Low-cost emerging market SoC/phone discussion

[Arun]

I believe 8x30 was intended to be the follow-on to 7227.

But it's still higher cost than 7227. Heck, it even supports 64-bit LPDDR1! It's true that the MSM7x30 has basically taken the price segment reserved by the MSM7x20A in the early days of the Android market though, but that's only because other components have become slightly cheaper AFAICT.

You'll likely see a single-core Krait rather than Scorpion at 28

Ah yes, that makes sense. Reminds me I still want to figure out how much performance/clock has improved over Scorpion. The only single-core Krait announced so far is the MSM8930 but that supports LTE so it's hardly the same market segment. Hopefully there will be some more clarity on the roadmap this year rather than having to wait until MWC12. If nothing is sampling this year, this creates a real opportunity for Broadcom and ST-Ericsson IMO.

 

[metafor]

But it's still higher cost than 7227. Heck, it even supports 64-bit LPDDR1! It's true that the MSM7x30 has basically taken the price segment reserved by the MSM7x20A in the early days of the Android market though, but that's only because other components have become slightly cheaper AFAICT.

True, but there just isn't anything targeted at the cost level of 7227. There was a 7227A but I don't think that ever made it anywhere.

Ah yes, that makes sense. Reminds me I still want to figure out how much performance/clock has improved over Scorpion.

Probably should wait until a chip is out

The only single-core Krait announced so far is the MSM8930 but that supports LTE so it's hardly the same market segment. Hopefully there will be some more clarity on the roadmap this year rather than having to wait until MWC12. If nothing is sampling this year, this creates a real opportunity for Broadcom and ST-Ericsson IMO.

What do BCM and ST have in that cost level that are supposed to rival 7x30? I'm sure they'll have plenty sampling this year, I've just not been following.

 

[Arun]

What do BCM and ST have in that cost level that are supposed to rival 7x30? I'm sure they'll have plenty sampling this year, I've just not been following.

Broadcom: http://www.broadcom.com/products/Cellular/3G-Baseband-Processors/BCM21654
ST-Ericsson: http://www.stericsson.com/products/u4500-novathor.jsp

Both are single-core Cortex-A9s on 40LP with a 7.2Mbps baseband, VGA/D1 video encode/decode, a basic OpenGL ES 2.0 GPU, and 32-bit LPDDR2. Needless to say, that's quite a bit cheaper than the MSM7x30 with a 14.4Mbps baseband, 720p video, a slightly faster GPU, and 64-bit LPDDR1/2.

There is obviously a place for the MSM7x30, but my point is I don't see what Qualcomm has to compete with the BCM21654 and the U4500 as both of them are sampling this quarter (Q2 2011). So it seems to me they've got a big opportunity to win the price segment that's currently using the MSM7227, and that's a very fast growing part of the market as well for smartphones. The original QSC7230 was supposed to be much closer to this (except it still had an ARM11 - ugh!) but it was cancelled in favour of what we have now. That's probably a good thing in itself, but it also leaves a gap in Qualcomm's roadmap.

 

[metafor]

Broadcom: http://www.broadcom.com/products/Cellular/3G-Baseband-Processors/BCM21654
ST-Ericsson: http://www.stericsson.com/products/u4500-novathor.jsp

Both are single-core Cortex-A9s on 40LP with a 7.2Mbps baseband, VGA/D1 video encode/decode, a basic OpenGL ES 2.0 GPU, and 32-bit LPDDR2. Needless to say, that's quite a bit cheaper than the MSM7x30 with a 14.4Mbps baseband, 720p video, a slightly faster GPU, and 64-bit LPDDR1/2.

There is obviously a place for the MSM7x30, but my point is I don't see what Qualcomm has to compete with the BCM21654 and the U4500 as both of them are sampling this quarter (Q2 2011). So it seems to me they've got a big opportunity to win the price segment that's currently using the MSM7227, and that's a very fast growing part of the market as well for smartphones. The original QSC7230 was supposed to be much closer to this (except it still had an ARM11 - ugh!) but it was cancelled in favour of what we have now. That's probably a good thing in itself, but it also leaves a gap in Qualcomm's roadmap.

