And as every year here is Gary's ARM new Cortex breakdown! I have to say that I am mildly interested to not all that impressed by the new ARM cores, I have seen a video by Geekerswan a few months ago where they did the power curves via SPEC for many ARM cores and since ARM V9 A7X0 cores have not been amazing in power efficiency terms and I hoped that this time around they managed to slash the power consumption when running the cores at their maximum to bring it down to the A78 levels. Let's see how manufacturers implement these as I suspect they will be running them at full blast to win benchmarks...
I'm excited what the SoC Qualcomm is going to produce based on this new cores design this year🎉 Finally they will get rid of 32 bits and no longer a710 cores inside new generation of chips. And no longer 4 cluster chip architecture, which was basically created to support 32 but apps to the detriment of efficiency.
5:53 Actually, the Cortex-X4 looks to have *removed* the Mop (micro-op) cache, so this uArch has a somewhat bigger shift than usual. Maybe this section needs a correction.
The mechanisms are still there but the way it works now there is no need for a separate mop cache. I specifically asked about that during the tech briefing.
@@GaryExplains Ah, very interesting. No separate Mop$, but snuck in somewhere else? Thank you for asking that, as the manual alludes that it was removed and not quite clear what replaced it (besides generally widening the decode / front-end)
Basically the instruction cache now holds MOPs and normal instructions, and there is no difference in dispatch width or pipeline depth between normal instructions and micro ops.
Funny how ARM started really putting effort in their cores and especially the X cores once it learned that Qualcomm bought Nuvia. Otherwise we'd still be 3 generations behind the Apple chips in terms of single threaded performance.
@@GaryExplains Yeah but the improvements were minimal. And the efficiency was getting worse and worse. Since 2021-2022 the X core has been getting 10-15 faster while staying at basically around the same power draw.
2021 to 2022 is the first revision of the X1 to the X2 and it made double digit performance gains. CPUs take 2 years to design. Even if you say 1 year (from the X1 to X2) then the X2 was already designed and sent to partners before the Nuvia sale was announced. You are trying to see a connection where there isn't one.
@@GaryExplains Again I'm talking about efficiency. The X2 was way worse than the X1 in power efficiency even though it offered double digit improvement. The X3 and X4 have been the only worthy successors to their predecessors offering double digit improvements with basically no increased power consumption.
Now I am confused... In your first comment you say, "Otherwise we'd still be 3 generations behind the Apple chips in terms of single threaded performance" but now you are saying, "I'm talking about efficiency" 🤷♂️
I wonder if a laptop chip set up with the full 32MB L3 and a 12+0+2 layout would perform. 12 threads of X4 would be rather power hungry, but the performance would be stellar. The 2 little cores stick around for standby tasks and maybe keeping certain OS functions going when the big cores are loaded.
I think we need a proper laptop chip from these updates to get a start in finally switching to a decent Windows on ARM laptop. A couple of X4 might do the trick, while hopefully not being to power hungry. Not expecting M1 performance just yet but close would be nice!
I think 6+8+0 or 8+6+0 makes with max l2 and l3 makes more sense in a laptop. A720 cores are super flexible and they consume the same power (on lower clocks) as A520 as evident by the graph however they can also perform a lot more on higher clocks and they can also sustain those clocks and be thermal efficient as shown in slides. So it makes a lot more sense to include those in a laptop soc along with x4 cores. All x4 will be not only be power hungry but also hard to cool.
@@stancex404 8+6 makes a lot of sense, but I still think favoring the X4s has some potential. I could see 8+4+2 being used as well, to have a couple of very low power cores for standby. Maybe not even A520, but something even smaller like an A320 core of just the super-slimmed-down A520 shown here. Now that I mention it, I wonder how an all A520 chip with the differentiation being in cache and clocks would behave. It would be similar to Zen4 vs 4C, but tiny and ARM. I could see a 4+4 phone SoC doing that.
@@MrFilip121 I would question that Filip - just one brand of SBC (The Raspberry Pi) is out-selling the x86 PC. The sales also support the claim that the higher-performance boards sell at least as well as the low-power ones. I know the SBC sales do not earn a lot per board, but that is still a serious incentive.
I think they should reduce A520 cores to just 2 because it isn't meant for multi threaded performance, just for background and light tasks. The most performance range is covered by A720 cores. Its quite flexible as seen from the graph. They should try to include more of these as they have better sustained performance and thermal efficiency. 1+5+3 or 2+4+2 makes the most sense with larger l2 and l3 sizes.
I definitely think there should be a variant with A720 at different frequencies, too! After a certain point, the gains in performance aren't as good for the power consumed.
I typically upgrade every year. This year, the power and efficiency gains were very substantial. Based on your assessment, I can't wait to see what next year brings!
@@RajeshPachaikani Nah. They will clock these super aggressively making using of every single bit of performance and leave no room for efficiency because thats how they win benchmarks.
I just replaced my four year old Galaxy note 20 with a Galaxy S23 Ultra. Synthetic benchmarks are twice as fast and all apps feel faster and smoother. you definitely noticed the 3.3 GHz X core. And the battery life is much longer. What a great time to be alive!
10:10 по сути, тут хорошо видно, что чистый выигрыш в производительности по архитектуре равен 10%, все, что выше, достигается большей частотой ядер, большим размером кэш памяти и так далее.
Professor Gary, I don't know if you'll see this comment, I know you don't do Speedtest G anymore, because it had a low audience and it was extremely laborious and costly to carry out the various tests. But here's my request from a fan of yours, when you feel like being able to do them again, but with a much lower intensity, that is, test only one device of each main SoC, that is, the SD Gen 2, A16 and D9200. Regarding the tests, a CPU peak test can be done, but with the introduction of energy consumption analysis and the average performance, with the GPU the same as the CPU. It can also introduce ray-tracing tests and more complex ISP and AI tests to truly test the capabilities of current and future SoCs.
