Got to say the only reason why I didn't upgrade from my SD845 device to an 865 this year was all the reports of issues with the 120hz AMOLED panels i.e. green tint, image retention, black crush.
Thanks Andrei! I look forward to your first test and hopefully a deep dive once you were able to put an 888 through your tests.
As an aside, my standard gripe with all 2020/21 ARM chips that are not from Apple: why are we still on A55 with its in-order design for the efficiency cores? Apple's efficiency cores seem to be a lot more "efficient" (perf/wh), pun intended. What's up with that, ARM, QC and others?
There are three components to measuring efficiency: work/watt, die area, and absolute watt.
Power is consumed performing work, so work/watt tells us how much time is necessary powering the efficiency cores.
Die area consumes power even when idle, so a smaller die area will boost efficiency by reducing the amount of powered transistors when no work is done.
You can also measure power usage, without measuring time.
So in that sense the efficiency cores ARM has, the A55, are a balance of the three, Apple is prioritizing perf/watt since they don't care about the silicon costs (die size) and reduce power usage there by using the smaller 5nm transistors.
I suspect that later on we will have a refresh of this SoC as S888+, similar to what was done with S865/S865+ .... where clocks are bumped as yields improve. Not needed now, but probably something QualComm is holding back... The S865 to S865+ move was a good way for QualComm to refresh it's product line to get premium dollars without a full redesign or generation jump. It's not hard to imagine a repeat of the strategy with S888/S888+
Well, there are probably lots of reasons that Qualcomm follows up on flagship SoC releases with a Plus version. If that happens this time the reason will be simple. The 5nm LPE process node used for the SD888 was a bit under par. Generally higher core frequencies (say, 5 - 10% higher) across the board for all CPU cores and thus higher performance should have been a goal within the reach of the designers of the SD888 without running into power consumption or thermal troubles. So, if a better silicon production process from Samsung should become available that doesn't necessitate lots of expensive rework by Qualcomm we may well see a Plus variant. Otherwise, perhaps, we won't. After shoehorning a very powerful GPU and modem in there it isn't clear that Qualcomm any longer has the leeway it needs to just push ahead on the performance front without regard to the power and thermal consequences.
^ This statement incorrectly implies Qualcomm is poor or doesn't have the resources to try to match Apple's superiority in the field. Qualcomm has be strong-arming manufacturers and making everyone pay a mafia cut for over a decade now and has all the money for this. Just not the will since they're free loading on the patent/royalty cuts and churning out garbage chips since there isn't a competitor in the Android world. Its crazy how apple is crushing every Qualcomm SoC with their mobile CPUs.
What this statement ignores is that there are many major competitors in the SOC world. All of them lag far behind Apple, but in second is pretty consistently Qualcomm. Qualcomm does peak better, it does sustained better, and it does graphics better than any of its non-Apple competition. Huawei comes close occasionally. Samsung, Mediatech, and Rockchip don't.
Apple makes expensive chips that they know they can sell, and they fab farther away from the ARM reference than the other manufacturers.
Android SoC vendors aren't willing to spend a lot of area on big cores and huge caches: here we have 4x512KB L2 plus a tiny 4MB L3 vs A14's 12MB L2 and 16MB L3.
That has nothing to do with it at all. Android SOC vendors - really just Qualcomm, MediaTek and Samsung unless you want to count Huawei - chose to use more cores instead of bigger ones. While the competition - Apple - was until recently still using 2 cores and no one was making a real effort to make ARM laptops beyond a few cheap Chromebooks, it wasn't a problem. But when Apple went to 4 and then 6 cores and started crushing Android SOCs practically overnight, then there was a problem. MediaTek tried to address it with a 10 core design with 2 fastest, 4 fast and 4 efficiency cores, but it only gave them a 10% performance increase over their octacore designs. So ARM and Samsung went back to the drawing board to create a much better single core design: the X1. But this is a first generation design. There are still issues that they need to work out in order to have this core perform even faster. Also, they need to go from 1+3+4 design to a 2+2+4 design, from 1 Cortex X1 to 2, if they are going to have any chance of rivaling the A14. The problem is that by the time they perfect the design, Apple may well have 4 Firestorm performance cores instead of 2 for their smartphone chips, and on a 3nm process at that. But the biggest problem - coming up with their own big core design - has been solved.
Ultimately, there is no "need" to play catchup with Apple smartphone speeds anyway. The vast majority of Android phones sold cost under $400, and the expensive Android phones are so because they offer a ton of features that iPhones won't have for 3-4 years. The real need is to come out with a PC chip for laptops and desktops to compete with the M1. That is the real value of the Cortex X1. We might even see Qualcomm make another try at being a supplier for ARM servers again - they began to in 2015 but gave up and pulled the plug in 2018 - with Samsung potentially joining them.
No - multithreaded performance is not an issue. That's already within 10% of A14 as the results above show. However single-threaded GB performance is 40% lower. The big issue is that the big cores don't have nearly enough cache, so they don't run as fast as they could. That can be solved by adding more cache, and lots and lots of it. L2 could become 1MB per core like in the Neoverse cores, L3 quadrupled to 16MB and 8-16MB as a system cache. That with an improved X1 core at similar frequency as SD865+ should match A14 both single and multithreaded.
