Arm's first Armv9 CPU and GPU designs offer generational leap in performance

[nanoguy]

In brief: Arm already dominates mobile, but now it's gearing up for a world where everything from small embedded devices to servers and supercomputers can be built using its scalable architecture. For consumers, Armv9 means we'll see faster smartwatches, phones, tablets, laptops, and even desktops built to compete with x86 systems, fueled by Apple's transition to custom silicon.

Arm announced its next generation processor architecture -- Armv9 -- earlier this year marking the first true generational upgrade since 2011's Armv8. The company hopes this will accelerate the shift away from x86 by allowing chip makers to utilize technologies that are at the heart of the world's fastest supercomputers and deliver better machine learning and digital signal processing capabilities.

For the enterprise infrastructure, those capabilities are encompassed in the Neoverse V1 and Neoverse N2 CPU IPs, building on the strong foundation of the Neoverse N1 platform that Ampere and other companies are using to compete with Intel and AMD in the data center.

At first glance, it may look as though Armv9 is all about high performance computing scenarios, but it does have some serious implications in the consumer space, too.

This is where the new Cortex-X2, Cortex-A710, and Cortex-A510 CPU designs come in. As you may have guessed, these are the successors of last year's Cortex-X1, Cortex-A78, and Cortex-A55. All three add support for SVE2 vector extensions and come with much-needed IPC and efficiency improvements.

Arm says the Cortex-X2 is 30 percent faster than its predecessor, which is used in custom designs in the Qualcomm Snapdragon 888 and Samsung Exynos 2100 chipsets.

The Cortex-A710 core is 10 percent faster and 30 percent more energy efficient than the Cortex-A78. The new "LITTLE" Cortex-A510 core is 35 percent faster than the Cortex-A55, while also delivering three times the performance in machine learning applications.

The company also revealed four new Mali GPUs that come with similar improvements over their predecessors. At the high end, there's the Mali-G710 with a 20 percent performance boost, but the most impressive of the bunch is actually the Mali-G510 that will double the performance you can get in mid-range smartphones and smart TVs.

Armv9 definitely sounds impressive, but don't expect to see these designs show up this year. Chip makers will most likely announce their custom designs in the coming weeks and months with a launch window set for the first half of 2022. There's a lot of potential for new and exciting Chrome OS and Windows 10 on Arm hardware, and everything down to the lowly smartwatch will likely benefit from the new architecture.

Permalink to story.

https://www.techspot.com/news/89817-arm-first-armv9-cpu-gpu-designs-offer-generational.html

 

[psycros]

As always, it comes down to power efficiency. If the performance gains come at the cost of battery life its not really an advance. Still, if anyone can find that balance its ARM.

 

[Geralt]

As always, it comes down to power efficiency. If the performance gains come at the cost of battery life its not really an advance. Still, if anyone can find that balance its ARM.

Will ARM find its way to desktop PC's too? Will we have to upgrade our Ryzen/Core i9, etc. to ARM in the future? Is this so drastic?

 

[Bullwinkle M]

Will ARM find its way to desktop PC's too? Will we have to upgrade our Ryzen/Core i9, etc. to ARM in the future? Is this so drastic?

Oh No, The Apple / Mac M1 Arm chips will never be on a desktop PC

LoL

Not Mine anyway

 

[Kosmoz]

I don't see either AMD, nor Intel losing the desktop race to ARM. In the worst case scenario at least one of them will still hold the fort, so far I'm counting on AMD, at least.

 

[GregonMaui]

I thought the article said "generational leap". The snapdragon 888 is not that particularly fast. The "leap" might make it close to as fast as the A14. I was expecting leap to put in in M1 territory.

 

[GregonMaui]

I don't see either AMD, nor Intel losing the desktop race to ARM. In the worst case scenario at least one of them will still hold the fort, so far I'm counting on AMD, at least.

maybe lose is the wrong word. ARM will be very competitive. One of the worlds fastest supercomputers is already based on ARM, the server market is being aggressively entered into by ARM. so lose, or compete with? And why is AMD developing its own ARM chip - interesting

 

[Hardware Geek]

The current x86-64 machines won't be displace anytime soon, but I wouldn't be surprised to see a lot of the low and middle cost machines leading a slow transition to arm. It will be primarily consumers that drive the transition. Corporate computing environments won't switch for a long time due to extensive use of legacy software that will need to be nearlt flawlessly emulated before they switch.

 

[GregonMaui]

The current x86-64 machines won't be displace anytime soon, but I wouldn't be surprised to see a lot of the low and middle cost machines leading a slow transition to arm. It will be primarily consumers that drive the transition. Corporate computing environments won't switch for a long time due to extensive use of legacy software that will need to be nearlt flawlessly emulated before they switch.

