So the modem is discrete ? Usually you specify when it's integrated and not even sure anymore how they call the modem in SD810, used to be 9x35 but ,maybe they upgraded the name.
The modem/baseband is integrated just as the name suggests. It lists MSM8994, and I also don't know of an 810 SoC without modem. MSM means Mobile Station Modem, their model name for pure application processors is APQ, like the 805 is. https://www.qualcomm.com/products/snapdragon/proce...
Thanks, you always mention when listing the connectivity that it's iP block (when it is), this time you didn't and the integrated modem in the SD810 used to be called MDM9x35 so that was confusing. I thought maybe they are not using the integrated one for so odd reason.
The discrete version i.e. MDM9x35 doesn't support 3x20MHz carrier aggregation. MDM9x45 is pretty recently announced and is a Cat 10 device. So the IP-block that makes up the modem is somewhere between those two products.
Since these are ARM cores, not custom cores, I'm curious what was the goal frequency of the design. In the past, ARM cores have been designed for much lower frequencies than phone makers have wanted to sell, so they either overclocked the cores or used custom cores designed for higher frequencies.
Maybe the problem is just lack of headroom here, so that the old overclocking tricks don't work. If that's the case, then a custom core is the only solution, and the only company that has a custom core is Apple.
I don't think you understand what you are saying. First off, you talk about Apple having the only custom core design, but Qualcomm has had Krait for a while, along with Nvidia's Denver cores in their Tegra K1 that went into the Nexus 9. Secondly you talk about using custom cores designed for higher frequencies, but Apple's main benefit is their really high IPC, so their clock speeds are significantly lower than the typical ARM CPU. Thirdly, name a phone manufacturer whose SoC clock speeds are running at higher than they are spec'd at, otherwise known as overclocking. I can't think of any, at least any major ones, but if you can prove your own statement, maybe I would believe you.
I was only thinking about ARMv8 cores, therefore Krait was not a consideration. Yes I did forget about the Nvidia Denver core, which is an ARMv8 core. So, Apple and Nvidia have ARMv8 custom cores. However everyone else seems to be going with A57 for the time being, and it has been reported that the A57 is having trouble running at the frequencies phone manufacturers desire while remaining within their thermal envelope.
"Then there’s the frequency discussion. Brian and I have long been hinting at the sort of ridiculous frequency/voltage combinations mobile SoC vendors have been shipping at for nothing more than marketing purposes. I remember ARM telling me the ideal target for a Cortex A15 core in a smartphone was 1.2GHz. Samsung’s Exynos 5410 stuck four Cortex A15s in a phone with a max clock of 1.6GHz. The 5420 increases that to 1.7GHz. The problem with frequency scaling alone is that it typically comes at the price of higher voltage. There’s a quadratic relationship between voltage and power consumption, so it’s quite possibly one of the worst ways to get more performance. Brian even tweeted an image showing the frequency/voltage curve for a high-end mobile SoC. Note the huge increase in voltage required to deliver what amounts to another 100MHz in frequency."
Is 8 cores necessary? Seems like if they could drop it down to 6 cores, and be able to run the cpu at a higher frequency, it would be better overall. Apple is still on 2 cores. It seems like the meaningless quest to pack as many "cores"as possible is backfiring - at least for use in phones.
All the octa-core designs I've seen have been 4 high-power cores coupled with 4 low-power cores, with no real provision to use all 8 cores simultaneously - just one set of 4 at a time. It's really a solution tailored for always-on devices which spend most of their time idling or doing occasional background tasks, but have to do (relatively) very heavy computational work at random intervals. e.g. Smartphones.
As for 4 cores vs 2, I suspect the optimal number of cores is actually e. That is, 2.718. That turns out to be the optimal number for a lot of things, including word encoding. i.e. If you're trying to create a written language, at one extreme is Chinese which has a different character for every word. At the other end is computer binary which has just 2 characters (0 and 1) which you combine into a long string to create words. What's the number of characters which best balances word complexity and word length to minimize memory consumption? Turns out to be e. So a trinary encoding (0, 1, 2) is most space-efficient.