I hear what you are saying but now you want me to develop ray tracing tests, AI tests, and energy efficiency tests. That is a huge investment of my time. Having said that I do want to do more efficiency tests like I did for microcontrollers.
@@GaryExplains Thank you very much for answering, but if this is really bad and hard to do then don't need to. Your health and well being are more important! But if one day this test comes back, I would be very happy!
Currently A16 wins by around 25-30% in single thread compared to 8G2. So a 15% uplift will make it closer to A16. A16 will still probably be around 10-12% faster. A17 will be a lot faster since it will be on 3nm while 8G3 will still be on 4nm and 3nm is a full die shrink so expect around 20-30% performance uplift compared to A16. That means A17 will be 35-45% faster than 8G3 in single thread.
Will any of the other ARM licensees come up with a SoC comparable or better than Apple's M2 or A16? The RISC side of computing is getting really interesting for us consumers. From SBC to tablet/mobile, laptop and desktop space.
@@NexuJin Actually I have several videos about it!!! 😜 th-cam.com/video/daGm4JmrmZ4/w-d-xo.html and th-cam.com/video/Fi0AKrprWxg/w-d-xo.html and th-cam.com/users/shortsaOl_g4vluFs?feature=share and th-cam.com/video/PA5KEjO6NqE/w-d-xo.html
Cortex a7 came in 2014 with Philips designed NEON technology catering to media, which is text, sound, and imagery. And images for gaming are simple arithmatic for ray-tracing.. But they are no good for number crunching required for general computing. Isn't that so Gary?. Your comments please.
Funny, I thought the second fastest supercomputer in the world used Arm processors. What proof or source are you using to say that Arm processors are not good for number crunching? 2014 is a long long time ago. The A7 was a 32-bit Arvmv7 processor. Things have moved a lot since then: SVE, SVE2, etc. Plus Armv8, Arm9, Neoverse, and more.
@GaryExplains what's so funny, my friend?. I am only telling you what I read in Prof. Tannenbaun's (who wrote the MINIX os which is the forerunner to all powerful Linux) book in its latest edition. I am not a computer expert but learn a lot from videos such as yours. I do some android app development and was baffled by the inability to convert them to Apple with M1 and M2, which run on ARM architecture. They seem unable to handle the type of mathematical function.
Sorry, saying "Funny, I thought..." is a English colloquialism, I shouldn't have used it. It means that since the second fastest supercomputer in the world uses Arm processors then your statement that they are no good at number crunching is obviously not true. What where you trying to covert to the Apple with M1? An Android program? What do you mean by "They seem unable to handle the type of mathematical function"?
@GaryExplains my earlier reply seems not accepted. Anyway, what I was trying to do was converting one app I have already uploaded (but not marketed yet) and another for structural design and drafting to be converted to iPhone apps as most engineers use Apple these days. But I found Apple silicon no longer support java. I like java as it has the lambda function capability(which you should consider for GaryBasic). I suppose this is due to a problem with ARM architecture. Tried to use Flutter but found that also do not support what want to do.
I see. OK. So Android apps are developed using Java or Kotlin and iPhone apps are developed using Obj-C or Swift. I don't think there was ever a time when iPhone apps could be built using Java. This is one of the key challenges in developing for mobile, i.e. cross-platform compatibility. Flutter is cross platform, so that was a good idea. You might want to look into Unity. All this has nothing to do with Arm or the CPU, it is just how Android / iOS are developed.
On average, how much overall power does the CPU use in a phone? I would have thought that the screen is the main thing, then you still have loads of other hardware suckling from the battery, I have no idea where the CPU fits in here. So, let's say phone X lasts 24 hours under normal use. Then only the CPU is upgrade and this new CPU is 20% more power efficient. As a rough guess, how much would this generally affect the lifespan of a phone under the same conditions of normal personal usage? +20% on 24hrs is 4.8hrs but we know it's not going to be close to that, but is it negligible to the overall energy usage of a phone or ~30mins, any other guesses?
You're not going to see that much improvement because the quoted number for power efficiency is under conditions that show it the best, typically at least moderate to full load. Most phones have bursty workloads so the actual battery savings will be much less. The improvement is worthwhile! It's good to strive for greatness. However, this is definitely not life-changing efficiency improvement. I would say not more than a few percent difference.
From what I remember Qualcomm specs something like a 10W TDP for their chips. A 15% reduction gets you down to 8.5W under full load. We'll likely see pretty much the same battery life though, as manufacturers will see this efficiency gain as a way to juice out more performance.
That number is kind of meaningless now it just means it's more compact not that the transistors at one nanometer. They still got quite a few shenanigans for them to burn through still like vertical gaafet, 3D stack transistors, nanotube gaafet, underside power wires and after that exotic transistors materials like gallium nitride, silicon carbide, vacuum channels, graphene, it's not gonna be cheap though. Though this video mainly focused on architecture and they are starting to learn how to improve an architecture without adding more transistors.
I look forward to seeing what Apple does with these new core designs as reference for their M series of chips on their macs and such. I would also love to see what Qualcomm will be doing and I’m crossing my fingers that something can make it’s way to the laptop space in mind for windows and Linux. Obviously the phone and tablet space with be the primary use case for consumers but I don’t really feel that much of a difference between generations when it comes to mobile use
@@GaryExplains holy cow, thanks for responding to my comment. I knew Apple designed their own stuff but i didn’t realize they didn’t rely on Arm’s specs at all. Thanks for clarifying this for me
IIRC it's similar to Alder Lake at the same frequency. I don't have any X4 silicon to test, but given how X3 vs Tiger Lake plays out I wouldn't be surprised.