I'd say 4 big cores is more than enough (whether it is 1+3 or 2+2). Replacing the 4 little cores with one A78 variant optimized for best perf/Watt would be a good idea too.
Whether it is worth chasing A14 is a good question indeed, but the key problem is willingness to spend lots of area on big cores and big caches (which either means higher cost or having to cut down GPU or AI performance).
You can see the diminishing returns in increasing on-die cache with A14 vs M1 L2 cache increase from 8 to 12MB. One core can't access all of the increase, and there are more big cores in the M1, but still a 33% increase in what one core can access (6 vs 8MB?) only gives a 7% increase in single core performance.
That 7% might be worth it on a MB Pro, but it's most definitely not worth it for a phone SoC. QC SoCs are smaller than Apple SoCs, the 845 was the first to have a SLC/L3, and the 6/700-series SoCs still don't have it. More on-die cache is always good, it's a matter of price, though I would agree QC could do better.
The A14 also has 4 memory channels, 16K pages, wider cores fed by more OoO execution and larger caches, and the perf/watt is impressive enough for low power, that I would say we should be seeing some of the same things from other ARM designs and even AMD/Intel eventually.
You are completely ignorant. Larger caches absolutely play a part in higher IPC. What do you think contributes to "larger" cores? They dont just throw more transistors into a FPU and call it a day.
They don't really. AMD has had 2x the cache Intel had and Intel always had better IPC than them. Until recently when cores were drastically redesigned.
If we take some rumors Apple will be focusing on power efficiency with A15 and potentially it can be 4+4 design. They do not need to push sigle thread performance much but not sure how they can fit 2 more cores on die. Also CPU performance in mobile world becoming nearly irrelevant as chips are blazing fast and more focus is now put on AI and graphic. From this point is 888 interesting and I wonder how all will be able to utilize. Let see what it brings to users and if it will leapfrog Apple in this. In desktop ARM designs I wonder if they catch up in two years somehow or some other vendor.
The problem is, nothing can touch Apple's single core performance. There is nothing stopping any chip maker from topping Apple's multi-core performance. Qualcomm could simply just add more cores.
Qualcomm doesn't design their own CPU cores anymore, so it's really up to ARM to decide if that's where they want to go. I thought that was what the X1 was supposed to be for. The X1 seems to be what the A78 should have been.
Agreed that there is no real need to add more cores for a phone chip. There are already thermal limitations to deal with. Qualcomm had done the right thing by focusing on AI/ML and making their GPU more competitive.
1. All the mobile vendors, save Apple, only make tiny changes to the arm core design ← this is by far the dominant factor, but there's a reason why the X-1 is their first "big" core & that's.... 2. Cost. Any of the core designers at the various companies could explain how to design a competitive core to Apple's but the core will be huge... just like Apple's ← for fast non-mobile arm-based cores look to arm's neoverse, Qualcomm's falkor, Calcium's thunderx/2,
Because selling such chips would be commercially unviable. Apple's chip division has no such restriction and can therefore build bigger chips or get in early on the newest process. Most of apple's success is actually a result of tsmc rather than apple.
One: Qualcomm is using a much worse process node, Samsung 5nm is around 2/3 as dense as TSMC 5nm. This has a huge impact on especially low wattage performance.
Secondly, cache. Cache is probably one of the more expensive things you can do to increase performance of a chip, e.g. Architecture change is once off, but cache will be paid linearly, so it's usually one of the cases where cost cuts are seen first if they can get away with it.
Qualcomm doesn't need to beat Apple, they just need to beat the rest of the Android competition.
Why not just ship reference design devices to reviewers ?
Nothing aggainst preventing thespred of viruses, but just posting supposed numbers as “best of the best” in every category seems suspicious as if there was a caveat or two.
Do we at least know if RD was a tablet or smartphone ? Or any info at all ?
Doesn't matter. The first batch of phones containing this new chip will be out in less than month, whether it will be the Samsung Galaxy S21 phones that launch January 14th or a Chinese phone that beats it to market by a week or 2. Either way it won't be much longer before we have independent benchmarks.
When was the last time qcom provided unreliable perf? They're obviously going to cherry pick the particular benchmarks but their actual results should be replicable. In the past these initial benchmarks tend to be on the lower side. In this case, there's no reason to be suspicious... HOWEVER, power draw is a completely different matter.
+23.5% performance is a nice improvement, but I can’t help but be a little disappointed when Arm was claiming +30% IPC *and* frequency increases. Here’s hoping that Qualcomm is just being conservative with peak performance for the smartphone chips to save power, and that the 8CX equivalent will be able to run at higher frequencies!
No, I believe this was Anandtech's misinterpretation. I remember reading ARM's information on the X1 and disagreeing with Anandtech's assessment. ARM's claim of 30% performance improvement for X1 assumes process node bump AND clock speed of 3Ghz. You can see this in their slides, it's in the fine print. ARM didn't mislead anyone, in this case Anandtech's analysis did.