In terms of cores, that would mean 3 cores is most efficient. But because computers are designed around powers of 2, it's easier to make 4 cores than 3. My real-world experience seems to back his up too. Dual core systems occasionally lag and stutter, especially if there's one very intensive processing task going on. Quad core systems rarely do.
Basically only the first Exynos SoCs used cluster migration (either the high-power or the low-power clusters could be online; system only saw 4 CPUs).
Everything since then has used CPU migration (for lack of a better term) where the OS sees all 8 CPUs and can online/offline individual cores as needed.
With the exception of the nVidia Tegra X1 (of course) which uses their own custom interconnect and power management system and a variant of cluster migration.
The nvidia Tegra 4 works much better on 4 cores than 2. I think it's one of the few devices in which you can set to two or four cores and it's definitely better on 4.
As for why 8 cores, I have seen a video showing less power consumption using the big.LITTLE combination so, given the vast performance of CPUs these days, who am I to disagree with the hypothesis.
Given these cores are all on their own power plane they can run one or two big cores flat out, and some few number of little cores to handle timer events and the like. Yeah, gts allows for this. My guess is that lg hasn't found the right balance of parameters to feed to the scheduler yet.
I really like my Optimus G, but until I see a more consistent software support pattern from LG I'll never own another one. Considering it's basically the same hardware as the Nexus 4 it's pathetic that they essentially never updated the software beyond a few desperately needed bug fixes. There were rumors of a 4.4.4 update but that was about it. Guess I'll have to root it, none of the newest phones really make me want to drop another $$$ on another phone right now.
I have an LG Optimus G Pro, and the hardware is great, but lack of software support has been frustrating. There were some huge flaws that effected the normal email app, that weren;t fixed until an update 1 year after I bought the phone. 1 year! I tried a few different ROMS, but they just introduced different bugs, so went back to stock. SO yeah - I don't see myself getting another LG phone.
Seems QC on the QC 810 might have been overlooked a little. Wasn't this chip actually earmarked for the Nexus 6 back last year? It got swept under carpet an life went on, and here we are seeing many low clock speed 64 but nothing higher end with the 420. Qualcomm must have a million stepping or bins of the 810 by now I'd think an no creme yet. This year will be slightly different going forward from this one chip. I better X1 can use it with a heat sink in tablets or car dash boards but not no phone.
You mean rushing a 64-bit reference design with no optimizations to market purely for the marketing tag was a bad idea? Who could have seen this coming? That guy Qualcomm demoted for saying just that? Me? Who can tell..
Are you referring to the guy Qualcomm demoted for claiming that Apple's A7 was a marketing gimmick? Because the A7 is not "a 64-bit reference design with no optimizations".
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
24 Comments
Back to Article
jjj - Friday, January 9, 2015 - link
So the modem is discrete ? Usually you specify when it's integrated and not even sure anymore how they call the modem in SD810, used to be 9x35 but ,maybe they upgraded the name.Penti - Friday, January 9, 2015 - link
The modem/baseband is integrated just as the name suggests. It lists MSM8994, and I also don't know of an 810 SoC without modem. MSM means Mobile Station Modem, their model name for pure application processors is APQ, like the 805 is. https://www.qualcomm.com/products/snapdragon/proce...JoshHo - Friday, January 9, 2015 - link
The modem is integrated as denoted by the MSM designation in the SoC. The 9x45 reference is the IP block in the SoC.richworks - Friday, January 9, 2015 - link
The 9x45 is the RF transceiver chip which is independent of the SoCJoshHo - Friday, January 9, 2015 - link
The referenced modem is not a transceiver. The transceiver is most likely the WTR3925.jjj - Friday, January 9, 2015 - link
Thanks, you always mention when listing the connectivity that it's iP block (when it is), this time you didn't and the integrated modem in the SD810 used to be called MDM9x35 so that was confusing.I thought maybe they are not using the integrated one for so odd reason.
