3. (where the accelerator requirements are a function of targeted resolutions, of course).

 

[wco81]

Is there an advantage in integrating the baseband with the application processor?

Or how common is the integrated approach? Probably not common since a big volume vendor like Apple is shipping baseband separately from their A4 chips?

What about HTC, Samsung, etc.?

 

[mczak]

Is there an advantage in integrating the baseband with the application processor?

Or how common is the integrated approach? Probably not common since a big volume vendor like Apple is shipping baseband separately from their A4 chips?

What about HTC, Samsung, etc.?

Well the MSM7227 is very popular (qualcomm really the only one offering integrated chips for now). Looks though overall today the total marketshare of integrated baseband/app chips isn't that high. If you believe market studies it's bound to grow however.
I think the advantage is it's just plain cheaper...

 

[wco81]

Wondering if the firmware can be upgraded or downgraded independently, as is the case now for jailbreaks.

Or would integrated designs make this tougher? Apple may be using Qualcom basebands but looks like they're going to roll their own APU.

 

[metafor]

Well the MSM7227 is very popular (qualcomm really the only one offering integrated chips for now). Looks though overall today the total marketshare of integrated baseband/app chips isn't that high. If you believe market studies it's bound to grow however.
I think the advantage is it's just plain cheaper...

I believe Broadcom's AP offerings also offer integrated baseband. It looks as if Intel and nVidia are moving in that direction as well. There are obvious advantages for mid-tier phones due to cost and board space.

I agree that it may not make much sense at the high-end and tablet space. However, that may change with LTE since the area and cost of a separate modem can be prohibitive.

 

[fehu]

If a sole A8 is all you have to rely on for flash, you're dead in the water. At the same time, an A8 with some 'accelerator' support can be quite adequate for flash player 10.1 through .3. (where the accelerator requirements are a function of targeted resolutions, of course).

here is nook A8@800MHz saying hallo

 

[darkblu]

here is nook A8@800MHz saying hallo

Unless you know for a fact the various 'media accelerators' in the 3621 (i.e. sgx, c64x, overlay engine, etc) twiddle their thumbs while the device does flash, I'm willing to wager the A8 is not the sole worker under that scenario.

 

[Arun]

Well the MSM7227 is very popular (qualcomm really the only one offering integrated chips for now). Looks though overall today the total marketshare of integrated baseband/app chips isn't that high. If you believe market studies it's bound to grow however.
I think the advantage is it's just plain cheaper...

I believe Broadcom's AP offerings also offer integrated baseband. It looks as if Intel and nVidia are moving in that direction as well. There are obvious advantages for mid-tier phones due to cost and board space.

Actually integrated basebands are currently VERY common because:
1) Qualcomm is the market leader.
2) All Nokia S60 devices with 10.2Mbps HSPA (i.e. a truckload of them) use the same custom 45nm integrated baseband from TI. I believe in the past some of their solutions were SiPs but this one is definitely a SoC as indicated by Slide 11 of this presentation: http://www.ti.com/corp/docs/investor/analyst2008/pdfs/Ritchie.pdf
3) Broadcom is also shipping a 3G baseband in (large) volume to Samsung with a dedicated 312MHz ARM11 application processor next to the 208MHz ARM9 modem: http://www.eetrend.com/files-eetrend/BCM2153 datasheet.pdf - the BCM2157 is a similar chip targeted at Android that boosts the ARM11 clock speed to 500MHz+ among other things. They are also sampling Cortex-A9 chips both standalone and integrated (with a focus on the latter going forward).
4) Marvell ships a variety of integrated solutions including the PXA910 and PXA920: http://www.marvell.com/cn/platforms/assets/Marvell-Pantheon-Platform-Brief.pdf

And this is only counting chips with a dedicated ARM core for application processing (thus excluding the vast majority of feature phones which reuse the modem ARM core to run the OS). Other companies that will soon be shipping integrated solutions with a dedicated application processor include:
5) ST-Ericsson with the U8500. Ironically, some customers don't want the Nokia-designed 14.4Mbps baseband and would rather use the EMP-designed 21.6Mbps M5730 baseband next to it. In theory, that solution is called the U9500. While thz A9540 and A9600 are discrete, the U4500 is integrated and should be very high volume. They've also got integrated derivatives on their 28nm roadmap.
6) Mediatek with the MT6573 (7.2Mbps HSUPA baseband with a dedicated 650MHz ARM11).

Actually Intel and NVIDIA aren't being too aggressive with baseband integration. The main reasoning behind their respective acquisitions was bundling and being able to provide direct support for the whole platform. I'd expect Intel to integrate a LTE baseband into their second-generation 22nm SoC and NVIDIA to include one in a lower-end derivative on 20nm... but not before.

I agree that it may not make much sense at the high-end and tablet space. However, that may change with LTE since the area and cost of a separate modem can be prohibitive.

I completely fail to see how that is the case. If you only need 50Mbps LTE (e.g. AT&T) and don't need the 2600MHz band, Icera's ICE8060 does it in a 7x7 package with stacked DRAM and a die size of only 15.6mm². Their 28nm solution (which was recently still expected to be back from the fab in Q4) will support 150Mbps LTE with an even slightly smaller die size. What's prohibitive about that?

The only thing ridiculous about LTE is the number of power amplifiers required if you want to support multiple countries or even sometimes just multiple operators in the same country. If anything, that makes modules an even more attractive solution for tablet manufacturers.