10 ms·

I'd be shocked if this ever made it to production. I'd be more shocked if they did this for 2 generations of pixels. The problem with vertical integration is

I'd be shocked if this ever made it to production. I'd be more shocked if they did this for 2 generations of pixels. The problem with vertical integration is how fast you can be obsolete by competitors. You only need 1 competitor to make a better product and your internal Engineers are going to be begging for it. Unlike Apple customers, the rest of the population is sensitive to performance and cost. If Google can't compete people move on. Edit- Samsung is not a flagship company, their high pricing has nothing to do with performance. At best they are mid tier.

rovr138 6y ago
The problem I see is, will they support it? Or will they send it to the graveyard?
Zenst 6y ago
Maybe a logical progression for the TPU chips and having on-board cores to manage the TPU side of things for performance does make sense and if they can scale this for other products, then why not. Given the whole mobile CPU performance caught up and power usage become the biggest saving area on the table PR financially as well as direct financial savings. Then I'd hazard a educated stab that maybe what is afoot here. However it pans, the Linux kernel is sure going to see some nice ARM optimisations over the comming years out of all this drive.
kevin_thibedeau 6y ago
Apple has a first class in house design team. Google is subbing out so they don't get any long term benefit from an experienced employee base and they have to share a slice of the extra profit with someone else in addition to paying larger license fees to ARM than Apple needs to. This is probably just a bow shot to Qualcomm to get a better deal out of them the same way PC vendors threaten to switch to AMD to squeeze Intel.
- cogman10 6y ago
  I think that's very much the case. Qualcomm has been price gouging while limiting features from each new release (for example, the new 888 doesn't have AV1 decode support... WTF!) This is google forcing them to up their game if they want to keep playing. Will google keep doing this? Probably not (IMO). Rather, it's likely just a demonstration of "Hey, we don't need you, so straighten up!" That being said, if Qualcomm ignores them (since Pixel isn't a major player in the phone market) google may keep producing their own chips. Another interesting aspect here is that Google may decide that they don't want to keep paying for the ARM licenses. This is an avenue for them to start getting RISC-V as a first class Android citizen.
  - kevin_thibedeau 6y ago
    Android on RISC-V will go over as well as it did on Atom. Apps with native ARM code will suck or be unusable.
    - cogman10 6y ago
      I actually think there's a pretty easy way around this for Google. They could create a ARM -> RISC-V translation layer. That works well because RISC-V already has a very limited instruction set. In order to make things fast, you don't need to do a lot of fancy transforms. The issue with doing the same ARM->x86 translation is the x86 instruction set is vast and to get the best performance requires you effectively use those instructions. That requires you to merge instructions that you wouldn't have merged (think of things like the lea instruction). There's less of an instruction impedance mismatch going from ARM to RISC-V. Interestingly, I think going from x86->ARM or RISC-V would present less of a performance problem. That's because most of the work would be splitting one instruction into many. The evidence of this is the x86->IA-64 which resulted in something like a 10->20% performance loss vs a natively compiled IA-64.
  - lern_too_spel 6y ago
    The real risk to Qualcomm is if Samsung uses this in their phones and Chromebooks as well.
acdha 6y ago
> Unlike Apple customers, the rest of the population is sensitive to performance and cost. If Google can't compete people move on. This is an odd phrasing: Apple's pricing isn't higher than the competition — flagship Android phones frequently cost noticeably more — and when it comes to performance it's very hard to beat the $400 iPhone SE2 at any price. Vertical integration is what allows Apple to be ahead on both price and performance. The only way Google is matching that is with a strong long-term commitment to make similar investments. Apple will falter at some point but after years of ignoring the basics Google won't be able to take advantage of that — especially since at this point they're chasing Apple's previous generation so it'd take an extended sag for Google to catch up much less surpass with their current half-hearted strategy.
- Dylan16807 6y ago
  > Vertical integration is what allows Apple to be ahead on both price and performance. I don't know about that. The processor is the main thing they have vertically integrated, right? And if I go look at a flagship like a Galaxy S20 Ultra 5G I see a BOM of $57 for the processor. (It does have a hideously expensive modem for 5G but it didn't need to have that.)
  - cmsj 6y ago
    Consider the knock-on effects though - that processor isn't necessarily optimising the things Samsung might want, nor is Android, which costs them more in other ways (I'm particularly thinking of the fact that Android phones seem to have to include a lot more RAM than iPhones, so that's another chunk of BOM).
  - acdha 6y ago
    They control the entire system on a chip — remember how long it was before Android devices started to have the security features which Apple introduced in the iPhone 3GS? That took Google getting phone vendors to get the CPU vendors to all work together, whereas at Apple that's a couple of teams all working on the same product with the same incentives to make it succeed.
schoolornot 6y ago
Apple has a 10 year lead time on Google for running OSes on in-house Silicon. If there is no intention to compete with Qualcomm, how are they going to get the economies of scale problem down to where the chips are affordable + performant? The volumes of Chromebooks/Pixels sold in a week vs. iPhones in a week is stark.
- cromwellian 6y ago