So what's your estimate geek bench single core score To my gues it will be bw 1650-1720 in geekbench whts yours and to me it's so sad that it won't even match the apple a16 which wa basically a15 I hope they are able to atleast match the efficiency
@@GaryExplains arm is very efficient, we have yet to see am arm soc at m1 or m2 proficiency level. The plan is to run Linux on m2 ultra as bare metal as possible. Since that's no yet possible. We need an OEM arm chip that comes to as close to m1 or m2 in performance.
I think the gap is slowly closing not because ARM is improving faster but because Apple has not improved as much YoY. It's harder to improve when you are pushing the boundaries.
@@Henfredemars Dunno mate , I suspect its more to do profit , apple has been a mile ahead for years in smartphone performance until recently , and if you are ahead , why spend more coin , i`d say it is very much like features and accessories that each new phone gets each year , clearly these phone makers could throw in many more features and add even more upgrades and value if they wanted , but a smarter business game is to just add enough of a carrot and some marketing to get the masses buying new phones , while allowing a few carrots in the bag for next year and the year after , i`m pretty sure these big companies play the long game with their strategy's and marketing , its all about the money ;) .
@@mikldude9376 apple tries to do everything in house but when companies that just do 1 thing get good at it, it is hard to compete. qualcomm, mediatek, arm is just soc develepment. that is all they do
For sure, the SD8G2 is ARMv9, whilst the A16 is ARMv8.6. So we should see better ARM architecture in the A17. Then you factor in the new nodes, N3/N3E, and you should get a big generational leap.
Let's hope they take proper advantage of this I'm sick of $2000 phones and they're spending like 100 bucks on silicon. Maybe they can do a smart phone with 2 + 4 + 6 with full cashe.
Wish we can get a chip that have a similar performance to Apple M-series chip. 1) A powerful ARM based Windows Laptop 2) A powerful Linux based Laptop or Handheld gaming device 3) Next gen Nintendo Switch with maybe 2TF to 3TF GPU performance 4) Powerful VR headset 5) Ipad level Android-base Tablet.
That is reported by ARM not by testing of real handsets, and they don't typically provide detailed reports as to how they make that estimate. It's a claim.
Those new CPUs seems promising, but a bit unrelated... my "dream" is to use an extremely low power-demanding PC able to do daily browsing tasks reasonably well below the 1W mark. (Something like 1080p, steady framerates.)
1 watt you said? You have no idea what you are saying. Any decent BASIC cpu consumes at least 20-35 watts while busy, TODAY! Yes, there are 10 watt ARM CPUs but you would not want them for basic Destop duties!
If I remember correctly, Apple M1 can handle word processing and watching online videos (if the video can be hardware decoded by SOC) within 1W (only considering SOC wattage), perhaps it takes a couple years of fab for it to be possible, but considering the increasing processing power requirements of web browsing, I seriously doubt that
Phone CPU 's around 5 watts yeah and laptop 15 to 30 watts So what you're describing sounds like a smartwatch processor or IOT embedded those are pretty slow.
I don't think the 15% increase even reaches apples A15 in single and multicore. Pathetic. Let alone when the A17 comes out And with the power draw being less, is it independent of the upcoming 3nm chips as this will also drastically reduce power draw 🤔
N4P and 15% improvement is gonna be A16 levels of single core. Which is excellent. Android used to be 3 years behind in single core. Now they're 1. Eventually at this right will be right there.
@@Bibbatron A16 levels isn't good. This is because the A16 is just an overclock of the A15 meaning that it's not that Qualcomm has become better but apple had a stop gap year that's why they are closing in
@@ifrankenstein7455 Well still they're only 1 year behind. If Apple fumbled last year I don't really care. At the end I care about the overall competitiveness.
It almost never scales linearly. Usually a drastically larger cache also means worse latency in terms of cycles. For example when Intel moved from the c2q to i7 family architecture, l2 cache size shrank all the way down to just 256K but latency improved so much it was a big net improvement.
You need to consider the market. All of Apple's devices will use their cores, so they can afford to make expensive chips not very area efficient. ARM/SoC vendor has to make a profit at that stage and their market of OEMs is not willing to pay that much.
@@Henfredemars well, apple is not using their own cores as far as I know, they are using ARMv8.5-A Everest and Sawtooth cores, created by ARM in the latest A16 Bionic. Doesn't that mean anyone can use these type of cores? It's enough to buy a licence, isn't?
Qualcomm has never been able to maximise the potential of ARM, but they have made good strides lately, having sacked Samsung Fabs as their node of choice. But it could be argued that Apple's hardware and software integration is better implemented.
So can we all recognize that ARM went to producing chips themselves. Aka now the subs like Apple, phones, etc. Now buy actual ARM chips. Making dev 100 times easier!!!
Qualcomm went against ARM in the Nvidia purchase. There has to be a lot of hard feelings about that and in 2024 when Qualcomm starts making their Nuvia chips, as the lawsuit shows. Arm wants Qualcomm to destroy their last 3 years of chip design. Going to get very ugly.
Qualcomm didn't go against Arm and there aren't hard feelings. The problem is about licensing. I have several videos about the whole saga here on this channel. Arm doesn't have any hard feelings against Apple, or other partners who have architectural licenses. In fact Qualcomm used to design its own Arm based CPUs as well, that is why Arm grants architectural licenses, it is part of its business plan.