Apple has two very high performance cores, Qualcomm has one high performance and three medium-high performance cores.
Apple dedicates around twice as much die area as Qualcomm to CPU cores and private cache, yet Qualcomm gets 70-90% of their performance at similar efficiency. That’s very quite decent. Not very exciting, but pragmatic.
Apple also controles the entire stack. We cant run android on an iphone to test this, but I'd wager part of the A core's performance is due to top to bottom optimization that just doesnt happen on android devices.
I have my doubts the optimization are anything only iOS would benefit from. More so they optimize against specific use cases common on an iPhone or iPad, such as web browser JavaScript, which the M1 is exceedingly proficient at and other browsers also benefit from.
People always say that but I disagree. Apple really makes great general purpose CPUs. Many benchmarks and applications compiled for standard ARM64 flies on them and even ARM Windows shows fantastic performance running inside a VM. They have lots of general purpose features: Like that 8-wide decoding with a huge ROB and large+fast caches. The only other CPU with similar architecture is the big iron POWER10 (as mentioned by Anandtech).
The reason is that the X1 core is brand new. It was a joint design project by ARM Holdings and Samsung to catch up to Apple Ax and it was just completed last year. Qualcomm, Samsung and the rest went from a big.Little 4+4 architecture to a 1+3+4 architecture awhile back because they knew it was on the way. Getting 2 X1 cores to match the Apple A14 will be next. I would be shocked if Samsung isn't already working on it because that is basically their only path to competing with Qualcomm since they have given up on custom cores. After that it will be getting to 4 Cortex X1 chips like the M1. Hopefully they will achieve it by the time they get to a 3nm process in 2022.
So you're telling me samsung was involved in the x1's development, but still used 2+2+4 config and was also working on thier custom m cores? All the while knowing that x1 was about to be usable soon?
Well, the current SQ2 used in the Surface Pro X has 4 Cortex A76 performance cores - not the latest but the A78 isn't that much faster anyway - and 4 Cortex A55 efficiency cores. What needs to happen is for the SQ3 to have 2 Cortex X1 performance cores and 2 Cortex A78 performance cores to go with the 4 Cortex A55s. That would be an absolute minimum.
I would love to see someone get bold and just go with 4 Cortex X1 chips. Even if it has to go in a Chromebox (ChromeOS but in a desktop form factor) or something with a fan.
Also, the reason for the lack of good desktop Qualcomm and MediaTEK CPUs until now is their just being stupid. No one is willing to go beyond the 8 core barrier. (Except Apple, and when they come out with their 12 and 16 core Mx chips next year it is going to make them all look like fools.) Intel and AMD use 6 and even 8 performance cores in their Intel Core -7 and AMD 4800U chips right? What on earth keeps Samsung or MediaTek from doing the same? Keep the 4 efficiency cores but add performance cores until you reach diminishing returns or the laptop gets hot enough to fry an egg. If it works for ARM servers ... oh never mind. Again, they're just being stupid, lazy and risk averse.
Honestly, I'm more curious about what all this means for mid-range chips. When shopping around, I can always find a mid-ranger with good battery life for like $200-300, so that's what I'm keeping my eyes on. But one thing at a time...
Mid range honetly reached good enough years ago. My moto z play has a 625, and still feels perfectly snappy under normal load, and can still play games. The 665 in my moto g power is way faster. The phone feels slower, but that's because of lack of software otpimization.
Yup, I've had a 625 that was good enough (dropped it broke the screen) and now I'm rocking a 710. But there's always room for better connectivity and lower battery drain ;)
Games, I really don't care for. The games I have installed always show up in Files as "apps you haven't used in awhile".
I think that these results are disappointing. So, the 888 will net even be able to match the A13 in single threat performance. You must consider that by the time the Smartphone featuring the 888 will come out, Apple will be somewhat 6 months from releasing an even more powerful A15. I'm not an Apple fan boy, I'm just frustrated that Qualcomm isn't able to deliver the power that the Android community deserves.
A similar story is happening between Intel an AMD, where AMD is starting to crush Intel on every single level. At least, all Windows users are having the opportunity benefiting from it.
Qualcomm’s TOP number is for all CPU/GPU/Neural Engine combined. That’s not in the fine print. Apple gives their number 10 or 16 TOPs for just the Neural Engine processing.
p.s. I'll admit the number for the CPU is a guesstimate on my part, but pretty objective. AFAIK, even the highest end desktop 32 core AMD Zen 2 doesn't cross 1TFLOPS. Regardless, doesn't make a difference to my point.
Qualcomm's smartphone chip crushed the M1 in AI department
I didn't. You suggested that we include TFLOPS along with TOPS in your first reply..right? Anyway, TOPS are the general measure and TFLOPS are more specific.
Since tflops are more specific, wouldnt that mean its able to do more tops than tflops? And from apples page 'with a powerful 8‑core GPU, machine learning accelerators and the Neural Engine, the entire M1 chip is designed to excel at machine learning.'
Uh..I was hoping the logic would come to you. T0PS means trillion operations per seconds.. so, a general measure. TFLOPS means trillion floating operations per second, so more specific part of the general measure.
