This website is probably in it's current state precisely because it's owners won't hire the necessary amount of competent people. I assume that the business model of writing quality articles and cashing in through ads isn't very good for the bloodsuckers (I.e. investors). Clickbait articles have a much better return on investment.
Maybe something about the power efficiency of the recent Snapdragon and Dimensity Socs can be found in this video, except Exynos 2200. But we can only use English subtitles to understand it, wish they can translate it into English as soon as possible. https://www.youtube.com/watch?v=kkrzM7i64oE&li...
The logic to Qualcomm's new numbering system has already gone out the window -- or is a "Snapdragon 8+" supposed to be superior to a future "Snapdragon 8 Gen 2"?
Put it another way, shouldn't the "+" have been attached to the "Gen 1" bit?
Qualcomm does it: "Generation.Class", thus backwards to AMD & Intel. 8+ Gen 1 with 8+ = class and Gen 1 = generation. The 7, 8, etc. is Qualcomm's class, so this is the same generation, just a higher class.
I can't explain MediaTek's numbering at all, though.
Except for the fact that Qualcomm also announced the Snapdragon 7 gen 1 on the same day (or at least it was reported here the same day) and it is clearly a lower performance class of soc. So, it is obvious that the intended order is class.generation, not generation.class!
I agree, it’s totally dumb. The + should be tied to the gen part since it’s a gen improvement. The 8 is bound to be the class and it’s exactly the same. Marketing genius striked again.
Look at Apple’s chips. For example their 12th generation. A12 was their lower performance phone and tablet chip. A12X was the same 12th generation but with beefed up core count and memory. A12Z added another GPU core.
Or their 1st generation high end part; M1 was their lowest end part. M1 Pro was their next more powerful, followed by M1 Max and then M1 Ultra, but all were ‘Gen 1’.
The same is true here. Gen 1 hasn’t changed between the 8 and 8+; the 8 and 8+ are like the difference between AX and AZ, or MPro and MMax.
No, the naming conventions are clear. The generation number has no qualifiers and the chip series number has three: 'c' indicating 'compute' (for laptop chips etc.), 'x' indicating 'extreme' (which might suggest a major upgrade to the GPU over the standard part ot something of that order) and '+' which indicates, well, 'plus' (pushed clockrates, for instance, rather than any architectural changes from the standard part). (Note: While Qualcomm already has 'cx' chips I doubt we will ever see an 'x+' or 'cx+' parts.)
The greater significance of this naming scheme seems to be that Qualcomm is intending to cut down on the total number of chip variations. If you just name the chip series, without some extended chip name like 'Snapdragon 778G' or 'Snapdragon 888', say, then you have effectively limited the chip variations that might ship with those designators. So, 'Snapdragon 7' or 'Snapdragon 8' plus a small numbers of qualifiers doesn't allow for a lot of variation. Also, within any generation of a chip all variation will be indicated by that chip series number and the mentioned qualifiers. Will Qualcomm continue to create consistent chip names? Time will tell.
Why does this matter? The most important device to use the 8 Gen 1 is the Galaxy S22, and that's already shipped. THe new Pixel won't have this either. So is it just a couple of gamer phones?
What does sL3 mean? - shared L3? (ie shared across all CPUs, so like every L3 ever pretty much) - synthetic L3 (ie data stored in other L2's can be used by a particular CPU via sideways transfers, ala IBM)? - something else?
Apart from a few Snapdragons I'm not sure what non Exynos IP they've made.
The supposed 780G (5nm) is almost completly MIA, especially when compared to the widely used 778 (6nm TSMC) . and their 888 was sometimes shown to be no better than the (cheaper and 7nm TSMC) 870 just because it thermally throttle before any extra performance could be got out of it - outside short term burst use.
Not aware of a single order Mediatek has given them in the same period.
Plus these comparisons with TSMC's 'equivalent' 4nm process makes Samsung look like a half node behind. Not a good look.
afaik the actual foundry financials are hidden and not separated from their flash/DRAM business which is surely the source of that divisions massive market share/profits.
