Can Qualcomm Kryo do what Krait did

[FORTHEWIND]

With Kryo set to become Qualcomm new flagship custom core, can it bring back Qualcomm's glory days akin to when Snapdragon 600 and 800 came?

We all know how a mess 810 brings (Throttle + pretty lack luster in perf compare to Exynos 7420). So can 820 put the past behind?

And with ARM Cortex A72 going to be out soon, is it performance future-proof enough?:\

 

[Mondozei]

No. The reason is simple. Qualcomm essentially had no competition back then in the Android space.

The only ARM player was Apple and they were and are irrelevant for 80% of the ARM market. Today you have not only Samsung but also LG making high-end ARM SoCs. Mediatek's offerings are much more appealing in terms of prices for mid-to-low end phones while offering decent perfomance. And it is in that price range where the market has been going for the last two years.

 

[FORTHEWIND]

No. The reason is simple. Qualcomm essentially had no competition back then in the Android space.

The only ARM player was Apple and they were and are irrelevant for 80% of the ARM market. Today you have not only Samsung but also LG making high-end ARM SoCs. Mediatek's offerings are much more appealing in terms of prices for mid-to-low end phones, where is where the volume has been going for the last two years.

But as Qualcomm is still in the high-end offering for the time being. Can it make it still before everyone go in-house? *cough*samsung*cough*

IMHO, Kryo and their Zeroth platform still look like it can win back some customer. Except Samsung :\ (Project Mongoose?)

 

[el etro]

Yes, can do.

 

[hemedans]

No. The reason is simple. Qualcomm essentially had no competition back then in the Android space.

The only ARM player was Apple and they were and are irrelevant for 80% of the ARM market. Today you have not only Samsung but also LG making high-end ARM SoCs. Mediatek's offerings are much more appealing in terms of prices for mid-to-low end phones while offering decent perfomance. And it is in that price range where the market has been going for the last two years.

if i remember correct galaxy s2 and s3 use exynos soc, htc one x use tegra soc, nokia use broadcom, etc. there were many competitors but krait crush them all.

 

[podspi]

Hopefully we'll find out soon enough...


Their new SoCs sound like they have a lot of different features (above and beyond just AP cores) that will also differentiate them from the competition. From their press releases and AT's stories on those press releases it sounds like the 820 is going to be a great platform, but of course in the end we'll have to wait and see.

 

[jhu]

Sure, if it can avoid bad press and run cooler this time.

 

[Mopetar]

if i remember correct galaxy s2 and s3 use exynos soc, htc one x use tegra soc, nokia use broadcom, etc. there were many competitors but krait crush them all.

I believe that a large part of the reason Qualcomm was so successful is that they were making their own customized version of the ARM core, whereas the other companies used the stock design. With the 810, Qualcomm did not have a custom core ready or their in-house design did not achieve the results that they wanted so they went with the standard ARM design that everyone else was using.

This meant companies like Samsung could use their own chips without the same historical performance gap, but Samsung also had another advantage in that their own fab tech was able to surpass TSMC at the 20nm node, which meant that they could make an SoC that would give them even better performance.

Kyro may well give Qualcomm back some of their edge, but another part of their falloff was that their HMP implementation wasn't very good which meant that their big.LITTLE design didn't work as well as Samsung's chips. If they're going back to their own core, they may move away from a mixed-core approach, as Apple seems to do well with only on type of core in their SoC so there's no reason Qualcomm couldn't also succeed in such an endeavor.

 

[antihelten]

I believe that a large part of the reason Qualcomm was so successful is that they were making their own customized version of the ARM core, whereas the other companies used the stock design.

The vast majority of the reason why Qualcomm was so successful, was their modem tech. When it came to 3G you either got a Qualcomm chip, or you paid licensing (5% per handset) to Qualcomm (and if you didn't get a Qualcomm SoC, you often had to go with a multi-chip solution).

Qualcomm doesn't have nearly the same patent stranglehold on 4G as they do on 3G, which is why current gen (and future) SoCs will start to see a lot more diversification.

The core architecture itself really has very little to do with the current and past market share makeup. Of course it might matter more going forward if the various SoC manufacturers start seeing more parity in the modem area, and thus have to differentiate themselves in other areas to stay competitive. It might also go in the opposite direction with CPU performance becoming even less important from a competitive standpoint, since ARM SoCs are arguably entering the realm of diminishing returns.