Penti - Saturday, January 10, 2015 - link
The discrete version i.e. MDM9x35 doesn't support 3x20MHz carrier aggregation. MDM9x45 is pretty recently announced and is a Cat 10 device. So the IP-block that makes up the modem is somewhere between those two products.milan03 - Sunday, January 11, 2015 - link
- MDM9x45 is Qualcomm's 5th generation discrete baseband processor. It's Cat 10 since it supports Carrier Aggregation on the Uplink as well.-Snapdragon 810 integrated IP Stack is basically MDM9x45, minus that Carrier Aggregation on the Uplink which makes it Category 9.
They're both capable of aggregating up to three 20MHz Component Carriers on the Downlink for the maximum aggregate capacity of 450Mbps.
extide - Sunday, January 11, 2015 - link
You are thinking of the 801/805. 810 is and always has been 9x45SydneyBlue120d - Friday, January 9, 2015 - link
So the Snapdragon 810 issues are all true?http://vr-zone.com/articles/snapdragon-810-delay-r...
ws3 - Friday, January 9, 2015 - link
Since these are ARM cores, not custom cores, I'm curious what was the goal frequency of the design. In the past, ARM cores have been designed for much lower frequencies than phone makers have wanted to sell, so they either overclocked the cores or used custom cores designed for higher frequencies.Maybe the problem is just lack of headroom here, so that the old overclocking tricks don't work. If that's the case, then a custom core is the only solution, and the only company that has a custom core is Apple.
metayoshi - Friday, January 9, 2015 - link
I don't think you understand what you are saying. First off, you talk about Apple having the only custom core design, but Qualcomm has had Krait for a while, along with Nvidia's Denver cores in their Tegra K1 that went into the Nexus 9. Secondly you talk about using custom cores designed for higher frequencies, but Apple's main benefit is their really high IPC, so their clock speeds are significantly lower than the typical ARM CPU. Thirdly, name a phone manufacturer whose SoC clock speeds are running at higher than they are spec'd at, otherwise known as overclocking. I can't think of any, at least any major ones, but if you can prove your own statement, maybe I would believe you.ws3 - Saturday, January 10, 2015 - link
I was only thinking about ARMv8 cores, therefore Krait was not a consideration.Yes I did forget about the Nvidia Denver core, which is an ARMv8 core. So, Apple and Nvidia have ARMv8 custom cores. However everyone else seems to be going with A57 for the time being, and it has been reported that the A57 is having trouble running at the frequencies phone manufacturers desire while remaining within their thermal envelope.
That is what led into my statements about overlocking. That comes from this Anandtech article: http://www.anandtech.com/show/7335/the-iphone-5s-r... which states:
"Then there’s the frequency discussion. Brian and I have long been hinting at the sort of ridiculous frequency/voltage combinations mobile SoC vendors have been shipping at for nothing more than marketing purposes. I remember ARM telling me the ideal target for a Cortex A15 core in a smartphone was 1.2GHz. Samsung’s Exynos 5410 stuck four Cortex A15s in a phone with a max clock of 1.6GHz. The 5420 increases that to 1.7GHz. The problem with frequency scaling alone is that it typically comes at the price of higher voltage. There’s a quadratic relationship between voltage and power consumption, so it’s quite possibly one of the worst ways to get more performance. Brian even tweeted an image showing the frequency/voltage curve for a high-end mobile SoC. Note the huge increase in voltage required to deliver what amounts to another 100MHz in frequency."