Hopefully they will make a massive comeback at some point .. maybe first to GAAFET ? (Just like Intel went from paper launching 10nm for over a year to releasing the excellent Alder Lake chips)
..
Can't help thinking their non memory foundry business must be an absolute dumpster fire at the moment. Would love to read the story behind this.
Think back a few years when Apple had 50% of its iPhone chips made on Samsungs 14nm, and 50% made on TSMC’s 16nm. The TSMC chips were 20% more efficient and almost 20% faster. Apple had to work to equalize the performance, but the efficiency differences remained. After everything in a phone are accounted for, such as the screen and the separate modem, the overall difference between the phones came down to about 5%.
But Samsung’s process has never been world beating.
The important thing is that everyone has known about it, and that's why Samsung has been around 30% cheaper than TSMC on similar (should've been) nodes.
Agreed. The QSD 855 chipset from years past was pretty great. Since then all the performance gains came with the increase of energy and heat. In fact, that same YouTube quoted above was able to customise it (Xiaomi Mi 9) with a +45% GPU gain and +35% CPU gain. That allowed it to narrowly surpass the QC 8g1 in performance, and it did it at slightly less power. Basically a tie. So from early 2019 to late 2021, Qualcomm hasn't made an overall improvement.
I think Samsung's lithography is to blame. Their 5nm nodes are more-like 8nm nodes, while their midrange 8nm node is more like 12nm node. That's still ahead of Intel, and far ahead of GlobalFoundaries and SMIC. However, TSMC is just in a different league with their 4nm and 6nm nodes.
Based on a lot of reports from people who have the OnePlus 8 and the Samsung S20 line of phones. It's basically because the Cortex-A77 is less efficient than the Cortex-A76 in lower voltages, but the A77 can increase it's performance further than the A76 based on ARM's projections and designs. The other factor is the thirsty 5G modem in the 865, whereas this was an external option in 855 devices.
If you run only synthetic benchmarks, the QSD 865 will look more efficient. But when you look at the complete package, the QSD 855 stacks up competitively. The QSD 865+ was merely an overclocked QSD 865. However, the QSD 870 actually uses a new process-node in its construct and it is using a new software stack/updated drivers (probably the 5G as well). I haven't heard of QSD 870 phones throttling or having heat issues, and testing shows it has actually lower drain than the 855 and higher sustained performance than 865+. So that's a decent chipset, but nothing worth celebrating when looking at the Apple A13-Bionic, let alone the A14 and A15 variants.
...with all of this said, the chipset is not all there is to a phone/device; I'd much prefer the Samsung A52-s over the Moto Edge X30.
I don't know, just saw a lot of bad feedback about the 865. I can't shrug the feeling that it is basically a side-grade, whereas moving from the QSD 845 to the QSD 855 was an upgrade.
The QSD 870 apparently is using the second-revision of TSMC 7nm node. This is because Qualcomm had provisions there, whilst they sourced Samsung 5nm for the QSD 888/888+. There is likely improvements in the kernels and drivers too. I think there were lot's of improvements in the 5G modem, it doesn't drain the battery and heat the SoC as much as it used to. Which is why this chipset is a proper upgrade (albeit not huge) without drawbacks over the ageing QSD 855.
Meanwhile, the QSD 860 is the same as the QSD 855, simply re-badged and sold cheaply.
Cant say anything about the modem tests, havent seen any. Gpu drivers can make enough of a difference, 870 phones would usually have the newer drivers but golden reviewer tests always put 865s gpu ahead in efficiency.
These data confirm my view that, on first approximation, a supposedly similar node from Samsung is basically a whole generation behind the "same" from TSMC. A while ago, Andrei showed that to be true for Samsung's EUV 5 nm, which performed (efficiency) as well as TSMC's N7 (mostly conventional DUV).
I would love to see Google's next large SoC fabbed by TSMC; those large X cores might finally get to shine. I know it's unlikely, as Google used a lot of Samsung's IP in their Pixel 6 SoCs.