  "X has an insurmountable lead" almost always turns out to be wrong. You could have made the same argument with x86. Intel/AMD used to at least have the argument that they had vertically integrated because they not only designed the chips, but also designed the fabs. Apple is a fabless designer, other vendors have access to TSMC as well, as long as they have space capacity. Realistically, other ARM vendors are about 2 years behind. A Snapdragon 875 scores the same as an Apple A12 on Single Thread Geekbench. On GPU and TPU, Apple isn't that far ahead at all. TPUs are relatively simple devices, and the Snapdragon 888 already has almost double the A14 performance in TOPS on paper. GPU wise, the A14 (PowerVR derived) GPU is roughly equivalent to a 2016 era NVidia 1060. AMD's RDNA2 based cores for SOCs are likely more powerful and more capable (e.g. raytracing acceleration) Much is made of how far Qualcomm is behind Apple, but like with AMD and Intel, Qualcomm has a wider market focus to address. Apple sells only a few SKUs, Qualcomm needs to make chips for a much wider array of demands, and so they're a jack of all trades, master of none. In much the same way, AMD and Intel have desktops and enterprise vendors to satisfy, including people running SMP systems. Does the M1 support multiple M1 SMP? Does it support ECC? The other vendors are kind of hamstrung by trying to make one architecture that pleases too many markets at once. Look at the Anandtech Snapdragon 888 article: https://www.anandtech.com/show/16271/qualcomm-snapdragon-888-deep-dive https://www.anandtech.com/show/16271/qualcomm-snapdragon-888... "Qualcomm’s 25% generational boost is also less than Arm’s advertised 30% as the new S888 continues to use a 4MB L3 cache for the CPU cluster, versus Arm’s envisioned 8MB configuration for a high-end 5nm SoC with the new X1 cores. Qualcomm explained to us that this was simply a balance between cost, implementation effort, and diminishing returns of a higher cache configuration design." Basically, Qualcomm needs to worry about what their customers are willing to pay for the chip and how much work it takes to integrate. Apple doesn't, if need be, they can bump the price for a more expensive SOC, Qualcomm Android vendors can't. 25% CPU performance uplift and 30% GPU performance uplift for the 888. the A14 gets 1583 Single Thread in Geekbench, the 888 will likely turn in a final Single Core perf between 1300-1450, only 9% less than the A14. The situation is the same for multi-core, roughly on par, and on Antutu, the Snapdragon 875 beats the A14 substantially, and wins on several 3dMark tests. Apple does not have an insurmountable lead over the other vendors in terms of SOC, I'd argue that their primary advantage right now is their software has always been better. In particular, WebKit's JIT for ARM does a lot better than V8 for ARM, and Android Dalvik does not produce code that runs as well as LLVM on Swift (plus there's generational GC vs ARC+refcounting). Otherwise, a 9% CPU advantage in single thread performance is nothing to write home about.
nxc18 6y ago
I don’t get how you can look at Apple’s mobile offerings, which have been years ahead of all the Android competition, and come to the conclusion that Apple customers don’t care about performance. I’d say Apple’s customers are far more sensitive to performance than Google’s given what Android users have been willing to put up with (e.g. flagship SoCs that are years behind what’s in last year’s iPhones). If you factor in broader aspects of performance (e.g. Face ID, fingerprint unlock speed) it’s clear that the _only_ thing most Android users care about is cost. I’m not saying this to diss them, I just think the Android manufacturers owe it to their users to actually produce competitive hardware and software. P.S. if you factor in longevity, Apple products are usually cheaper, too.
- aylmao 6y ago
  > Unlike Apple customers, the rest of the population is sensitive to performance and cost. If Google can't compete people move on. Maybe they meant both independently. As an example: - A friend of mine just upgraded out of his iPhone 6. Until recently performance and storage seemingly just weren't that important to them. - Another friend of mine will just buy iPhones, no matter the cost. Both of these friends stick with Apple because they like and care about other things about the ecosystem more so than performance or cost (independently), whereas OP was perhaps implying that Android users tend to care more about "performance to cost ratio" since they could as well just buy a different phone from another maker and get the same Android experience. This of course, in the context sticking with one phone maker or another.
  - jdhn 6y ago
    As someone who upgraded this year from the 6 to the SE 2, I held out not only because I like the ecosystem, but also because I'm not shelling out almost one grand for a phone. The SE hit the sweet spot, so I ordered one the first day it came out.
    - aylmao 6y ago
      I wish I had waited and gotten an SE 2, especially because Touch ID is way more convenient than Face ID nowadays that one's often wearing a face mask haha
cmsj 6y ago
The smart move for Google here would be to debut their own chips in Pixels, but then sell the chips to other phone manufacturers to use in their phones. Assuming the chips provide enough raw performance and interesting hardware acceleration, that would make it a reasonable business for Google to stay involved in.
vinay427 6y ago
> Samsung is not a flagship company, their high pricing has nothing to do with performance. At best they are mid tier. Their high pricing has to do with the slew of software and hardware features they include, many of which are executed at least reasonably well and later trickle down to mid-tier brands like Google's Pixel line or more conservative high-end brands like Apple. For instance: DeX, more flexible fast USB C and wireless charging than most manufacturers, pen/remote support with in-device charging and corresponding apps, Samsung Pay with magnetic stripe emulation, etc. Samsung isn't focused on raw computing power, because they largely depend on the same chips as most of their manufacturers, but if you've used a newer Galaxy S/Note device the level of hardware polish and random "gimmicky" but occasionally useful features far outpaces a Pixel or similar. Disclaimer: I currently own a Pixel (because there are other things that Google competes on) but last owned a Note 9.