I get so annoyed at the inconsistency of naming things. I see it as “ARM” and “Arm” and “arm.” I’m fine with all caps or all lower-case (as a stylistic choice), but since it’s an acronym, I hate seeing “Arm.” Intel does the same thing referring to their new GPUs as “ARC” and “Arc.” About as annoying as acronyms with too many letters or start using characters in the middles of words just to make a preferred name work. e.g. “TBT” for “throwback Thursday” when it should be “TT.” Sorry, I’m in a mood. :-)
@@GaryExplains they say companies can choose to configure the A720 with minimum performance and area to reduce costs, while consumers will not be able to tell which phones have small A720 and which ones have big A720.
The transparency in hardware has been lost years, go anyway. When enterprises like Crapple ignore all specs while integrating components, making the components run outside their specs and have them throttle near permanently for example ..
@@wiggattelo I think with 2 cortex-X cores, then the A720 performance might not matter too much, so I'm looking forwards to what Google will be doing for their G4
Cpu performance isn't that relevant these days. It is running a phone but not a super computer. Ai performance is more relevant to run matrix workload with big memory requirement. That means tpu is the key.. Speech recognition on a phone is still really poor
Perhaps CPU performance is not as critical, but higher performing CPUs we can often run at lower power to achieve the same amount of work. We care greatly about battery performance in mobile devices.
@@DigitalJedi Especially smartphones. Average at anything, good at nothing. No significant development since ten years, but loss of user swappable batteries, audio jacks and extension slots (SD). Servers always where boring, because they are nothing but workhourses. ngreedya's only non boring property is them being frustrating. Microcontrollers have a thing for them, but the more isteresting ones don't use ARM Well, SOHO routers using ARM are a good thing.
@@elmariachi5133 You are one of the usual suspects always complaining of no SD, removable battery or jack! Get over it, tech moves on, and they are dead and buried.
And as every year here is Gary's ARM new Cortex breakdown!
I have to say that I am mildly interested to not all that impressed by the new ARM cores, I have seen a video by Geekerswan a few months ago where they did the power curves via SPEC for many ARM cores and since ARM V9 A7X0 cores have not been amazing in power efficiency terms and I hoped that this time around they managed to slash the power consumption when running the cores at their maximum to bring it down to the A78 levels.
Let's see how manufacturers implement these as I suspect they will be running them at full blast to win benchmarks...
Thank you professor, I always love your tech videos, so informative, i wish speed test g to come back, oh the memories
I'm excited what the SoC Qualcomm is going to produce based on this new cores design this year🎉 Finally they will get rid of 32 bits and no longer a710 cores inside new generation of chips. And no longer 4 cluster chip architecture, which was basically created to support 32 but apps to the detriment of efficiency.
I am much more interested in the chips Qualcomm will produce a year later based on Nuvia technology.
5:53 Actually, the Cortex-X4 looks to have *removed* the Mop (micro-op) cache, so this uArch has a somewhat bigger shift than usual. Maybe this section needs a correction.
I call it the micro-op cache, but of course, Arm calls it the macro-op cache: "L0 Macro-OP (MOP) cache". This is no longer present on the Cortex-X4.
The mechanisms are still there but the way it works now there is no need for a separate mop cache. I specifically asked about that during the tech briefing.
@@GaryExplains Ah, very interesting. No separate Mop$, but snuck in somewhere else?
Thank you for asking that, as the manual alludes that it was removed and not quite clear what replaced it (besides generally widening the decode / front-end)
Basically the instruction cache now holds MOPs and normal instructions, and there is no difference in dispatch width or pipeline depth between normal instructions and micro ops.
Whenever I watch these videos, I can feel my phone getting slower!
Funny how ARM started really putting effort in their cores and especially the X cores once it learned that Qualcomm bought Nuvia. Otherwise we'd still be 3 generations behind the Apple chips in terms of single threaded performance.
The Cortex-X program started long before Qualcomm bought Nuvia. In fact Qualcomm was the principal partner for the Cortex-X program.
@@GaryExplains Yeah but the improvements were minimal. And the efficiency was getting worse and worse. Since 2021-2022 the X core has been getting 10-15 faster while staying at basically around the same power draw.
2021 to 2022 is the first revision of the X1 to the X2 and it made double digit performance gains. CPUs take 2 years to design. Even if you say 1 year (from the X1 to X2) then the X2 was already designed and sent to partners before the Nuvia sale was announced. You are trying to see a connection where there isn't one.
@@GaryExplains Again I'm talking about efficiency. The X2 was way worse than the X1 in power efficiency even though it offered double digit improvement. The X3 and X4 have been the only worthy successors to their predecessors offering double digit improvements with basically no increased power consumption.
Now I am confused... In your first comment you say, "Otherwise we'd still be 3 generations behind the Apple chips in terms of single threaded performance" but now you are saying, "I'm talking about efficiency" 🤷♂️
I wonder if a laptop chip set up with the full 32MB L3 and a 12+0+2 layout would perform. 12 threads of X4 would be rather power hungry, but the performance would be stellar. The 2 little cores stick around for standby tasks and maybe keeping certain OS functions going when the big cores are loaded.
I think we need a proper laptop chip from these updates to get a start in finally switching to a decent Windows on ARM laptop. A couple of X4 might do the trick, while hopefully not being to power hungry. Not expecting M1 performance just yet but close would be nice!