That is a pretty gaping hole. IIRC, there really isn't anything that goes for that level of cost except for 8x30. There may be CDMA/UTMS only variants of that I'm not aware of but who knows.

 

[Arun]

That is a pretty gaping hole. IIRC, there really isn't anything that goes for that level of cost except for 8x30. There may be CDMA/UTMS only variants of that I'm not aware of but who knows.

There's the QSC6x95, but that's a feature phone product with a single 480MHz ARM11 (shared with the 14.4 baseband). So yes, even with as many chips as Qualcomm, it's apparently still possible to have a gap in your roadmap heh...

 

[metafor]

There's the QSC6x95, but that's a feature phone product with a single 480MHz ARM11 (shared with the 14.4 baseband). So yes, even with as many chips as Qualcomm, it's apparently still possible to have a gap in your roadmap heh...

There really aren't that many MSM chips. So far, there's 7x30 (same as 8x55), 8x50, 8x60 and the aging 7227. For now, the future's focused on high-end devices more than low-tier. I agree that the low-end is a giant market that needs to be addressed but oh well.

 

[convergedw]

I never noticed it before but Qualcomm has some unlabeled chips (MSM7xxx and MSM8xxx) on their roadmap in the mass market smartphone segment.

Page 10.

http://www.teamlightbulb.com/Telecom/Qualcomm Presentation.pdf

The position of the chips indicate that they should be successors to the MSM7227/7230 and were scheduled to sample in the first half of 2011. I have heard nothing about these chips so perhaps there was a delay?

 

[Erinyes]

Broadcom: http://www.broadcom.com/products/Cellular/3G-Baseband-Processors/BCM21654
ST-Ericsson: http://www.stericsson.com/products/u4500-novathor.jsp

Both are single-core Cortex-A9s on 40LP with a 7.2Mbps baseband, VGA/D1 video encode/decode, a basic OpenGL ES 2.0 GPU, and 32-bit LPDDR2. Needless to say, that's quite a bit cheaper than the MSM7x30 with a 14.4Mbps baseband, 720p video, a slightly faster GPU, and 64-bit LPDDR1/2.

There is obviously a place for the MSM7x30, but my point is I don't see what Qualcomm has to compete with the BCM21654 and the U4500 as both of them are sampling this quarter (Q2 2011). So it seems to me they've got a big opportunity to win the price segment that's currently using the MSM7227, and that's a very fast growing part of the market as well for smartphones. The original QSC7230 was supposed to be much closer to this (except it still had an ARM11 - ugh!) but it was cancelled in favour of what we have now. That's probably a good thing in itself, but it also leaves a gap in Qualcomm's roadmap.

Wouldnt it have been a better design choice to have incorporated 720p encode/decode instead of just VGA/D1? Wouldnt have been too much extra work would it? Graphics may also be a bit of a weakspot but most common android games should work. These chips will replace the older high end chips and yesterday's high end phones will now become midrange(i.e. WVGA screen, Cortex A8/A9 processor, 512 MB RAM, 5MP camera with 720p video). Some upper mid-range phones which already have these spefications and are out already are the Moto Defy and Samsung Galaxy SL which sell for roughly $350-$400. These sort of phones should slip down even to even lower prices and occupy even the lower mid-range market by late this year/early next year.

720p HD video (oooh its HD ) is still a big marketing feature and is a big advantage for the chips already available in this range (MSM 7230, OMAP 36xx, etc). By the time the ST and Broadcom chips are available they would have been in mass production for more than a year and a half and pricing would have dropped as well. The lack of 720p is going to be a big handicap for them. Of course their pricing will have to be lower but how much of a difference would it be? Except for the CPU(which of course is a huge difference), the MSM 7227 pretty much has all the features that these new chips have, and its been available for what two years now?

 

[Arun]

Some upper mid-range phones which already have these spefications and are out already are the Moto Defy and Samsung Galaxy SL which sell for roughly $350-$400. These sort of phones should slip down even to even lower prices and occupy even the lower mid-range market by late this year/early next year.