rovr138 6y ago

The problem I see is, will they support it? Or will they send it to the graveyard?

Zenst 6y ago

Maybe a logical progression for the TPU chips and having on-board cores to manage the TPU side of things for performance does make sense and if they can scale this for other products, then why not. Given the whole mobile CPU performance caught up and power usage become the biggest saving area on the table PR financially as well as direct financial savings. Then I'd hazard a educated stab that maybe what is afoot here. However it pans, the Linux kernel is sure going to see some nice ARM optimisations over the comming years out of all this drive.

kevin_thibedeau 6y ago

Apple has a first class in house design team. Google is subbing out so they don't get any long term benefit from an experienced employee base and they have to share a slice of the extra profit with someone else in addition to paying larger license fees to ARM than Apple needs to. This is probably just a bow shot to Qualcomm to get a better deal out of them the same way PC vendors threaten to switch to AMD to squeeze Intel.

cogman10 6y ago
I think that's very much the case. Qualcomm has been price gouging while limiting features from each new release (for example, the new 888 doesn't have AV1 decode support... WTF!) This is google forcing them to up their game if they want to keep playing. Will google keep doing this? Probably not (IMO). Rather, it's likely just a demonstration of "Hey, we don't need you, so straighten up!" That being said, if Qualcomm ignores them (since Pixel isn't a major player in the phone market) google may keep producing their own chips. Another interesting aspect here is that Google may decide that they don't want to keep paying for the ARM licenses. This is an avenue for them to start getting RISC-V as a first class Android citizen.
- kevin_thibedeau 6y ago
  Android on RISC-V will go over as well as it did on Atom. Apps with native ARM code will suck or be unusable.
  - cogman10 6y ago
    I actually think there's a pretty easy way around this for Google. They could create a ARM -> RISC-V translation layer. That works well because RISC-V already has a very limited instruction set. In order to make things fast, you don't need to do a lot of fancy transforms. The issue with doing the same ARM->x86 translation is the x86 instruction set is vast and to get the best performance requires you effectively use those instructions. That requires you to merge instructions that you wouldn't have merged (think of things like the lea instruction). There's less of an instruction impedance mismatch going from ARM to RISC-V. Interestingly, I think going from x86->ARM or RISC-V would present less of a performance problem. That's because most of the work would be splitting one instruction into many. The evidence of this is the x86->IA-64 which resulted in something like a 10->20% performance loss vs a natively compiled IA-64.
- lern_too_spel 6y ago
  The real risk to Qualcomm is if Samsung uses this in their phones and Chromebooks as well.