That's exactly what I am saying..so where is this going then?
TFLOPS are still TOPS. For Snapdragon 888, if we count the TOPS for entirety of the soc, then the GPU part should be around 1.6TOPS( because Adreno 660 ~ 1.6 TFLOPS), and the CPU is much less than a 1 TFLOPS.
That means the AI engine alone should be able to do at least 24 TOPS.
Bottom line - well over 2x that of the M1. Get over it.
What? Dont tell me you think calculating floating points require the same processing power as any other operation. Tops could be any kind of operation, it could be the most simple one. No source, no nothing, 'dont know what else to say' huh, you dont even know much it seems.
Regarding the much hyped snapdragon 888, because of overheating issues and over power consumption issues you're better off just getting the lower line 865+ since it has neither one of those problems. If you do want to get the 888 Wait until qualcomm eliminates those 2 problems. If you have the apple a14 then you're already the winner since it also does not have either problem and beats the 888's benchmarks decisively
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
75 Comments
Back to Article
darkchazz - Friday, December 18, 2020 - link
Got to say the only reason why I didn't upgrade from my SD845 device to an 865 this year was all the reports of issues with the 120hz AMOLED panels i.e. green tint, image retention, black crush.Seems like the wait will be worth it.
Sharma_Ji - Sunday, December 20, 2020 - link
Only batches of all HRR devices have tint issues, and for black crush, what can i say, no nobody except apple knows how to deal with it perfectlyeastcoast_pete - Friday, December 18, 2020 - link
Thanks Andrei! I look forward to your first test and hopefully a deep dive once you were able to put an 888 through your tests.As an aside, my standard gripe with all 2020/21 ARM chips that are not from Apple: why are we still on A55 with its in-order design for the efficiency cores? Apple's efficiency cores seem to be a lot more "efficient" (perf/wh), pun intended. What's up with that, ARM, QC and others?
michael2k - Friday, December 18, 2020 - link
There are three components to measuring efficiency: work/watt, die area, and absolute watt.Power is consumed performing work, so work/watt tells us how much time is necessary powering the efficiency cores.
Die area consumes power even when idle, so a smaller die area will boost efficiency by reducing the amount of powered transistors when no work is done.
You can also measure power usage, without measuring time.
So in that sense the efficiency cores ARM has, the A55, are a balance of the three, Apple is prioritizing perf/watt since they don't care about the silicon costs (die size) and reduce power usage there by using the smaller 5nm transistors.
HardwareDufus - Friday, December 18, 2020 - link
I suspect that later on we will have a refresh of this SoC as S888+, similar to what was done with S865/S865+ .... where clocks are bumped as yields improve. Not needed now, but probably something QualComm is holding back... The S865 to S865+ move was a good way for QualComm to refresh it's product line to get premium dollars without a full redesign or generation jump. It's not hard to imagine a repeat of the strategy with S888/S888+ChrisGX - Sunday, December 20, 2020 - link
Well, there are probably lots of reasons that Qualcomm follows up on flagship SoC releases with a Plus version. If that happens this time the reason will be simple. The 5nm LPE process node used for the SD888 was a bit under par. Generally higher core frequencies (say, 5 - 10% higher) across the board for all CPU cores and thus higher performance should have been a goal within the reach of the designers of the SD888 without running into power consumption or thermal troubles. So, if a better silicon production process from Samsung should become available that doesn't necessitate lots of expensive rework by Qualcomm we may well see a Plus variant. Otherwise, perhaps, we won't. After shoehorning a very powerful GPU and modem in there it isn't clear that Qualcomm any longer has the leeway it needs to just push ahead on the performance front without regard to the power and thermal consequences.https://semiwiki.com/semiconductor-manufacturers/s...
https://www.anandtech.com/show/16271/qualcomm-snap...
ChrisGX - Sunday, December 20, 2020 - link
A few of my words weren't well chosen. I should have described the Samsung 5nm LPE process node as "subpar".admnor - Friday, December 18, 2020 - link
Still a looooong way behind Apple in single-core CPU performance though. Why is it so hard for everyone to catch up to them?toyeboy89 - Friday, December 18, 2020 - link
Apple is spending their billions into their chips and it's showing. Hard to compete with a tech super power with endless pockets.milkywayer - Sunday, December 20, 2020 - link
^ This statement incorrectly implies Qualcomm is poor or doesn't have the resources to try to match Apple's superiority in the field. Qualcomm has be strong-arming manufacturers and making everyone pay a mafia cut for over a decade now and has all the money for this. Just not the will since they're free loading on the patent/royalty cuts and churning out garbage chips since there isn't a competitor in the Android world. Its crazy how apple is crushing every Qualcomm SoC with their mobile CPUs.johnathanblade - Saturday, January 30, 2021 - link
What this statement ignores is that there are many major competitors in the SOC world. All of them lag far behind Apple, but in second is pretty consistently Qualcomm. Qualcomm does peak better, it does sustained better, and it does graphics better than any of its non-Apple competition. Huawei comes close occasionally. Samsung, Mediatech, and Rockchip don't.Apple makes expensive chips that they know they can sell, and they fab farther away from the ARM reference than the other manufacturers.
s.yu - Thursday, December 24, 2020 - link
>a tech super power with endless pocketsYou mean Intel, with world No.1 R&D expenditure? ;)
RobJoy - Tuesday, January 5, 2021 - link
Qualcom is loaded, so money is no issue.I think it is the patents.