We're not talking $250 phones here. We are potentially talking as low as $100 unsubsidised with a BoM of less than $75. Every cent counts. The main differences with the MSM7227 are the CPU for faster applications and browsing (the most important use cases by far) and, even more importantly, lower cost.

I agree, however, that a MSM7230A on 28LP would be a very big deal. 1.4GHz Krait, 720p encode/decode, 14.4Mbps HSPA+ and all at a cost very similar to these 40nm chips. It remains to be seen whether that will happen and in what timeframe.

 

[mczak]

The lack of 720p is going to be a big handicap for them.

Strangely, the MSM7227 seems to be in just about any midrange smartphone these days (so is the chip really expensive? In any case looks to be the fastest ARM11 chip sold in smart phones, might not even be far behind some of the slower-clocked Cortex-A8), and roughly half of them only support video recording with QVGA/15 fps - way below what the chip is supposed to be able to handle...

Where are the Cortex-A5 chips btw? Wouldn't something like dual Cortex-A5 be great for next gen cheap smartphone? The perf/W and perf/area efficiency of the Cortex-A5 seems to be quite a bit better than any of the other recent-ish ARM chips (arm11, cortex-A9, and presumably cortex-A8). But stil no signs of any Cortex-A5 chip, I guess everybody just selling existing ARM11 designs instead...

 

[metafor]

Where are the Cortex-A5 chips btw? Wouldn't something like dual Cortex-A5 be great for next gen cheap smartphone? The perf/W and perf/area efficiency of the Cortex-A5 seems to be quite a bit better than any of the other recent-ish ARM chips (arm11, cortex-A9, and presumably cortex-A8). But stil no signs of any Cortex-A5 chip, I guess everybody just selling existing ARM11 designs instead...

The problem is that the A9 is pretty high on the perf/W and perf/area that it kinda negates the need for an A5 or even dual A5 in the majority of cases. At some point, shrinking the size of the CPU gets you diminishing returns in chip area and an A9 without NEON serves the low-end smartphone segment quite well.

 

[mczak]

The problem is that the A9 is pretty high on the perf/W and perf/area that it kinda negates the need for an A5 or even dual A5 in the majority of cases. At some point, shrinking the size of the CPU gets you diminishing returns in chip area and an A9 without NEON serves the low-end smartphone segment quite well.

Well, arm quotes the A5 as "1/3 the power and die area of A9". Granted it's probably only half as fast, and of course if you look at the whole chip the die size difference won't be that much. Still, the A5 is also quoted as having roughly twice the DMips/mW (granted probably not a very good metric) and I thought that licensing costs are also higher for the more high-end chips (but I've no idea there really) - just all sounds too perfect for lower end phones (and in contrast to the existing arm11 designs, they could run flash - I doubt adobe will ever figure out the compiler switches for armv6...).
Do Cortex-A9 chips without NEON exist yet? Though (unlike A8) both the A5 and A9 seem to have a decent enough VFP implementation, so skipping NEON might be acceptable (unlike the A8 though they don't support the vector VFP instructions, if you'd still want to run such code).

 

[Laurent06]

Do Cortex-A9 chips without NEON exist yet? Though (unlike A8) both the A5 and A9 seem to have a decent enough VFP implementation, so skipping NEON might be acceptable

Tegra2 doesn't have NEON.

(unlike the A8 though they don't support the vector VFP instructions, if you'd still want to run such code).

I have never seen any code using vectors. Anyway on A8 if you use vectors, you can't use the faster NFP pipeline, so it's not that interesting

 

[mczak]

Tegra2 doesn't have NEON.

Oops yes forgot about that. But the Cortex-A8 designs all had NEON right?

I have never seen any code using vectors. Anyway on A8 if you use vectors, you can't use the faster NFP pipeline, so it's not that interesting

Yes that would be more for code written for armv6 with VFP unit - the A8 VFP unit really looks very weak, only there for compatibility reasons.