Its not like the VP9/HEVC situation either. AV1 is more competitive, and almost no one wants to touch VVC.
Then again, Qcom aren't going to lose many Snapdragon 8 sales over this. And they don't really care about the longevity of 8-based phones, in fact that would rather have you dump it and buy a newer generation.
Qualcomm is one of the key patent holders for VVC (h.266). They seem to be doing everything they can to ensure VVC wins the format war against AV1 so they can maximize royalties.
I would like to spin an idea around. What about Anandtech introduce a business model similar to CNBC pro? Keep all the review for free but in-depth analysis behind a pay wall. I’ve been coming here daily for 20 years and I cannot imagine this site going offline. I would gladly pay 60$/year if it could bring back the quality we have been used to.
Whereas the Snapdragon 8 Gen 1 built on Samsung's 4nm node is an awful part the Snapdragon 8+ Gen 1 built on TSMC's N4 is actually rather good. Performance and energy efficiency are a lot better and power consumption appears to be have been reduced.
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
51 Comments
Back to Article
Archer_Legend - Friday, May 20, 2022 - link
It would be nice to have a deep dive review of the Exynos 2200, the Snapdragon 8 gen 1 and the Snapdragon 8 gen 1 plus and the Dimensity 9000.Le Geek - Friday, May 20, 2022 - link
I second this, we didn't have such an article this year.Archer_Legend - Friday, May 20, 2022 - link
I know, and those deep dives where really unique, you cannot find such articles on other sites, I hope they start working on it.ballsystemlord - Friday, May 20, 2022 - link
I agree. Go for the deep dive!Kangal - Sunday, May 22, 2022 - link
Does AnandTech have the staff and expertise anymore for those deep-dives?yeeeeman - Sunday, May 22, 2022 - link
Seems like they are in a bit of a bad momentCellar Door - Friday, May 20, 2022 - link
There won't be one - Andrei Frumusanu, who did all those excellent reviews left Anandtech.garblah - Friday, May 20, 2022 - link
Maybe it won't be as in depth and knowledgeable as what Andrei would have done, but hopefully they take a stab at it anyway.DeathArrow - Saturday, May 21, 2022 - link
They just need to hire someone competent enough to write that kind of articles.Wereweeb - Saturday, May 21, 2022 - link
This website is probably in it's current state precisely because it's owners won't hire the necessary amount of competent people. I assume that the business model of writing quality articles and cashing in through ads isn't very good for the bloodsuckers (I.e. investors). Clickbait articles have a much better return on investment.s.yu - Saturday, May 21, 2022 - link
...No wonder we're missing so many in depth articles here.ZdxdUR89 - Friday, May 20, 2022 - link
Maybe something about the power efficiency of the recent Snapdragon and Dimensity Socs can be found in this video, except Exynos 2200. But we can only use English subtitles to understand it, wish they can translate it into English as soon as possible. https://www.youtube.com/watch?v=kkrzM7i64oE&li...jamesindevon - Friday, May 20, 2022 - link
That didn't take long.The logic to Qualcomm's new numbering system has already gone out the window -- or is a "Snapdragon 8+" supposed to be superior to a future "Snapdragon 8 Gen 2"?
Put it another way, shouldn't the "+" have been attached to the "Gen 1" bit?
michael2k - Friday, May 20, 2022 - link
It’s the same design, so I can see why there is no Gen 1+; changing the design is how we got Gen 2, right?So it kinda makes sense to add the plus to 8; though if another variant appears where they do tweak the design means we might see an 8+ Gen 1+
ikjadoon - Friday, May 20, 2022 - link
Yes, it seems the order of "generation.class" is backwards compared to AMD & Intel.Intel & AMD: "Class.Generation.SKU" = i7-8700K with i7 = class, 8 = generation, 700K = SKU. Or Ryzen 9 5950X with Ryzen 9 = class, 5 = generation, 950X = SKU.
Qualcomm does it: "Generation.Class", thus backwards to AMD & Intel. 8+ Gen 1 with 8+ = class and Gen 1 = generation. The 7, 8, etc. is Qualcomm's class, so this is the same generation, just a higher class.