That is probably fair. maybe this is why Microsoft won't release Windows on ARM! And its so fashionable today to sue for anti-trust. coalition of the "Free Us from Intel".. Just kidding

 

[Michael7]

With Intel adopting bigg.little architecture later this year and AMD following in about 2 years I don't see ARM displacing x86 on Windows.

 

[GregonMaui]

With Intel adopting bigg.little architecture later this year and AMD following in about 2 years I don't see ARM displacing x86 on Windows.

Help me with this. You are saying both Intel and AMD are developing arm architectures, but won’t replace x86 on windows?

 

[Markoni35]

Allegedly Intel is working on a new RISC CPU called "LEG" while will be able to kick ARM back to its place.

 

[GregonMaui]

Allegedly Intel is working on a new RISC CPU called "LEG" while will be able to kick ARM back to its place.

LOL. Bye buy Intel.

 

[Michael7]

Help me with this. You are saying both Intel and AMD are developing arm architectures, but won’t replace x86 on windows?

They are not developing ARM but hybrid architecture just like ARM bigg.little. Intel will debut its Alder Lake with next gen Core and follow up to Atom cores in it.

 

[GregonMaui]

They are not developing ARM but hybrid architecture just like ARM bigg.little. Intel will debut its Alder Lake with next gen Core and follow up to Atom cores in it.

https://www.tomsguide.com/news/look-out-apple-m1-amd-reportedly-developing-arm-silicon-chip. Some beg to differ

 

[HardReset]

maybe lose is the wrong word. ARM will be very competitive. One of the worlds fastest supercomputers is already based on ARM, the server market is being aggressively entered into by ARM. so lose, or compete with? And why is AMD developing its own ARM chip - interesting

That supercomputer is very inefficient one. You can build "world fastest supercomputer" using ANY chips that have ever existed. Even supercomputer using Motorola 68000 CPU's could be world's fastest, just leave efficiency out.

Server market enters ARM still quite slowly, some custom designs and such but since x86 is still performance king, ARM has long way to go.

AMD have been developing own ARM chip since 2013, one design was already abandoned and next comes who knows when.

https://www.tomsguide.com/news/look-out-apple-m1-amd-reportedly-developing-arm-silicon-chip. Some beg to differ

That news is 6 months old now. AMD didn't say anything about it on CES January. AMD has not told anything about upcoming ARM chips for years.

 

[GregonMaui]

That supercomputer is very inefficient one. You can build "world fastest supercomputer" using ANY chips that have ever existed. Even supercomputer using Motorola 68000 CPU's could be world's fastest, just leave efficiency out.

Server market enters ARM still quite slowly, some custom designs and such but since x86 is still performance king, ARM has long way to go.

AMD have been developing own ARM chip since 2013, one design was already abandoned and next comes who knows when.

That news is 6 months old now. AMD didn't say anything about it on CES January. AMD has not told anything about upcoming ARM chips for years.

Maybe, I’ll give you credit for being right even though some of your points can be disputed, because none of your points matter. The fact that early versions of arm are able to have these accomplishments is the point period. Is the arm supercomputer the best ever built, probably not, but as proof of concept, it is remarkably successful. It will probably spur additional designs, more funding and better accomplishments in the next few years

x86 and wintel have dominated for decades sucking up most of the development and funding dollars. Then Intel got fat dumb and lazy, stagnating horribly for several years. It took AMD to produce better, faster and cheaper alternatives to Intel to push Intel back on track. And yet they can’t even manufacture chips to modern 5nm and below specs whilst others can. Sure they will be back in the game, and are moving rapidly there. But arm with RISC architecture, despite much lowers r&d and development dollars has become highly competitive because of work by Apple and others making faster and faster socs

implicitly there is no real advantage of CISC over RISC. Even the x86 instruction set is internally processed as a series of simpler instructions. The same thing can be achieved in software to map x86 to a series of RISC instructions and compiled. Aka Rosetta 2

so the point arm will keep getting better, probably faster than x86 can keep getting better and there will be competing architectures achieving similar results. After all 2+2 is still 4 whether on RISC or CISC

 

[HardReset]

Maybe, I’ll give you credit for being right even though some of your points can be disputed, because none of your points matter. The fact that early versions of arm are able to have these accomplishments is the point period. Is the arm supercomputer the best ever built, probably not, but as proof of concept, it is remarkably successful. It will probably spur additional designs, more funding and better accomplishments in the next few years

It's not any kind of "proof of concept". We have seen supercomputers built on 15 year old chips that were quite fast at that time. Just put enough CPU's on same computer and it will be fast on simple software like Linpack.