I think 6+8+0 or 8+6+0 makes with max l2 and l3 makes more sense in a laptop. A720 cores are super flexible and they consume the same power (on lower clocks) as A520 as evident by the graph however they can also perform a lot more on higher clocks and they can also sustain those clocks and be thermal efficient as shown in slides. So it makes a lot more sense to include those in a laptop soc along with x4 cores. All x4 will be not only be power hungry but also hard to cool.
@@stancex404 8+6 makes a lot of sense, but I still think favoring the X4s has some potential. I could see 8+4+2 being used as well, to have a couple of very low power cores for standby. Maybe not even A520, but something even smaller like an A320 core of just the super-slimmed-down A520 shown here.
Now that I mention it, I wonder how an all A520 chip with the differentiation being in cache and clocks would behave. It would be similar to Zen4 vs 4C, but tiny and ARM. I could see a 4+4 phone SoC doing that.
I am far more interested to see a single board computer based on the X4 than a phone, or tablet.
Did any Cortex-X CPU make its way into a SBC?
I don't think there is enough demand to financially justify it, maybe some windows laptop
@@MrFilip121 I would question that Filip - just one brand of SBC (The Raspberry Pi) is out-selling the x86 PC. The sales also support the claim that the higher-performance boards sell at least as well as the low-power ones.
I know the SBC sales do not earn a lot per board, but that is still a serious incentive.
@@GaryExplains Sorry, but I do not know Gary.
Nice overview
Great presentation.
Any idea what's going on with the microcontroller cores, are they keeping the current m cores the same?
I think they should reduce A520 cores to just 2 because it isn't meant for multi threaded performance, just for background and light tasks. The most performance range is covered by A720 cores. Its quite flexible as seen from the graph. They should try to include more of these as they have better sustained performance and thermal efficiency. 1+5+3 or 2+4+2 makes the most sense with larger l2 and l3 sizes.
I definitely think there should be a variant with A720 at different frequencies, too! After a certain point, the gains in performance aren't as good for the power consumed.
With a 20% energy saving will they reduce the battery size by 20% to make them lighter, thinner, or smaller?
I hope not.
5:50 But the macro-op cache isn't there, they removed it?
It has been merged with the I cache.
@@GaryExplains Oh I see, thank you for explaining that! I wondered why they'd remove such a beneficial feature but now it makes more sense!
I typically upgrade every year. This year, the power and efficiency gains were very substantial. Based on your assessment, I can't wait to see what next year brings!
Most likely manufacturers will reduce the battery size to offset any gains achieved by efficient CPU.
I upgrade every 5 years
@@toby9999 I do every 3-4 years. Last time in June 2020. Plan to upgrade next year to this platform with wif 7.
@@RajeshPachaikani Nah. They will clock these super aggressively making using of every single bit of performance and leave no room for efficiency because thats how they win benchmarks.
Think of all that E waste yearly upgrades generate. Your corporate masters approve.
I just replaced my four year old Galaxy note 20 with a Galaxy S23 Ultra. Synthetic benchmarks are twice as fast and all apps feel faster and smoother. you definitely noticed the 3.3 GHz X core. And the battery life is much longer. What a great time to be alive!
Oh yes, indeed Cortex X3 is a monster of performance.
More performance is always nice , better battery too , oh and to have had one of these in my galaxy note 4 , 9 years ago :) .
Existence of Cache L2 & L3 in one processor are ultimately crucial.
10:10 по сути, тут хорошо видно, что чистый выигрыш в производительности по архитектуре равен 10%, все, что выше, достигается большей частотой ядер, большим размером кэш памяти и так далее.
This means Speed Test G's next update will be dope.
very exciting!
Professor Gary, I don't know if you'll see this comment, I know you don't do Speedtest G anymore, because it had a low audience and it was extremely laborious and costly to carry out the various tests. But here's my request from a fan of yours, when you feel like being able to do them again, but with a much lower intensity, that is, test only one device of each main SoC, that is, the SD Gen 2, A16 and D9200. Regarding the tests, a CPU peak test can be done, but with the introduction of energy consumption analysis and the average performance, with the GPU the same as the CPU. It can also introduce ray-tracing tests and more complex ISP and AI tests to truly test the capabilities of current and future SoCs.
I hear what you are saying but now you want me to develop ray tracing tests, AI tests, and energy efficiency tests. That is a huge investment of my time. Having said that I do want to do more efficiency tests like I did for microcontrollers.
@@GaryExplains Thank you very much for answering, but if this is really bad and hard to do then don't need to. Your health and well being are more important! But if one day this test comes back, I would be very happy!
What's data gathering hint feature ???
I think X4's 15% performance upllft can't beat Apple's A17 single core performance, Hope Qualcomm is ready with Orion CPU😌
Currently A16 wins by around 25-30% in single thread compared to 8G2. So a 15% uplift will make it closer to A16. A16 will still probably be around 10-12% faster. A17 will be a lot faster since it will be on 3nm while 8G3 will still be on 4nm and 3nm is a full die shrink so expect around 20-30% performance uplift compared to A16. That means A17 will be 35-45% faster than 8G3 in single thread.
More performance per Watt that's how smart design should looks like👍
Can someone help me find information about Cortex-X4 pipelines? Thank you for your attention.
Will any of the other ARM licensees come up with a SoC comparable or better than Apple's M2 or A16?
The RISC side of computing is getting really interesting for us consumers. From SBC to tablet/mobile, laptop and desktop space.
Check out my videos on Qualcomm's Nuvia based CPU called Oryon.
@@GaryExplains It seems I have missed that upload. Going to check it out. Thanks Gary!