I can't explain MediaTek's numbering at all, though.
vlad42 - Sunday, May 22, 2022 - link
Except for the fact that Qualcomm also announced the Snapdragon 7 gen 1 on the same day (or at least it was reported here the same day) and it is clearly a lower performance class of soc. So, it is obvious that the intended order is class.generation, not generation.class!geoxile - Friday, May 20, 2022 - link
The Plus models have been a thing for years. Normally they're binned chips on the same node but in this case it's a port to a better node.rmfx - Friday, May 20, 2022 - link
I agree, it’s totally dumb. The + should be tied to the gen part since it’s a gen improvement. The 8 is bound to be the class and it’s exactly the same. Marketing genius striked again.michael2k - Saturday, May 21, 2022 - link
It’s not that weird.Look at Apple’s chips. For example their 12th generation. A12 was their lower performance phone and tablet chip. A12X was the same 12th generation but with beefed up core count and memory. A12Z added another GPU core.
Or their 1st generation high end part; M1 was their lowest end part. M1 Pro was their next more powerful, followed by M1 Max and then M1 Ultra, but all were ‘Gen 1’.
The same is true here. Gen 1 hasn’t changed between the 8 and 8+; the 8 and 8+ are like the difference between AX and AZ, or MPro and MMax.
name99 - Friday, May 20, 2022 - link
I believe that, to make this more clear, QC will adopt the terminology of USB, so the next name is going to be Snapdragon 8 Gen 2x2 Superspeed.ChrisGX - Tuesday, May 31, 2022 - link
No, the naming conventions are clear. The generation number has no qualifiers and the chip series number has three: 'c' indicating 'compute' (for laptop chips etc.), 'x' indicating 'extreme' (which might suggest a major upgrade to the GPU over the standard part ot something of that order) and '+' which indicates, well, 'plus' (pushed clockrates, for instance, rather than any architectural changes from the standard part). (Note: While Qualcomm already has 'cx' chips I doubt we will ever see an 'x+' or 'cx+' parts.)The greater significance of this naming scheme seems to be that Qualcomm is intending to cut down on the total number of chip variations. If you just name the chip series, without some extended chip name like 'Snapdragon 778G' or 'Snapdragon 888', say, then you have effectively limited the chip variations that might ship with those designators. So, 'Snapdragon 7' or 'Snapdragon 8' plus a small numbers of qualifiers doesn't allow for a lot of variation. Also, within any generation of a chip all variation will be indicated by that chip series number and the mentioned qualifiers. Will Qualcomm continue to create consistent chip names? Time will tell.
SydneyBlue120d - Friday, May 20, 2022 - link
When can we expect the first Qualcomm SOC made by Intel?Dante Verizon - Friday, May 20, 2022 - link
Late 2033*syxbit - Friday, May 20, 2022 - link
Why does this matter?The most important device to use the 8 Gen 1 is the Galaxy S22, and that's already shipped. THe new Pixel won't have this either. So is it just a couple of gamer phones?
iphonebestgamephone - Saturday, May 21, 2022 - link
Ultra phones from some chinese brands, xiaomi oneplus etc.hnlog - Monday, May 23, 2022 - link
Galaxy (Z) Fold4name99 - Friday, May 20, 2022 - link
What does sL3 mean?- shared L3? (ie shared across all CPUs, so like every L3 ever pretty much)
- synthetic L3 (ie data stored in other L2's can be used by a particular CPU via sideways transfers, ala IBM)?
- something else?
Ryan Smith - Monday, May 23, 2022 - link
You had it right with "shared L3"erotomania - Friday, May 20, 2022 - link
5800X3D? Final writeup?xol - Friday, May 20, 2022 - link
Feels like Samsung has been struggling post 8nm.Apart from a few Snapdragons I'm not sure what non Exynos IP they've made.
