Arm Unveils Next-Gen 76-Series Mobile CPU, GPU Cores (hothardware.com) 54
MojoKid writes: Last week, Arm showed off its new Machine Learning Processor design, but today it has lifted the veil on its next-generation Cortex and Mali CPU, GPU, and VPU architectures, destined for 2019 smartphones and mobile devices. The Arm Cortex-A76 CPU, Mali-G76 GPU, and Mali-V76 VPU designs all step up performance and efficiency over previous generation designs, though there are architectural and layout changes and more advanced manufacturing processes.
Arm believes its A76 core, which can be clocked at 3GHz+ when produced on a 7nm process, can perform within 10 percent of an Intel Skylake core within the same thermal constraints, but at approximately half the footprint. The Mali-G76 improves density and energy efficiency by 30 percent over the previous generation G72, while providing a 2.7x uplift in machine learning workloads. And the Mali-V76 VPU improves on the recently announced V52 by adding support for 8K UltraHD content, among many other improvements.
Arm believes its A76 core, which can be clocked at 3GHz+ when produced on a 7nm process, can perform within 10 percent of an Intel Skylake core within the same thermal constraints, but at approximately half the footprint. The Mali-G76 improves density and energy efficiency by 30 percent over the previous generation G72, while providing a 2.7x uplift in machine learning workloads. And the Mali-V76 VPU improves on the recently announced V52 by adding support for 8K UltraHD content, among many other improvements.
Wait what? (Score:2, Informative)
Getting to within 10% of a skylake core (you know it's really probably 15 to 20%) with the same heat dissipation and at 7nm is terrible even if the footprint is 50%.
Skylake is a 14 nm chip. That means if Intel blindly shrunk it to 7nm, it would perform way faster and take up approximately half the footprint too .. AND have less heat dissapation. So I am not sure what ARM is bringing to the table. I expected way better from ARM.
Switch to RISC V .. ARM is dying.
Re: (Score:2)
WTF are you talking about? Getting to 90% of Intel single-core throughput is freaking awesome for a chip that costs a lot less and mainly operates at the low end of its thermal profile.
Re: (Score:3)
They're comparing theoretical performance of a 7nm chip with a 14nm one, while both consuming the same power yet performing not as well.
So what they're really saying is they're getting 10% less performance per watt with much better, more efficient process.
Re: (Score:2)
Also, for what it's worth Intel's 14 nm process (14++) is also performing better than their own first generation 10 nm process, which is about th
Re: (Score:2)
My takeaway from this announcement is: ARM already beats Atom in every way and consequently is now taking aim at Core arch, which also beats Atom, so there you go. If the ARM chip costs less and runs cooler, I'm most probably in for my next laptop. Likely to hit the streets in Chromebook form and be a hit with school kids. With a decent amount of flash and confirmed Debian friendliness, I'll take a flyer.
Re: (Score:2)
"f the ARM chip costs less and runs cooler, ..."
Wait, haven't you already declared that it does both these things?
"Likely to hit the streets in Chromebook form and be a hit with school kids. With a decent amount of flash and confirmed Debian friendliness, I'll take a flyer."
Critical thinking isn't your specialty, is it? Good thing that "confirmed Debian friendliness", how will Intel compete? No wonder it will be a hit with school kids, Debian all the rage there.
It's sad what ignorant tribalists people ha
Re: (Score:2)
I feel more stupid after reading your post.
Re: (Score:2)
I'm afraid we can't really compare the claims about power and speed, because ARM is referring to a 7 nanometer process that doesn't really exist in mass production, and comparing it to an Intel process that is getting rather long in the tooth.
TSMC, Glofo and Samsung 7nm is roughly comparable to, maybe a bit better than, Intel's stalled 10nm node. AMD has 7nm samples in the lab. That means that TSMC has caught up with Intel, maybe just a few months lag now, or maybe it is in reality a tie race in terms of commercially viable production.
Re: (Score:2)
Don't forget speculative execution too.
Re: (Score:2)
A76 cores won't be cheap, at a smaller process than the Skylakes they can't beat in thermal or speed, they're still half the area size at 7nm than Skylake is at 14nm.
They're massive compared to the cheap ARM cores, like A53. They're adding massive numbers of transistors to get this extra performance. A76's apparently won't run some 32bit code either, you'll need a big.little chip so you can run Aarch32 code on the little cores.
Don't get me wrong, I still want ARM to succeed outside the smartphone, tablet an
Re: (Score:3)
Considering that the article never said it "mainly operates at the low end of its thermal profile" (nor that it was differentiated from Skylake in that respect), nor that it "costs a lot less" presumably than Skylake, nor that it gets "90% of Intel single-core throughput", it appears "freaking awesome" to you based on facts that you've 100% made up.
Re: (Score:2)
You jerk. You know that ARM costs a lot less than Intel because of the Arm Holdings licensing policy, and it typically operates at the low end of its thermal profile because of the market it is aimed at. Don't be disingenuous.
Of all the issues with ARM, the biggest one for Intel is its low cost. Intel relies on a huge margin to continue to support itself in the manner to which it has become accustomed. Arm holdings is a leaner operation. When ARM gets anywhere close to Core arch at its lower price point, In
Re: (Score:2)
On the first slide of the article, the "overarching design goal" is to "outperform the competition at half the area and power". They then go on to say that the design offers 35% more performance and 40% better power efficiency compared to a Cortex-A75 core, yet less performance than Skylake with the same "thermal constraints" and half the footprint.