Wilco1 - Friday, December 18, 2020 - link
Android SoC vendors aren't willing to spend a lot of area on big cores and huge caches: here we have 4x512KB L2 plus a tiny 4MB L3 vs A14's 12MB L2 and 16MB L3.fishingbait - Friday, December 18, 2020 - link
That has nothing to do with it at all. Android SOC vendors - really just Qualcomm, MediaTek and Samsung unless you want to count Huawei - chose to use more cores instead of bigger ones. While the competition - Apple - was until recently still using 2 cores and no one was making a real effort to make ARM laptops beyond a few cheap Chromebooks, it wasn't a problem. But when Apple went to 4 and then 6 cores and started crushing Android SOCs practically overnight, then there was a problem. MediaTek tried to address it with a 10 core design with 2 fastest, 4 fast and 4 efficiency cores, but it only gave them a 10% performance increase over their octacore designs. So ARM and Samsung went back to the drawing board to create a much better single core design: the X1. But this is a first generation design. There are still issues that they need to work out in order to have this core perform even faster. Also, they need to go from 1+3+4 design to a 2+2+4 design, from 1 Cortex X1 to 2, if they are going to have any chance of rivaling the A14. The problem is that by the time they perfect the design, Apple may well have 4 Firestorm performance cores instead of 2 for their smartphone chips, and on a 3nm process at that. But the biggest problem - coming up with their own big core design - has been solved.Ultimately, there is no "need" to play catchup with Apple smartphone speeds anyway. The vast majority of Android phones sold cost under $400, and the expensive Android phones are so because they offer a ton of features that iPhones won't have for 3-4 years. The real need is to come out with a PC chip for laptops and desktops to compete with the M1. That is the real value of the Cortex X1. We might even see Qualcomm make another try at being a supplier for ARM servers again - they began to in 2015 but gave up and pulled the plug in 2018 - with Samsung potentially joining them.
Wilco1 - Friday, December 18, 2020 - link
No - multithreaded performance is not an issue. That's already within 10% of A14 as the results above show. However single-threaded GB performance is 40% lower. The big issue is that the big cores don't have nearly enough cache, so they don't run as fast as they could. That can be solved by adding more cache, and lots and lots of it. L2 could become 1MB per core like in the Neoverse cores, L3 quadrupled to 16MB and 8-16MB as a system cache. That with an improved X1 core at similar frequency as SD865+ should match A14 both single and multithreaded.I'd say 4 big cores is more than enough (whether it is 1+3 or 2+2). Replacing the 4 little cores with one A78 variant optimized for best perf/Watt would be a good idea too.
Whether it is worth chasing A14 is a good question indeed, but the key problem is willingness to spend lots of area on big cores and big caches (which either means higher cost or having to cut down GPU or AI performance).
dudedud - Sunday, December 20, 2020 - link
"That with an improved X1 core at similar frequency as SD865+ should match A14 both single and multithreaded."A13 single yes. A14 single nop.
ichaya - Monday, December 21, 2020 - link
You can see the diminishing returns in increasing on-die cache with A14 vs M1 L2 cache increase from 8 to 12MB. One core can't access all of the increase, and there are more big cores in the M1, but still a 33% increase in what one core can access (6 vs 8MB?) only gives a 7% increase in single core performance.That 7% might be worth it on a MB Pro, but it's most definitely not worth it for a phone SoC. QC SoCs are smaller than Apple SoCs, the 845 was the first to have a SLC/L3, and the 6/700-series SoCs still don't have it. More on-die cache is always good, it's a matter of price, though I would agree QC could do better.
The A14 also has 4 memory channels, 16K pages, wider cores fed by more OoO execution and larger caches, and the perf/watt is impressive enough for low power, that I would say we should be seeing some of the same things from other ARM designs and even AMD/Intel eventually.
TheinsanegamerN - Saturday, December 19, 2020 - link
"That has nothing to do with it at all"You are completely ignorant. Larger caches absolutely play a part in higher IPC. What do you think contributes to "larger" cores? They dont just throw more transistors into a FPU and call it a day.
RobJoy - Tuesday, January 5, 2021 - link
They don't really.AMD has had 2x the cache Intel had and Intel always had better IPC than them.
Until recently when cores were drastically redesigned.
Frantisek - Sunday, December 20, 2020 - link
If we take some rumors Apple will be focusing on power efficiency with A15 and potentially it can be 4+4 design. They do not need to push sigle thread performance much but not sure how they can fit 2 more cores on die.Also CPU performance in mobile world becoming nearly irrelevant as chips are blazing fast and more focus is now put on AI and graphic. From this point is 888 interesting and I wonder how all will be able to utilize. Let see what it brings to users and if it will leapfrog Apple in this.