@@NexuJin Actually I have several videos about it!!! 😜 th-cam.com/video/daGm4JmrmZ4/w-d-xo.html and th-cam.com/video/Fi0AKrprWxg/w-d-xo.html and th-cam.com/users/shortsaOl_g4vluFs?feature=share and th-cam.com/video/PA5KEjO6NqE/w-d-xo.html
The 't' in 'often' is silent.
Eh?
You do understand that in British English the T isn't silent... Please tell me you know that... Please....
Cortex a7 came in 2014 with Philips designed NEON technology catering to media, which is text, sound, and imagery. And images for gaming are simple arithmatic for ray-tracing.. But they are no good for number crunching required for general computing. Isn't that so Gary?. Your comments please.
Funny, I thought the second fastest supercomputer in the world used Arm processors. What proof or source are you using to say that Arm processors are not good for number crunching? 2014 is a long long time ago. The A7 was a 32-bit Arvmv7 processor. Things have moved a lot since then: SVE, SVE2, etc. Plus Armv8, Arm9, Neoverse, and more.
@GaryExplains what's so funny, my friend?. I am only telling you what I read in Prof. Tannenbaun's (who wrote the MINIX os which is the forerunner to all powerful Linux) book in its latest edition. I am not a computer expert but learn a lot from videos such as yours. I do some android app development and was baffled by the inability to convert them to Apple with M1 and M2, which run on ARM architecture. They seem unable to handle the type of mathematical function.
Sorry, saying "Funny, I thought..." is a English colloquialism, I shouldn't have used it. It means that since the second fastest supercomputer in the world uses Arm processors then your statement that they are no good at number crunching is obviously not true. What where you trying to covert to the Apple with M1? An Android program? What do you mean by "They seem unable to handle the type of mathematical function"?
@GaryExplains my earlier reply seems not accepted.
Anyway, what I was trying to do was converting one app I have already uploaded (but not marketed yet) and another for structural design and drafting to be converted to iPhone apps as most engineers use Apple these days. But I found Apple silicon no longer support java. I like java as it has the lambda function capability(which you should consider for GaryBasic). I suppose this is due to a problem with ARM architecture. Tried to use Flutter but found that also do not support what want to do.
I see. OK. So Android apps are developed using Java or Kotlin and iPhone apps are developed using Obj-C or Swift. I don't think there was ever a time when iPhone apps could be built using Java. This is one of the key challenges in developing for mobile, i.e. cross-platform compatibility. Flutter is cross platform, so that was a good idea. You might want to look into Unity. All this has nothing to do with Arm or the CPU, it is just how Android / iOS are developed.
Will those be used in next gen sd8 gen3 or the next gen 4
Snapdragon 8 Gen 3
On average, how much overall power does the CPU use in a phone? I would have thought that the screen is the main thing, then you still have loads of other hardware suckling from the battery, I have no idea where the CPU fits in here.
So, let's say phone X lasts 24 hours under normal use. Then only the CPU is upgrade and this new CPU is 20% more power efficient. As a rough guess, how much would this generally affect the lifespan of a phone under the same conditions of normal personal usage?
+20% on 24hrs is 4.8hrs but we know it's not going to be close to that, but is it negligible to the overall energy usage of a phone or ~30mins, any other guesses?
You're not going to see that much improvement because the quoted number for power efficiency is under conditions that show it the best, typically at least moderate to full load. Most phones have bursty workloads so the actual battery savings will be much less.
The improvement is worthwhile! It's good to strive for greatness. However, this is definitely not life-changing efficiency improvement.
I would say not more than a few percent difference.
From what I remember Qualcomm specs something like a 10W TDP for their chips. A 15% reduction gets you down to 8.5W under full load.
We'll likely see pretty much the same battery life though, as manufacturers will see this efficiency gain as a way to juice out more performance.
What will happen after 1nm processors. Will it be the end? So soon?
That number is kind of meaningless now it just means it's more compact not that the transistors at one nanometer. They still got quite a few shenanigans for them to burn through still like vertical gaafet, 3D stack transistors, nanotube gaafet, underside power wires and after that exotic transistors materials like gallium nitride, silicon carbide, vacuum channels, graphene, it's not gonna be cheap though. Though this video mainly focused on architecture and they are starting to learn how to improve an architecture without adding more transistors.
#Clairvoyance and #Omnipotence will be Yours.
All the other guys said plus they might start using photons. (But mostly claravoyance and omnipotence.)
@@technolus5742 those words have nothing to do with computers, harrball is bot spreading spiritual BS.
@@AndrewMellor-darkphoton lol I think he's just joking around. I was just kidding about it too - though the use of photons was serious.
Aren't apple chips also arm? But why don't they use arm cortex found in mediatek and snapdragon? Anyone can explain?
Apple designs its own CPUs that are Arm compatible.
I look forward to seeing what Apple does with these new core designs as reference for their M series of chips on their macs and such.
I would also love to see what Qualcomm will be doing and I’m crossing my fingers that something can make it’s way to the laptop space in mind for windows and Linux.
Obviously the phone and tablet space with be the primary use case for consumers but I don’t really feel that much of a difference between generations when it comes to mobile use
Apple doesn't use Arm's Cortex designs in any of its processors. It designs its own CPUs
@@GaryExplains holy cow, thanks for responding to my comment.
I knew Apple designed their own stuff but i didn’t realize they didn’t rely on Arm’s specs at all.
Thanks for clarifying this for me
Apple does rely on Arm's specs, it makes compatible CPUs according to the Armv8 or Armv9 documentation.
I wonder what would cortex X4 equivalent intel/and processor be when it comes to performance. I know it's different architecture
IIRC it's similar to Alder Lake at the same frequency. I don't have any X4 silicon to test, but given how X3 vs Tiger Lake plays out I wouldn't be surprised.