The conclusion should be that either "the competition" is Cortex-A75 or they failed to meet their "overarching design goal". Perhaps they reach their "thermal c
Re: (Score:2)
Yeah, it costs way less and have 90% of the performance of an already very good platform for desktop... it's dying!!!
90% of the performance only if you were to artificially handicap the Skylake core.
90% as good as a Skylake core with the same power (Score:2)
Cool, why did they compare it to a Kaby lake or a Coffee lake? They're both still 14nm chips, how bad can a 7nm A76 be compared to them?
Re:90% as good as a Skylake core with the same pow (Score:4, Interesting)
You're implying 7nm is actually that much smaller than a 14nm process.
In reality, it isn't. That's just describing the smallest feature size, usually the distance between components, not component size itself.
It's almost entirely marketing. TSMC's 7nm is in many respects exactly like intel's 10nm.
*SMARTPHONE* (Score:4, Interesting)
How can the A76 be compared to Skylake when it doesn't have Skylake's X86/X64's command set and cannot run Skylake's X86/X64 code natively?
The instruction set is absolutely completely irrelevant.
It's targetting smartphone/tablet/embed not desktops.
i.e.: a market that almost exclusively runs Android (save for Apple, a couple of things trying to add full blown GNU/Linux support (SBCs, after market OSes like Sailfish), and the big joke coming from Microsoft).
not a market that is stuck to Windows 10.
i.e.: a market where (parts of) the OS is available for free and can be compiled for your CPU (AOSP, Armbian/Sailfish/Etc.), and most of the applications are delivered as bytecode that get JIT/AOT during installation and will run on whatever architecture your CPU runs (except for a few apps packing native libraries, but those are usually available for both x86 and ARM arches)
not a market that is stuck with proprietary closed source blobs.
i.e.: each CPU will be running natively. the support for x86/x86_64 is completely irrelevant.
Nobody with any level of sanity wants to run Microsoft Windows binaries on a smartphone/tablet/SBC.
The problem of comparing lays elsewhere :
- the Skylake is a chip that actually exists in the real world. You could make the Android x86 benchmarks on it using some dev board.
- ARM Cortex-A76 is a core. A design. That a company could license and then ask TMSC to actually build. There are no current physical chips that you could benchmark Android on them. You need first for, e.g.: Qualcom to announce they'lll use this core in the upcomming Snapdragon 900 or whatever number they'll decide to slap on it. Once they start producing actuall chips, you'll finally be able to get real world numbers.
Until then all you have is engineers' speculations "can be clocked at 3GHz+ when produced on a 7nm process, can perform within 10 percent of an Intel Skylake core within the same thermal constrain". Not necessarily will.
Re: (Score:2)
"It's targetting smartphone/tablet/embed not desktops."
FTFA: "Arm claims laptop-class performance with the A76." "Designed to run flat-out in multi-core implementations".
Doesn't sound like "smartphone/tablet/embed" at all. "flat-out multi-core" and "laptop-class" is not how "smartphone/tablet/embed" works...ever.
"i.e.: a market that almost exclusively runs Android (save for Apple, a couple of things trying to add full blown GNU/Linux support (SBCs, after market OSes like Sailfish), and the big joke coming
Re: (Score:3)
"That's just describing the smallest feature size"
It's not even that. Calling a node "7nm" just means it comes after the node called "10nm" and before the node called "5nm". Even that is not a given since retroscaling is becoming popular these day.
There's a good chance that *no* features in a 7nm process are actually 7nm.
Behind Schedule (Score:2)
The V76 is really geared towards the next generation of 8K Ultra HD content, like planned coverage of the upcoming 2018 Winter Olympics in PyeongChang.
The 2018 Winter Olympics ended months ago [pyeongchang2018.com].
76 huh? (Score:2)
Future Raspberry Pis ? (Score:2)
Given that the Linux devs are also getting good support for the Video Core V and 6,
we might indeed get within a couple of years some Raspberry Pi running newer SoCs from broadcom.
Until then, you can count on the less known/widespread manufacturer to make SBC with A76 cores from other chip makers.
Question for someone who actually knows ARM (Score:2)
This is great. But so was A75. And I don't see any pickup of that. I'm looking in the Chromebook space. They all seem to be based on MUCH older ARM designs.
I see blurbs saying N Billion (with a B) ARM 53 cores have been used in things. Where are those? In Chromebooks? Or are these things really pitched at the set-top box/smart TV/embedded device market?
I have an old, Samsung series 3 Chromebook. It's 6 years old. And I'd love to replace it, but it's years old and I don't see ARM-based Chromebook wit
Re: (Score:2)
I see blurbs saying N Billion (with a B) ARM 53 cores have been used in things. Where are those? In Chromebooks? Or are these things really pitched at the set-top box/smart TV/embedded device market?
Cars electronics. Especially camera based driver assist systems often use ARMs.
Re: (Score:2)
Set-tops, TVs and SBCs .. Fire TVs, tablets, Android phones ..
Keep in mind that e.g. the ARM Cortex A53 is just a standard. The implementations/manufacturers are pushing bounds on clock speeds, bandwidth, power optimization for incremental improvements between standards.
I just finished moving to a Rock64 (https://www.pine64.org/?page_id=7147) as my main desktop machine. 4GB RAM, 4x cores, 4k LXLE desktop, media/TV server and 16x drives in two 16TB RAIDZ2 arrays.. and idling at a few watts.. for $50.
The n