x86 and wintel have dominated for decades sucking up most of the development and funding dollars. Then Intel got fat dumb and lazy, stagnating horribly for several years. It took AMD to produce better, faster and cheaper alternatives to Intel to push Intel back on track. And yet they can’t even manufacture chips to modern 5nm and below specs whilst others can. Sure they will be back in the game, and are moving rapidly there. But arm with RISC architecture, despite much lowers r&d and development dollars has become highly competitive because of work by Apple and others making faster and faster socs

implicitly there is no real advantage of CISC over RISC. Even the x86 instruction set is internally processed as a series of simpler instructions. The same thing can be achieved in software to map x86 to a series of RISC instructions and compiled. Aka Rosetta 2

so the point arm will keep getting better, probably faster than x86 can keep getting better and there will be competing architectures achieving similar results. After all 2+2 is still 4 whether on RISC or CISC

Apple's competitiveness is much about fact Apple uses most advanced process at this time. They have money but quite soon AMD will receive same process and then Apple's CPU's are not so fast any more. ARM has become "competitive" but problem is: it's easy to make CPU that gets close to high end x86 CPU's but getting that last performance out is very hard.

There is also not real advantage of RISC vs CISC. Software emulation is always slower than native and things like Rosetta 2 won't likely be very popular.

CISC vs RISC battle have been around 30 years already, interesting to see final result. If we will ever see it.

 

[SixTymes]

I said it before, ill say it again. Intel is currently making yet another big mistake, they should have skipped 10th or 11th gen and gone right to 12th, and should have been releasing 12th gen this year, not next. Since they didn't, it will provide ARM and others to gain more traction. That said, Intel still holds the reins on stability and long lasting platforms when it comes to their chipsets and procs, but, we are seeing less and less people care about that. Intel better push their release schedule up a bit, and hurry and get on with socket 1700+DDR5 like very soon.

 

[Michael7]

https://www.tomsguide.com/news/look-out-apple-m1-amd-reportedly-developing-arm-silicon-chip. Some beg to differ

This is just a rumour and old one at that. What is not a rumour is that the next Samsung Exynos SoC will have AMD RDNA2 graphics baked in with supposedly impressive performance gains. Since AMD helps Samsung build it's ARM SoC it does not look like it will build its own SoC because why help the competitor? Besides it would be very risky both for AMD and Intel to enter the market with well established Qualcomm, Samsung, Mediatek and Apple.

 

[HardReset]

This is just a rumour and old one at that. What is not a rumour is that the next Samsung Exynos SoC will have AMD RDNA2 graphics baked in with supposedly impressive performance gains. Since AMD helps Samsung build it's ARM SoC it does not look like it will build its own SoC because why help the competitor? Besides it would be very risky both for AMD and Intel to enter the market with well established Qualcomm, Samsung, Mediatek and Apple.

"Well established?"

Phone manufacturers will use any SOC that offers good performance with good price. They are not locked into single manufacturer (except Apple). Samsung makes their own chips but still seem like to use AMD GPU. Also if AMD ARM core is faster, not much more expensive and not much bigger than ARM's, manufacturers will gladly use it and ditch Qualcomm.

 

[Wereweeb]

I thought the article said "generational leap". The snapdragon 888 is not that particularly fast. The "leap" might make it close to as fast as the A14. I was expecting leap to put in in M1 territory.

1) This is not about the SD 888, so you didn't even read the article you're commenting on.

2) Anyone can make something as fast as the M1, they just need to use larger cores, bleeding edge silicon and a lot of accelerators in a closed ecossystem. Such a thing would be economically suicidal for any brand which hasn't organized a cult around itself.

 

[Michael7]

"Well established?"

Phone manufacturers will use any SOC that offers good performance with good price. They are not locked into single manufacturer (except Apple). Samsung makes their own chips but still seem like to use AMD GPU. Also if AMD ARM core is faster, not much more expensive and not much bigger than ARM's, manufacturers will gladly use it and ditch Qualcomm.

Sure they can ditch Qualcomm but what are the chances of any newcomer coming to ARM market and beating Qualcomm? It took Huawei 5 years to reach Qualcomm's level of performance and they had the advantage of putting their SoCs in their phones. Neither Intel nor AMD make smartphones so yes it will be risky for them.

 

[HardReset]

Sure they can ditch Qualcomm but what are the chances of any newcomer coming to ARM market and beating Qualcomm? It took Huawei 5 years to reach Qualcomm's level of performance and they had the advantage of putting their SoCs in their phones. Neither Intel nor AMD make smartphones so yes it will be risky for them.

Qualcomm uses ARM "default" cores. Nothing so special AMD couldn't compete with it just making better ARM cores and some level of integration. AMD's faster ARM core and faster GPU, Qualcomm would have no chance on performance side.

However we are still waiting AMD's CPU's with ARM cores.

 

[Markoni35]

Sure they can ditch Qualcomm but what are the chances of any newcomer coming to ARM market and beating Qualcomm?

Michael7, I'm looking at your profile picture and you look almost identical to me. It really is true that everyone has a twin somewhere.