@@DigitalJedi to be clear, are you saying it might be similar to alder lake in terms of having the same TDP for comparison? Or unconstrained?
for me its a lot of marketing talk.
it would be interesting how power efficient these cpus really are. how much watt do they use etc.
So what's your estimate geek bench single core score
To my gues it will be bw 1650-1720 in geekbench whts yours and to me it's so sad that it won't even match the apple a16 which wa basically a15
I hope they are able to atleast match the efficiency
I am curious if Apple is using the different various for their upcoming M3 or M4 processors. A combination with 3nm and their own mix would be killer.
Different various? Remember that Apple designs its own Arm compatible cores and doesn't use Arm's Cortex designs.
@@GaryExplains we need a company to make integrated memory with arm with Linux or Android on it.
@@univera1111 I am not quite sure what you mean by that. Can you rephrase it?
@@GaryExplains arm is very efficient, we have yet to see am arm soc at m1 or m2 proficiency level. The plan is to run Linux on m2 ultra as bare metal as possible. Since that's no yet possible. We need an OEM arm chip that comes to as close to m1 or m2 in performance.
It's an honor speaking with you Gary. From third world Nigeria.
Apples a17 should remain a good bit ahead still but this looks like a very nice improvement for arm.
I think the gap is slowly closing not because ARM is improving faster but because Apple has not improved as much YoY.
It's harder to improve when you are pushing the boundaries.
@@Henfredemars also I believe they have staffing issues surrounding chip development and associated divisions
@@Henfredemars Dunno mate , I suspect its more to do profit , apple has been a mile ahead for years in smartphone performance until recently , and if you are ahead , why spend more coin , i`d say it is very much like features and accessories that each new phone gets each year , clearly these phone makers could throw in many more features and add even more upgrades and value if they wanted , but a smarter business game is to just add enough of a carrot and some marketing to get the masses buying new phones , while allowing a few carrots in the bag for next year and the year after , i`m pretty sure these big companies play the long game with their strategy's and marketing , its all about the money ;) .
@@mikldude9376 apple tries to do everything in house but when companies that just do 1 thing get good at it, it is hard to compete.
qualcomm, mediatek, arm is just soc develepment. that is all they do
For sure, the SD8G2 is ARMv9, whilst the A16 is ARMv8.6. So we should see better ARM architecture in the A17. Then you factor in the new nodes, N3/N3E, and you should get a big generational leap.
Let's hope they take proper advantage of this I'm sick of $2000 phones and they're spending like 100 bucks on silicon. Maybe they can do a smart phone with 2 + 4 + 6 with full cashe.
so this means by the same area, we've just got 10% more performance after 3 generations.
0. a78
1. a710
2. a715
3. a720
Yes, more performance generally means more transistors, that is kind of the core of Moore's law.
Wish we can get a chip that have a similar performance to Apple M-series chip.
1) A powerful ARM based Windows Laptop
2) A powerful Linux based Laptop or Handheld gaming device
3) Next gen Nintendo Switch with maybe 2TF to 3TF GPU performance
4) Powerful VR headset
5) Ipad level Android-base Tablet.
Hopefully AMD releases a ARM soc with a 5g chip form Xilinx.
How did you measure all that speed ups?!?
That is reported by ARM not by testing of real handsets, and they don't typically provide detailed reports as to how they make that estimate.
It's a claim.
Can you do a video on NM process and what will happen when we go to a SoC 1nm processs. Like in future phone's.
At one point, the size will stop shrinking, and further less nm means more dense transistors in SoC
Those new CPUs seems promising, but a bit unrelated... my "dream" is to use an extremely low power-demanding PC able to do daily browsing tasks reasonably well below the 1W mark. (Something like 1080p, steady framerates.)
1 watt you said?
You have no idea what you are saying.
Any decent BASIC cpu consumes at least 20-35 watts while busy, TODAY!
Yes, there are 10 watt ARM CPUs but you would not want them for basic Destop duties!
@@pepeshopping Said the one who just let a WISH go over your head that easily.
If I remember correctly, Apple M1 can handle word processing and watching online videos (if the video can be hardware decoded by SOC) within 1W (only considering SOC wattage), perhaps it takes a couple years of fab for it to be possible, but considering the increasing processing power requirements of web browsing, I seriously doubt that
Iphones can't even do that, that's future tech.
Phone CPU 's around 5 watts yeah and laptop 15 to 30 watts So what you're describing sounds like a smartwatch processor or IOT embedded those are pretty slow.
Hope MTK makes a good designs from these CPUs, Cuz Qualcomm is no longer it used to be
I don't think the 15% increase even reaches apples A15 in single and multicore. Pathetic. Let alone when the A17 comes out
And with the power draw being less, is it independent of the upcoming 3nm chips as this will also drastically reduce power draw 🤔
IIRC ARM does power testing on the current flagship node, not the next one.
N4P and 15% improvement is gonna be A16 levels of single core. Which is excellent. Android used to be 3 years behind in single core. Now they're 1. Eventually at this right will be right there.
@@Bibbatron A16 levels isn't good. This is because the A16 is just an overclock of the A15 meaning that it's not that Qualcomm has become better but apple had a stop gap year that's why they are closing in
@@ifrankenstein7455 Well still they're only 1 year behind. If Apple fumbled last year I don't really care. At the end I care about the overall competitiveness.
But still 2mb of l2 cache is not even close to a16 bionic with 16 MB (performance cores)
4 MB (efficient cores)
24 MB (system cache)
Of l2 cache
How it is even possible that Apple ,which is using arm architecture, can bump up l2 cache to that crazy numbers?