In desktop ARM designs I wonder if they catch up in two years somehow or some other vendor.
techconc - Tuesday, December 22, 2020 - link
The problem is, nothing can touch Apple's single core performance. There is nothing stopping any chip maker from topping Apple's multi-core performance. Qualcomm could simply just add more cores.Qualcomm doesn't design their own CPU cores anymore, so it's really up to ARM to decide if that's where they want to go. I thought that was what the X1 was supposed to be for. The X1 seems to be what the A78 should have been.
Agreed that there is no real need to add more cores for a phone chip. There are already thermal limitations to deal with. Qualcomm had done the right thing by focusing on AI/ML and making their GPU more competitive.
tuxRoller - Saturday, December 19, 2020 - link
1. All the mobile vendors, save Apple, only make tiny changes to the arm core design ← this is by far the dominant factor, but there's a reason why the X-1 is their first "big" core & that's....2. Cost. Any of the core designers at the various companies could explain how to design a competitive core to Apple's but the core will be huge... just like Apple's ← for fast non-mobile arm-based cores look to arm's neoverse, Qualcomm's falkor, Calcium's thunderx/2,
adriaaaaan - Wednesday, December 23, 2020 - link
Because selling such chips would be commercially unviable. Apple's chip division has no such restriction and can therefore build bigger chips or get in early on the newest process. Most of apple's success is actually a result of tsmc rather than apple.RSAUser - Thursday, December 24, 2020 - link
One: Qualcomm is using a much worse process node, Samsung 5nm is around 2/3 as dense as TSMC 5nm. This has a huge impact on especially low wattage performance.Secondly, cache. Cache is probably one of the more expensive things you can do to increase performance of a chip, e.g. Architecture change is once off, but cache will be paid linearly, so it's usually one of the cases where cost cuts are seen first if they can get away with it.
Qualcomm doesn't need to beat Apple, they just need to beat the rest of the Android competition.
GC2:CS - Friday, December 18, 2020 - link
Why not just ship reference design devices to reviewers ?Nothing aggainst preventing thespred of viruses, but just posting supposed numbers as “best of the best” in every category seems suspicious as if there was a caveat or two.
Do we at least know if RD was a tablet or smartphone ? Or any info at all ?
fishingbait - Friday, December 18, 2020 - link
Doesn't matter. The first batch of phones containing this new chip will be out in less than month, whether it will be the Samsung Galaxy S21 phones that launch January 14th or a Chinese phone that beats it to market by a week or 2. Either way it won't be much longer before we have independent benchmarks.tuxRoller - Saturday, December 19, 2020 - link
When was the last time qcom provided unreliable perf?They're obviously going to cherry pick the particular benchmarks but their actual results should be replicable. In the past these initial benchmarks tend to be on the lower side.
In this case, there's no reason to be suspicious... HOWEVER, power draw is a completely different matter.
yeeeeman - Friday, December 18, 2020 - link
Another BIG unknown is the process performance, given that this will be made on samsung 5nm process and not on TSMC.techconc - Tuesday, December 22, 2020 - link
Agreed. We already know transistor density is not as good on the Samsung process. Probably similar for performance.Small Bison - Friday, December 18, 2020 - link
+23.5% performance is a nice improvement, but I can’t help but be a little disappointed when Arm was claiming +30% IPC *and* frequency increases. Here’s hoping that Qualcomm is just being conservative with peak performance for the smartphone chips to save power, and that the 8CX equivalent will be able to run at higher frequencies!arayoflight - Friday, December 18, 2020 - link
Qualcomm went with half the cache compared to ARM's configuration to save costs. Also they are doing 0 frequency increases.Qualcomm has nothing to gain from a faster chip. They have zero competition and it's better for margins to make a more cost efficient chip.
Small Bison - Friday, December 18, 2020 - link
Half the L3 cache for the whole cluster, but Qualcomm’s using the full 1 MB L2 cache for the X1 core.techconc - Tuesday, December 22, 2020 - link
No, I believe this was Anandtech's misinterpretation. I remember reading ARM's information on the X1 and disagreeing with Anandtech's assessment. ARM's claim of 30% performance improvement for X1 assumes process node bump AND clock speed of 3Ghz. You can see this in their slides, it's in the fine print. ARM didn't mislead anyone, in this case Anandtech's analysis did.toyeboy89 - Friday, December 18, 2020 - link
Okay I don't like the route they're going with a single- X1 core, Apple has 2 high performance cores in parallel isn't this way more powerful?fierysnowman - Friday, December 18, 2020 - link
Apple has two very high performance cores, Qualcomm has one high performance and three medium-high performance cores.Apple dedicates around twice as much die area as Qualcomm to CPU cores and private cache, yet Qualcomm gets 70-90% of their performance at similar efficiency. That’s very quite decent. Not very exciting, but pragmatic.