The supposed 780G (5nm) is almost completly MIA, especially when compared to the widely used 778 (6nm TSMC) . and their 888 was sometimes shown to be no better than the (cheaper and 7nm TSMC) 870 just because it thermally throttle before any extra performance could be got out of it - outside short term burst use.
Not aware of a single order Mediatek has given them in the same period.
Plus these comparisons with TSMC's 'equivalent' 4nm process makes Samsung look like a half node behind. Not a good look.
afaik the actual foundry financials are hidden and not separated from their flash/DRAM business which is surely the source of that divisions massive market share/profits.
Hopefully they will make a massive comeback at some point .. maybe first to GAAFET ? (Just like Intel went from paper launching 10nm for over a year to releasing the excellent Alder Lake chips)
..
Can't help thinking their non memory foundry business must be an absolute dumpster fire at the moment. Would love to read the story behind this.
Raqia - Friday, May 20, 2022 - link
It's likely they're not going to do much better at the 3nm node:https://www.phonearena.com/news/samsung-foundry-3n...
https://semiwiki.com/semiconductor-manufacturers/t...
nandnandnand - Friday, May 20, 2022 - link
It will be pretty rough if TSMC beats them as usual, but with Samsung having moved to GAAFETs while TSMC holds off on that until 2nm.Some analyses I saw were optimistic about Samsung since they were moving to GAAFET first.
brucethemoose - Sunday, May 22, 2022 - link
They were similarly optimistic about Samsung's aggressive EUV plans.Early GAAFET could be a blessing or a curse.
melgross - Saturday, May 21, 2022 - link
They’ve been struggling since 14nm, at least.Think back a few years when Apple had 50% of its iPhone chips made on Samsungs 14nm, and 50% made on TSMC’s 16nm. The TSMC chips were 20% more efficient and almost 20% faster. Apple had to work to equalize the performance, but the efficiency differences remained. After everything in a phone are accounted for, such as the screen and the separate modem, the overall difference between the phones came down to about 5%.
But Samsung’s process has never been world beating.
Zoolook - Saturday, May 28, 2022 - link
The important thing is that everyone has known about it, and that's why Samsung has been around 30% cheaper than TSMC on similar (should've been) nodes.Kangal - Sunday, May 22, 2022 - link
Agreed.The QSD 855 chipset from years past was pretty great. Since then all the performance gains came with the increase of energy and heat. In fact, that same YouTube quoted above was able to customise it (Xiaomi Mi 9) with a +45% GPU gain and +35% CPU gain. That allowed it to narrowly surpass the QC 8g1 in performance, and it did it at slightly less power. Basically a tie. So from early 2019 to late 2021, Qualcomm hasn't made an overall improvement.
I think Samsung's lithography is to blame. Their 5nm nodes are more-like 8nm nodes, while their midrange 8nm node is more like 12nm node. That's still ahead of Intel, and far ahead of GlobalFoundaries and SMIC. However, TSMC is just in a different league with their 4nm and 6nm nodes.
iphonebestgamephone - Sunday, May 22, 2022 - link
Why are you spreading this misinfo everywhere? 865 is much more efficient than 855. Go watch geekerwans latest video. Someone even linked it here.Kangal - Monday, May 23, 2022 - link
Based on a lot of reports from people who have the OnePlus 8 and the Samsung S20 line of phones. It's basically because the Cortex-A77 is less efficient than the Cortex-A76 in lower voltages, but the A77 can increase it's performance further than the A76 based on ARM's projections and designs. The other factor is the thirsty 5G modem in the 865, whereas this was an external option in 855 devices.If you run only synthetic benchmarks, the QSD 865 will look more efficient. But when you look at the complete package, the QSD 855 stacks up competitively. The QSD 865+ was merely an overclocked QSD 865. However, the QSD 870 actually uses a new process-node in its construct and it is using a new software stack/updated drivers (probably the 5G as well). I haven't heard of QSD 870 phones throttling or having heat issues, and testing shows it has actually lower drain than the 855 and higher sustained performance than 865+. So that's a decent chipset, but nothing worth celebrating when looking at the Apple A13-Bionic, let alone the A14 and A15 variants.