acdha 6y ago

> Unlike Apple customers, the rest of the population is sensitive to performance and cost. If Google can't compete people move on. This is an odd phrasing: Apple's pricing isn't higher than the competition — flagship Android phones frequently cost noticeably more — and when it comes to performance it's very hard to beat the $400 iPhone SE2 at any price. Vertical integration is what allows Apple to be ahead on both price and performance. The only way Google is matching that is with a strong long-term commitment to make similar investments. Apple will falter at some point but after years of ignoring the basics Google won't be able to take advantage of that — especially since at this point they're chasing Apple's previous generation so it'd take an extended sag for Google to catch up much less surpass with their current half-hearted strategy.

Dylan16807 6y ago
> Vertical integration is what allows Apple to be ahead on both price and performance. I don't know about that. The processor is the main thing they have vertically integrated, right? And if I go look at a flagship like a Galaxy S20 Ultra 5G I see a BOM of $57 for the processor. (It does have a hideously expensive modem for 5G but it didn't need to have that.)
- cmsj 6y ago
  Consider the knock-on effects though - that processor isn't necessarily optimising the things Samsung might want, nor is Android, which costs them more in other ways (I'm particularly thinking of the fact that Android phones seem to have to include a lot more RAM than iPhones, so that's another chunk of BOM).
- acdha 6y ago
  They control the entire system on a chip — remember how long it was before Android devices started to have the security features which Apple introduced in the iPhone 3GS? That took Google getting phone vendors to get the CPU vendors to all work together, whereas at Apple that's a couple of teams all working on the same product with the same incentives to make it succeed.

schoolornot 6y ago

Apple has a 10 year lead time on Google for running OSes on in-house Silicon. If there is no intention to compete with Qualcomm, how are they going to get the economies of scale problem down to where the chips are affordable + performant? The volumes of Chromebooks/Pixels sold in a week vs. iPhones in a week is stark.

cromwellian 6y ago
"X has an insurmountable lead" almost always turns out to be wrong. You could have made the same argument with x86. Intel/AMD used to at least have the argument that they had vertically integrated because they not only designed the chips, but also designed the fabs. Apple is a fabless designer, other vendors have access to TSMC as well, as long as they have space capacity. Realistically, other ARM vendors are about 2 years behind. A Snapdragon 875 scores the same as an Apple A12 on Single Thread Geekbench. On GPU and TPU, Apple isn't that far ahead at all. TPUs are relatively simple devices, and the Snapdragon 888 already has almost double the A14 performance in TOPS on paper. GPU wise, the A14 (PowerVR derived) GPU is roughly equivalent to a 2016 era NVidia 1060. AMD's RDNA2 based cores for SOCs are likely more powerful and more capable (e.g. raytracing acceleration) Much is made of how far Qualcomm is behind Apple, but like with AMD and Intel, Qualcomm has a wider market focus to address. Apple sells only a few SKUs, Qualcomm needs to make chips for a much wider array of demands, and so they're a jack of all trades, master of none. In much the same way, AMD and Intel have desktops and enterprise vendors to satisfy, including people running SMP systems. Does the M1 support multiple M1 SMP? Does it support ECC? The other vendors are kind of hamstrung by trying to make one architecture that pleases too many markets at once. Look at the Anandtech Snapdragon 888 article: https://www.anandtech.com/show/16271/qualcomm-snapdragon-888-deep-dive https://www.anandtech.com/show/16271/qualcomm-snapdragon-888... "Qualcomm’s 25% generational boost is also less than Arm’s advertised 30% as the new S888 continues to use a 4MB L3 cache for the CPU cluster, versus Arm’s envisioned 8MB configuration for a high-end 5nm SoC with the new X1 cores. Qualcomm explained to us that this was simply a balance between cost, implementation effort, and diminishing returns of a higher cache configuration design." Basically, Qualcomm needs to worry about what their customers are willing to pay for the chip and how much work it takes to integrate. Apple doesn't, if need be, they can bump the price for a more expensive SOC, Qualcomm Android vendors can't. 25% CPU performance uplift and 30% GPU performance uplift for the 888. the A14 gets 1583 Single Thread in Geekbench, the 888 will likely turn in a final Single Core perf between 1300-1450, only 9% less than the A14. The situation is the same for multi-core, roughly on par, and on Antutu, the Snapdragon 875 beats the A14 substantially, and wins on several 3dMark tests. Apple does not have an insurmountable lead over the other vendors in terms of SOC, I'd argue that their primary advantage right now is their software has always been better. In particular, WebKit's JIT for ARM does a lot better than V8 for ARM, and Android Dalvik does not produce code that runs as well as LLVM on Swift (plus there's generational GC vs ARC+refcounting). Otherwise, a 9% CPU advantage in single thread performance is nothing to write home about.