kmmatney - Friday, January 9, 2015 - link
Is 8 cores necessary? Seems like if they could drop it down to 6 cores, and be able to run the cpu at a higher frequency, it would be better overall. Apple is still on 2 cores. It seems like the meaningless quest to pack as many "cores"as possible is backfiring - at least for use in phones.Solandri - Friday, January 9, 2015 - link
All the octa-core designs I've seen have been 4 high-power cores coupled with 4 low-power cores, with no real provision to use all 8 cores simultaneously - just one set of 4 at a time. It's really a solution tailored for always-on devices which spend most of their time idling or doing occasional background tasks, but have to do (relatively) very heavy computational work at random intervals. e.g. Smartphones.As for 4 cores vs 2, I suspect the optimal number of cores is actually e. That is, 2.718. That turns out to be the optimal number for a lot of things, including word encoding. i.e. If you're trying to create a written language, at one extreme is Chinese which has a different character for every word. At the other end is computer binary which has just 2 characters (0 and 1) which you combine into a long string to create words. What's the number of characters which best balances word complexity and word length to minimize memory consumption? Turns out to be e. So a trinary encoding (0, 1, 2) is most space-efficient.
In terms of cores, that would mean 3 cores is most efficient. But because computers are designed around powers of 2, it's easier to make 4 cores than 3. My real-world experience seems to back his up too. Dual core systems occasionally lag and stutter, especially if there's one very intensive processing task going on. Quad core systems rarely do.
phoenix_rizzen - Friday, January 9, 2015 - link
Basically only the first Exynos SoCs used cluster migration (either the high-power or the low-power clusters could be online; system only saw 4 CPUs).Everything since then has used CPU migration (for lack of a better term) where the OS sees all 8 CPUs and can online/offline individual cores as needed.
With the exception of the nVidia Tegra X1 (of course) which uses their own custom interconnect and power management system and a variant of cluster migration.
dave1231 - Saturday, January 10, 2015 - link
The nvidia Tegra 4 works much better on 4 cores than 2. I think it's one of the few devices in which you can set to two or four cores and it's definitely better on 4.As for why 8 cores, I have seen a video showing less power consumption using the big.LITTLE combination so, given the vast performance of CPUs these days, who am I to disagree with the hypothesis.
tuxRoller - Friday, January 9, 2015 - link
Given these cores are all on their own power plane they can run one or two big cores flat out, and some few number of little cores to handle timer events and the like. Yeah, gts allows for this. My guess is that lg hasn't found the right balance of parameters to feed to the scheduler yet.djc208 - Friday, January 9, 2015 - link
I really like my Optimus G, but until I see a more consistent software support pattern from LG I'll never own another one. Considering it's basically the same hardware as the Nexus 4 it's pathetic that they essentially never updated the software beyond a few desperately needed bug fixes. There were rumors of a 4.4.4 update but that was about it. Guess I'll have to root it, none of the newest phones really make me want to drop another $$$ on another phone right now.kmmatney - Friday, January 9, 2015 - link
I have an LG Optimus G Pro, and the hardware is great, but lack of software support has been frustrating. There were some huge flaws that effected the normal email app, that weren;t fixed until an update 1 year after I bought the phone. 1 year! I tried a few different ROMS, but they just introduced different bugs, so went back to stock. SO yeah - I don't see myself getting another LG phone.GarageHermit - Friday, January 9, 2015 - link
Seems QC on the QC 810 might have been overlooked a little.Wasn't this chip actually earmarked for the Nexus 6 back last year?
It got swept under carpet an life went on, and here we are seeing many low clock speed 64 but nothing higher end with the 420.
Qualcomm must have a million stepping or bins of the 810 by now I'd think an no creme yet.
This year will be slightly different going forward from this one chip.
I better X1 can use it with a heat sink in tablets or car dash boards but not no phone.
dave1231 - Saturday, January 10, 2015 - link
What odds Qualcomm subconsciously had a downer on the ARM A57/A53 cores as they were not their own designs? Could have happened.coburn_c - Friday, January 9, 2015 - link
You mean rushing a 64-bit reference design with no optimizations to market purely for the marketing tag was a bad idea? Who could have seen this coming? That guy Qualcomm demoted for saying just that? Me? Who can tell..ws3 - Friday, January 9, 2015 - link
Are you referring to the guy Qualcomm demoted for claiming that Apple's A7 was a marketing gimmick? Because the A7 is not "a 64-bit reference design with no optimizations".