It almost never scales linearly. Usually a drastically larger cache also means worse latency in terms of cycles. For example when Intel moved from the c2q to i7 family architecture, l2 cache size shrank all the way down to just 256K but latency improved so much it was a big net improvement.
@@Claudius_Nero Well, it's interesting then how new chips with new core design will compete with latest Apple's SoC.
You need to consider the market. All of Apple's devices will use their cores, so they can afford to make expensive chips not very area efficient. ARM/SoC vendor has to make a profit at that stage and their market of OEMs is not willing to pay that much.
@@Henfredemars well, apple is not using their own cores as far as I know, they are using ARMv8.5-A Everest and Sawtooth cores, created by ARM in the latest A16 Bionic. Doesn't that mean anyone can use these type of cores? It's enough to buy a licence, isn't?
Apple going to hopefully design am even greater chip for future products
I love my M1 MacBook (not upgrading yet) but I'm hoping someone finally pulls their finger out to properly compete with them on desktop tier ARM SoCs.
Hopefully Qualcomm can make apple run for its money for once!
Qualcomm has never been able to maximise the potential of ARM, but they have made good strides lately, having sacked Samsung Fabs as their node of choice. But it could be argued that Apple's hardware and software integration is better implemented.
So can we all recognize that ARM went to producing chips themselves. Aka now the subs like Apple, phones, etc. Now buy actual ARM chips. Making dev 100 times easier!!!
Eh? Arm doesn't make chips.
Qualcomm went against ARM in the Nvidia purchase. There has to be a lot of hard feelings about that and in 2024 when Qualcomm starts making their Nuvia chips, as the lawsuit shows. Arm wants Qualcomm to destroy their last 3 years of chip design. Going to get very ugly.
Qualcomm didn't go against Arm and there aren't hard feelings. The problem is about licensing. I have several videos about the whole saga here on this channel. Arm doesn't have any hard feelings against Apple, or other partners who have architectural licenses. In fact Qualcomm used to design its own Arm based CPUs as well, that is why Arm grants architectural licenses, it is part of its business plan.
I’ll be mostly disappointed if the new PI5 doesn’t have a maxed out SOC with at least 10 Cortex-X4s. And the GPU obviously.
🤦♂️
It's unlikely it will, I doubt that those SOCs are cheap
I get so annoyed at the inconsistency of naming things. I see it as “ARM” and “Arm” and “arm.” I’m fine with all caps or all lower-case (as a stylistic choice), but since it’s an acronym, I hate seeing “Arm.” Intel does the same thing referring to their new GPUs as “ARC” and “Arc.” About as annoying as acronyms with too many letters or start using characters in the middles of words just to make a preferred name work. e.g. “TBT” for “throwback Thursday” when it should be “TT.” Sorry, I’m in a mood. :-)
ARM itself rebranded to Arm several years ago.
@@GaryExplains I know (and I hate that :-) ), but even they haven't been consistent about it.
I think Arm has been consistent about that, can you give me an example where it hasn't?
So companies can fuck us over with those a 720 cores since it doesnt seem like the worse and better ones are labeled differently. Great
Eh?
@@GaryExplains they say companies can choose to configure the A720 with minimum performance and area to reduce costs, while consumers will not be able to tell which phones have small A720 and which ones have big A720.
The transparency in hardware has been lost years, go anyway. When enterprises like Crapple ignore all specs while integrating components, making the components run outside their specs and have them throttle near permanently for example ..
@@YounesLayachi Exactly that, thanks. I mean, the difference in performance probably wont be noticeable in 90% of situations, but still.
@@wiggattelo I think with 2 cortex-X cores, then the A720 performance might not matter too much, so I'm looking forwards to what Google will be doing for their G4
x86_64 ftw!
Cpu performance isn't that relevant these days. It is running a phone but not a super computer. Ai performance is more relevant to run matrix workload with big memory requirement. That means tpu is the key.. Speech recognition on a phone is still really poor
Perhaps CPU performance is not as critical, but higher performing CPUs we can often run at lower power to achieve the same amount of work. We care greatly about battery performance in mobile devices.
1st Comment
Your second actually
@@AndrewMellor-darkphotonYou are correct except it's "you're" or "you are". Just saying :) It's one of my pet gripes.
@@AndrewMellor-darkphoton nvm
1 😁😁😁
just 15%....
This is just the generic Arm IP before tweaks from Apple, Qualcomm....... etc
Just? What other service or product do you use that offers double digit performance increases year on year? Asking for a friend.
@ATSounds Apple?
@@GaryExplains doesn't Apple use Arm's IP to design they're chips?
@ATSounds No, it uses its own designs.
hi
Hi
Normal intergenerational development. Nothing special.
And what is wrong with "normal" generational improvements?
ARM is boring.. besides SBCs there's no interesting product carrying these.
Eh?
So, every smartphone and tablet is boring, Nvidia and Ampere's enormous server chips are boring, and microcontrollers are boring?
@@DigitalJedi Especially smartphones. Average at anything, good at nothing. No significant development since ten years, but loss of user swappable batteries, audio jacks and extension slots (SD).
Servers always where boring, because they are nothing but workhourses. ngreedya's only non boring property is them being frustrating. Microcontrollers have a thing for them, but the more isteresting ones don't use ARM
Well, SOHO routers using ARM are a good thing.
@ElMariachi That means you find tech boring and you would say the same for Intel and AMD etc. Don't try to make your lack of interest Arm's fault.
@@elmariachi5133 You are one of the usual suspects always complaining of no SD, removable battery or jack! Get over it, tech moves on, and they are dead and buried.