 

[hemedans]

The vast majority of the reason why Qualcomm was so successful, was their modem tech. When it came to 3G you either got a Qualcomm chip, or you paid licensing (5% per handset) to Qualcomm (and if you didn't get a Qualcomm SoC, you often had to go with a multi-chip solution).

Qualcomm doesn't have nearly the same patent stranglehold on 4G as they do on 3G, which is why current gen (and future) SoCs will start to see a lot more diversification.

The core architecture itself really has very little to do with the current and past market share makeup. Of course it might matter more going forward if the various SoC manufacturers start seeing more parity in the modem area, and thus have to differentiate themselves in other areas to stay competitive. It might also go in the opposite direction with CPU performance becoming even less important from a competitive standpoint, since ARM SoCs are arguably entering the realm of diminishing returns.

Nokia is leader of both 3g and 4g patents, with 1/3 of 3g patents and 18-19% of 4g patents. Qualcomm is second to nokia with around 12% of 4g patents.

Nokia and Qualcomm have agreement of 15 years since 2008 to use each other patents in their business. so qualcomm still has around 30% of 4g patents

https://www.qualcomm.com/news/releases/2008/07/23/nokia-and-qualcomm-enter-new-agreement

 

[Mopetar]

The vast majority of the reason why Qualcomm was so successful, was their modem tech. When it came to 3G you either got a Qualcomm chip, or you paid licensing (5% per handset) to Qualcomm (and if you didn't get a Qualcomm SoC, you often had to go with a multi-chip solution).

Even today, Qualcomm generally has better basebands in their SoC because they were doing that before they were making ARM chips so it doesn't hurt either, but even through they were one of the first to move the baseband onto the SoC, they still had a definitive performance edge in terms of computational power over the other players. Their GPU tech (purchased from AMD when they got out of the SoC business) was also typically better than the stock ARM stuff as well, which made them better than almost everything out there are multiple counts.

If they get their custom core to work out, I think they'll enjoy enough of a performance edge to pick up some additional business from Samsung, especially if Qualcomm can get Samsung to fab the chips.

 

[StrangerGuy]

I have little doubt on Kyro at the technical side of things, but I highly doubt the commercial viability with current market conditions it will find itself in.

Remember Krait was released in a time everything else high end was barely able to run Android smoothly and LTE wasnt as widespread. Now, high-end Android and ASPs are in a decline, and you have to look hard to find a phone without LTE these days.

 

[antihelten]

Nokia is leader of both 3g and 4g patents, with 1/3 of 3g patents and 18-19% of 4g patents. Qualcomm is second to nokia with around 12% of 4g patents.

Nokia and Qualcomm have agreement of 15 years since 2008 to use each other patents in their business. so qualcomm still has around 30% of 4g patents

https://www.qualcomm.com/news/releases/2008/07/23/nokia-and-qualcomm-enter-new-agreement

That looks like a fairly old chart, at least it doesn't match up with any of the more recent stuff I could find.

From what I can tell the current 4G patent landscape is very diverse, and the biggest player is actually LG (btw, Optis Cellular Technology is an NPE, controlled by LG, as far as I can tell):

source

Even today, Qualcomm generally has better basebands in their SoC because they were doing that before they were making ARM chips so it doesn't hurt either, but even through they were one of the first to move the baseband onto the SoC, they still had a definitive performance edge in terms of computational power over the other players. Their GPU tech (purchased from AMD when they got out of the SoC business) was also typically better than the stock ARM stuff as well, which made them better than almost everything out there are multiple counts.

If they get their custom core to work out, I think they'll enjoy enough of a performance edge to pick up some additional business from Samsung, especially if Qualcomm can get Samsung to fab the chips.

I'm not saying that Qualcomm didn't have some very nice CPU and GPU tech, merely that even without that stuff they would likely still have taken the majority of the market share (remember without their custom stuff, they would merely have used the generic ARM designs, which wouldn't really have made them any worse off than any of the other ARM manufacturers)

 

[Exophase]

Krait wasn't really that impressive (great at some benches, but some bad glass jaws in real world apps), but Qualcomm beat ARM to the punch by over half a year with something to outdo Cortex-A9.

Then there was kind of an awkward transition period for ARM where Cortex-A15 needed better implementations and more core optimizations to have acceptable enough power consumption, and where a lot of software (and even hardware, early on) needed to be worked out to get big.LITTLE going. But those things have been settled and now they're more and more aggressively iterating on that. So it's going to be harder for Qualcomm to maintain a leadership position in this space.