nxc18 6y ago

I don’t get how you can look at Apple’s mobile offerings, which have been years ahead of all the Android competition, and come to the conclusion that Apple customers don’t care about performance. I’d say Apple’s customers are far more sensitive to performance than Google’s given what Android users have been willing to put up with (e.g. flagship SoCs that are years behind what’s in last year’s iPhones). If you factor in broader aspects of performance (e.g. Face ID, fingerprint unlock speed) it’s clear that the _only_ thing most Android users care about is cost. I’m not saying this to diss them, I just think the Android manufacturers owe it to their users to actually produce competitive hardware and software. P.S. if you factor in longevity, Apple products are usually cheaper, too.

aylmao 6y ago
> Unlike Apple customers, the rest of the population is sensitive to performance and cost. If Google can't compete people move on. Maybe they meant both independently. As an example: - A friend of mine just upgraded out of his iPhone 6. Until recently performance and storage seemingly just weren't that important to them. - Another friend of mine will just buy iPhones, no matter the cost. Both of these friends stick with Apple because they like and care about other things about the ecosystem more so than performance or cost (independently), whereas OP was perhaps implying that Android users tend to care more about "performance to cost ratio" since they could as well just buy a different phone from another maker and get the same Android experience. This of course, in the context sticking with one phone maker or another.
- jdhn 6y ago
  As someone who upgraded this year from the 6 to the SE 2, I held out not only because I like the ecosystem, but also because I'm not shelling out almost one grand for a phone. The SE hit the sweet spot, so I ordered one the first day it came out.
  - aylmao 6y ago
    I wish I had waited and gotten an SE 2, especially because Touch ID is way more convenient than Face ID nowadays that one's often wearing a face mask haha

cmsj 6y ago

The smart move for Google here would be to debut their own chips in Pixels, but then sell the chips to other phone manufacturers to use in their phones. Assuming the chips provide enough raw performance and interesting hardware acceleration, that would make it a reasonable business for Google to stay involved in.

vinay427 6y ago

> Samsung is not a flagship company, their high pricing has nothing to do with performance. At best they are mid tier. Their high pricing has to do with the slew of software and hardware features they include, many of which are executed at least reasonably well and later trickle down to mid-tier brands like Google's Pixel line or more conservative high-end brands like Apple. For instance: DeX, more flexible fast USB C and wireless charging than most manufacturers, pen/remote support with in-device charging and corresponding apps, Samsung Pay with magnetic stripe emulation, etc. Samsung isn't focused on raw computing power, because they largely depend on the same chips as most of their manufacturers, but if you've used a newer Galaxy S/Note device the level of hardware polish and random "gimmicky" but occasionally useful features far outpaces a Pixel or similar. Disclaimer: I currently own a Pixel (because there are other things that Google competes on) but last owned a Note 9.