TheinsanegamerN - Saturday, December 19, 2020 - link
Apple also controles the entire stack. We cant run android on an iphone to test this, but I'd wager part of the A core's performance is due to top to bottom optimization that just doesnt happen on android devices.fierysnowman - Monday, December 21, 2020 - link
I have my doubts the optimization are anything only iOS would benefit from. More so they optimize against specific use cases common on an iPhone or iPad, such as web browser JavaScript, which the M1 is exceedingly proficient at and other browsers also benefit from.Hyper72 - Friday, January 1, 2021 - link
People always say that but I disagree. Apple really makes great general purpose CPUs. Many benchmarks and applications compiled for standard ARM64 flies on them and even ARM Windows shows fantastic performance running inside a VM. They have lots of general purpose features: Like that 8-wide decoding with a huge ROB and large+fast caches. The only other CPU with similar architecture is the big iron POWER10 (as mentioned by Anandtech).fishingbait - Friday, December 18, 2020 - link
The reason is that the X1 core is brand new. It was a joint design project by ARM Holdings and Samsung to catch up to Apple Ax and it was just completed last year. Qualcomm, Samsung and the rest went from a big.Little 4+4 architecture to a 1+3+4 architecture awhile back because they knew it was on the way. Getting 2 X1 cores to match the Apple A14 will be next. I would be shocked if Samsung isn't already working on it because that is basically their only path to competing with Qualcomm since they have given up on custom cores. After that it will be getting to 4 Cortex X1 chips like the M1. Hopefully they will achieve it by the time they get to a 3nm process in 2022.iphonebestgamephone - Saturday, December 19, 2020 - link
So you're telling me samsung was involved in the x1's development, but still used 2+2+4 config and was also working on thier custom m cores? All the while knowing that x1 was about to be usable soon?Fulljack - Wednesday, December 23, 2020 - link
yes, because until X1 came out then Samsung need to use their own prime, big cores. that is the M-series.iphonebestgamephone - Thursday, December 24, 2020 - link
And is there any article mentioning the 'joint development' of x1 by samsung and arm?iphonebestgamephone - Thursday, December 24, 2020 - link
So arm used samsung to beta test or something?Vitor - Friday, December 18, 2020 - link
Impressive gains, feel like it could make a decent desktop cpu.fishingbait - Friday, December 18, 2020 - link
Well, the current SQ2 used in the Surface Pro X has 4 Cortex A76 performance cores - not the latest but the A78 isn't that much faster anyway - and 4 Cortex A55 efficiency cores. What needs to happen is for the SQ3 to have 2 Cortex X1 performance cores and 2 Cortex A78 performance cores to go with the 4 Cortex A55s. That would be an absolute minimum.I would love to see someone get bold and just go with 4 Cortex X1 chips. Even if it has to go in a Chromebox (ChromeOS but in a desktop form factor) or something with a fan.
Also, the reason for the lack of good desktop Qualcomm and MediaTEK CPUs until now is their just being stupid. No one is willing to go beyond the 8 core barrier. (Except Apple, and when they come out with their 12 and 16 core Mx chips next year it is going to make them all look like fools.) Intel and AMD use 6 and even 8 performance cores in their Intel Core -7 and AMD 4800U chips right? What on earth keeps Samsung or MediaTek from doing the same? Keep the 4 efficiency cores but add performance cores until you reach diminishing returns or the laptop gets hot enough to fry an egg. If it works for ARM servers ... oh never mind. Again, they're just being stupid, lazy and risk averse.
michael2k - Friday, December 18, 2020 - link
If they are using ARM's DynamIQ, they are forced to use a cluster of 8 CPUs:https://www.anandtech.com/show/11441/dynamiq-and-a...
If they wanted more then they would need to use a pair of DynamIQ clusters, which unlocks 16 CPUs.
You're asking why they haven't done so? It's because it costs money and if no one is paying Qualcomm to develop the product, they won't do so.
iphonebestgamephone - Saturday, December 19, 2020 - link
Yeah a78 is just what, 35% faster than a76. Yeah, not much. Next.Zoolook - Monday, December 21, 2020 - link
Most of Mediateks Helios X series had 10 cores, although only two performance cores.AdrianBc - Tuesday, December 22, 2020 - link
The ARM cores are designed to reach their maximum performance at around 2 W (Cortex-A78) or at around 3 W (Cortex-X1).A smartphone might provide up to 10 W for short bursts, but it would not be able to sustain such a high power.
Therefore a quadruple Cortex-X1 would need about 12 W. This power level might be acceptable for a large tablet, but certainly not for a smartphone.
The largest configuration that would make sense for a smartphone is 2 Cortex-X1 + 2 Cortex-A78 cores.
An 8-core Cortex-X1 would be great for a 15 W or 20 W laptop, but that could never be used in a smartphone or small tablet.
bug77 - Friday, December 18, 2020 - link
Honestly, I'm more curious about what all this means for mid-range chips. When shopping around, I can always find a mid-ranger with good battery life for like $200-300, so that's what I'm keeping my eyes on.But one thing at a time...