...with all of this said, the chipset is not all there is to a phone/device; I'd much prefer the Samsung A52-s over the Moto Edge X30.
iphonebestgamephone - Monday, May 23, 2022 - link
Just go watch the video. Look carefully at the efficiency curves of 855 and 865.iphonebestgamephone - Monday, May 23, 2022 - link
Oh and look carefully at the 870s curve too.Kangal - Wednesday, May 25, 2022 - link
I don't know, just saw a lot of bad feedback about the 865. I can't shrug the feeling that it is basically a side-grade, whereas moving from the QSD 845 to the QSD 855 was an upgrade.The QSD 870 apparently is using the second-revision of TSMC 7nm node. This is because Qualcomm had provisions there, whilst they sourced Samsung 5nm for the QSD 888/888+. There is likely improvements in the kernels and drivers too. I think there were lot's of improvements in the 5G modem, it doesn't drain the battery and heat the SoC as much as it used to. Which is why this chipset is a proper upgrade (albeit not huge) without drawbacks over the ageing QSD 855.
Meanwhile, the QSD 860 is the same as the QSD 855, simply re-badged and sold cheaply.
iphonebestgamephone - Wednesday, May 25, 2022 - link
Cant say anything about the modem tests, havent seen any. Gpu drivers can make enough of a difference, 870 phones would usually have the newer drivers but golden reviewer tests always put 865s gpu ahead in efficiency.wrkingclass_hero - Saturday, May 21, 2022 - link
Samsung's foundry is floundering so bad even Samsung themselves are going TSMC for their next SOC. The future does not look bright for them.eastcoast_pete - Saturday, May 21, 2022 - link
These data confirm my view that, on first approximation, a supposedly similar node from Samsung is basically a whole generation behind the "same" from TSMC. A while ago, Andrei showed that to be true for Samsung's EUV 5 nm, which performed (efficiency) as well as TSMC's N7 (mostly conventional DUV).I would love to see Google's next large SoC fabbed by TSMC; those large X cores might finally get to shine. I know it's unlikely, as Google used a lot of Samsung's IP in their Pixel 6 SoCs.
eastcoast_pete - Saturday, May 21, 2022 - link
And, oh yeah, boo QC for still boycotting AV1! Why? QC not getting any license fee from that? This is not befitting a " flagship" SoC in 2022/2023!brucethemoose - Sunday, May 22, 2022 - link
Its not like the VP9/HEVC situation either. AV1 is more competitive, and almost no one wants to touch VVC.Then again, Qcom aren't going to lose many Snapdragon 8 sales over this. And they don't really care about the longevity of 8-based phones, in fact that would rather have you dump it and buy a newer generation.
Kangal - Sunday, May 22, 2022 - link
...they're waiting for Apple?vlad42 - Sunday, May 22, 2022 - link
Qualcomm is one of the key patent holders for VVC (h.266). They seem to be doing everything they can to ensure VVC wins the format war against AV1 so they can maximize royalties.felixbrault - Sunday, May 22, 2022 - link
I would like to spin an idea around. What about Anandtech introduce a business model similar to CNBC pro? Keep all the review for free but in-depth analysis behind a pay wall. I’ve been coming here daily for 20 years and I cannot imagine this site going offline. I would gladly pay 60$/year if it could bring back the quality we have been used to.sophiacara - Thursday, June 9, 2022 - link
Hire ReactJS Developers from CronJ to leverage 9+ years of React handling and 15+ industrial experience at just $8 per hour!https://www.cronj.com/hire-react-js-developers.htm...
ChrisGX - Saturday, June 25, 2022 - link
Whereas the Snapdragon 8 Gen 1 built on Samsung's 4nm node is an awful part the Snapdragon 8+ Gen 1 built on TSMC's N4 is actually rather good. Performance and energy efficiency are a lot better and power consumption appears to be have been reduced.See this video review:
https://www.youtube.com/watch?v=sC_nOEGZjsQ
(Non-Mandarin speakers will need to configure suitable subtitles.)