 

[metafor]

Well, arm quotes the A5 as "1/3 the power and die area of A9". Granted it's probably only half as fast, and of course if you look at the whole chip the die size difference won't be that much. Still, the A5 is also quoted as having roughly twice the DMips/mW (granted probably not a very good metric) and I thought that licensing costs are also higher for the more high-end chips (but I've no idea there really) - just all sounds too perfect for lower end phones (and in contrast to the existing arm11 designs, they could run flash - I doubt adobe will ever figure out the compiler switches for armv6...).

I'm not sure the cost difference really is all that significant for the CPU core license. I would venture to guess that A9 is cheap enough, small enough and low-power enough in single-core, no-NEON configurations that you might as well go for it. The impact on the overall chip may not be big enough to justify it.

 

[Exophase]

Oops yes forgot about that. But the Cortex-A8 designs all had NEON right?

NEON's a core (tightly coupled) part of Cortex-A8 and Cortex-A15. I think there are some performance tradeoffs involved in making NEON easily modular. With Tegra 3 offering NEON I wouldn't be totally surprised if Tegra 2 ends up having been the only Cortex-A series chip made available without it.

 

[Deleted member 13524]

So is the A5 just a huge mistake? I'd have a hard time believing that the A9 can actually cover all the price points, but I haven't seen a single cortex A5 implementation yet, either..

 

[Exophase]

There are still companies using ARM11 though (I'm looking at you Nintendo), so you'd think that Cortex-A5 could have been a meaningful option instead..

But I think Nintendo takes issue with using any core not designed over 5 years ago.

 

[Arun]

Actually, the Cortex-A5 is very real:

Several vendors are already sampling Cortex-A5 processors, Bruce said, with an aim to put out A5 phones in either late 2011 or early 2012.

I'm willing to bet we'll see the Cortex-A5 mostly in... EDGE and TD-SCDMA chips! I don't know if anyone has noticed, but there hasn't been a SINGLE 40nm EDGE chip yet by anyone. And all current 65nm chips are still using ARM9 (except for Infineon using ARM11). A single-core 500MHz Cortex-A5 without L2 cache would be a massive upgrade that could even allow for a decent-ish Android experience.

The fact of the matter is that a Cortex-A9 is very small compared to a 3G baseband, but it's pretty big compared to a 2G one. TD-SCDMA basebands might be 3G and more expensive than EDGE, but they're still cheaper than a 7.2Mbps HSDPA one, and penny pinching is more likely in that market. Oh, Cortex-A5 is definitely also used in the upcoming 40nm CSR SiRFAtlasV with integrated GPS for PNDs (sampling in Q3 2011).

 

[mczak]

Actually, the Cortex-A5 is very real:

Yes but it's pretty nebulous - compared to A15 (which is newer) for instance.

I'm willing to bet we'll see the Cortex-A5 mostly in... EDGE and TD-SCDMA chips! I don't know if anyone has noticed, but there hasn't been a SINGLE 40nm EDGE chip yet by anyone. And all current 65nm chips are still using ARM9 (except for Infineon using ARM11). A single-core 500MHz Cortex-A5 without L2 cache would be a massive upgrade that could even allow for a decent-ish Android experience.

Skipping L2 cache even? Now that sounds like real cheap .

The fact of the matter is that a Cortex-A9 is very small compared to a 3G baseband, but it's pretty big compared to a 2G one. TD-SCDMA basebands might be 3G and more expensive than EDGE, but they're still cheaper than a 7.2Mbps HSDPA one, and penny pinching is more likely in that market. Oh, Cortex-A5 is definitely also used in the upcoming 40nm CSR SiRFAtlasV with integrated GPS for PNDs (sampling in Q3 2011).

Do you have any numbers for that? I've always wondered about die sizes of these SOCs (and how large the invidiual parts of it are) but no die shots tend to get published .
Also maybe that's different in the US but I'm not sure you could really sell a 2G smartphone (even if it's cheap) in Europe. The only 2G smartphone I even know of was the original iPhone...