TheinsanegamerN - Saturday, December 19, 2020 - link
Mid range honetly reached good enough years ago. My moto z play has a 625, and still feels perfectly snappy under normal load, and can still play games. The 665 in my moto g power is way faster. The phone feels slower, but that's because of lack of software otpimization.iphonebestgamephone - Saturday, December 19, 2020 - link
Yeah its still good for 30 fps and below high end games.bug77 - Saturday, December 19, 2020 - link
Yup, I've had a 625 that was good enough (dropped it broke the screen) and now I'm rocking a 710. But there's always room for better connectivity and lower battery drain ;)Games, I really don't care for. The games I have installed always show up in Files as "apps you haven't used in awhile".
sadick - Monday, December 21, 2020 - link
I think that these results are disappointing. So, the 888 will net even be able to match the A13 in single threat performance. You must consider that by the time the Smartphone featuring the 888 will come out, Apple will be somewhat 6 months from releasing an even more powerful A15. I'm not an Apple fan boy, I'm just frustrated that Qualcomm isn't able to deliver the power that the Android community deserves.A similar story is happening between Intel an AMD, where AMD is starting to crush Intel on every single level. At least, all Windows users are having the opportunity benefiting from it.
techconc - Tuesday, December 22, 2020 - link
"So, the 888 will net even be able to match the A13 in single threat performance."It basically matches the A12 on single thread performance. Also consider, that's with the process node advantage of the SD888.
darkich - Wednesday, December 23, 2020 - link
One thing the media will never mention - this chip has 160% more AI performance than than Apple's M1!!26 TOP's vs. 10 TOP's
iphonebestgamephone - Thursday, December 24, 2020 - link
Isnt the 26 of qualcomm from the whole thing,cpu, gpu , dsp etc, while m1 is just the neural engine?Irish910 - Thursday, December 24, 2020 - link
Qualcomm’s TOP number is for all CPU/GPU/Neural Engine combined. That’s not in the fine print. Apple gives their number 10 or 16 TOPs for just the Neural Engine processing.darkich - Friday, December 25, 2020 - link
..even if that's the case, the M1 doesn't come nowhere near..add about 2.6TFLOPS for the gpu and less than 500GFLOPS for the CPUiphonebestgamephone - Friday, December 25, 2020 - link
And where are those numbers from?darkich - Saturday, December 26, 2020 - link
Straight from Applehttps://www.apple.com/hr/newsroom/2020/11/apple-un...
darkich - Saturday, December 26, 2020 - link
p.s. I'll admit the number for the CPU is a guesstimate on my part, but pretty objective. AFAIK, even the highest end desktop 32 core AMD Zen 2 doesn't cross 1TFLOPS.Regardless, doesn't make a difference to my point.
Qualcomm's smartphone chip crushed the M1 in AI department
iphonebestgamephone - Saturday, December 26, 2020 - link
And how did you convert those tflops into tops?darkich - Sunday, December 27, 2020 - link
I didn't. You suggested that we include TFLOPS along with TOPS in your first reply..right?Anyway, TOPS are the general measure and TFLOPS are more specific.
iphonebestgamephone - Sunday, December 27, 2020 - link
Since tflops are more specific, wouldnt that mean its able to do more tops than tflops? And from apples page 'with a powerful 8‑core GPU, machine learning accelerators and the Neural Engine, the entire M1 chip is designed to excel at machine learning.'darkich - Sunday, December 27, 2020 - link
Uh..I was hoping the logic would come to you.T0PS means trillion operations per seconds.. so, a general measure. TFLOPS means trillion floating operations per second, so more specific part of the general measure.
iphonebestgamephone - Sunday, December 27, 2020 - link
What logic? I know what the full form is. Shouldnt the entire soc be able to do a higher number of general tasks (tops) than specific ones (tflops)?iphonebestgamephone - Sunday, December 27, 2020 - link
Wouldnt the specific task be just a part of the general tasks?darkich - Monday, December 28, 2020 - link
That's exactly what I am saying..so where is this going then?TFLOPS are still TOPS.
For Snapdragon 888, if we count the TOPS for entirety of the soc, then the GPU part should be around 1.6TOPS( because Adreno 660 ~ 1.6 TFLOPS), and the CPU is much less than a 1 TFLOPS.
That means the AI engine alone should be able to do at least 24 TOPS.
Bottom line - well over 2x that of the M1.
Get over it.
iphonebestgamephone - Monday, December 28, 2020 - link
And what is your source for 1 top = 1 tflop? Show me that and i'll 'get over it'.iphonebestgamephone - Monday, December 28, 2020 - link
Think about it, by your logic, the npu of m1 would be 11 tflops. And the whole qualcomm 888 would be 26 tflops.darkich - Tuesday, December 29, 2020 - link
Wow..I am trying to explain it to you the whole time. And you're telling me to think. I won't repeat myself, don't know what else to say.iphonebestgamephone - Tuesday, December 29, 2020 - link
What? Dont tell me you think calculating floating points require the same processing power as any other operation. Tops could be any kind of operation, it could be the most simple one. No source, no nothing, 'dont know what else to say' huh, you dont even know much it seems.Dneonu60 - Wednesday, January 20, 2021 - link
Regarding the much hyped snapdragon 888, because of overheating issues and over power consumption issues you're better off just getting the lower line 865+ since it has neither one of those problems. If you do want to get the 888Wait until qualcomm eliminates those 2 problems. If you have the apple a14 then you're already the winner since it also does not have either problem and beats the 888's benchmarks decisively