Discussion ARM unveils Cortex A78/X1

[Tabalan]

Hi,
this topic is about yesterdays ARM big announcement - Cortex A78 (which was expected by everyone) and Cortex X1 (which was surprise, I couldn't find any leaks or gossips about new, big core from ARM, closest thing I could find are some extremply sparse info about next year Mattlehorn core). Link to AnandTech article:

Arm's New Cortex-A78 and Cortex-X1 Microarchitectures: An Efficiency and Performance Divergence

www.anandtech.com

What surprised me is the fact that ARM decided to unveil Cortex X1 now. Currently, in mobile you have low performance Cortex A55, high performance A78 (A77 is described as "Leadership performance and efficiency for 5G mobile solutions ") and even higher performance X1. From my point of view, we lack medium core, which cxould be used in mid range SoC (Snapdragon 600, Kirin 600 and so on). I'm aware that Cortex A78 couldn't have same performance as A77 (but lower area and power consumption, shifting into middle core), because it would force ARM clients into noticeably bigger and more power hungry X1. This (pretty much forcing into X1) could negatively reflect on ARM-client relationship in future.
Next year ARMv9.0 should be announced, so, honestly, I expected new cores (huge, medium, small) to be released then. I think new small core will be mandatory, because there might be some problems with big.LITTLE combining ARMv8.2 and v9.0 cores (at least I assume so).

Another things, it's interesting to whether manufacturers will decide to go for X1 or stick with A78. I think all of them might jump on X1 train to not be left behind in benchmarks ("...but new Exynos is so slow, new Kirin/Snapdragon is 20% faster") with tri-cluster solutions (A78 would act as medium core, X1 would replace current A77 with higher clock speed), while mid-high (Kirin 800, SD700) series will adopt A78 + A55 combo.

What are your thoughts on this?
<<I'm just tech enthusiast, so pretty sure I'm wrong somewhere. Point it out or disregard it >>

Edit:
"What’s really shocking here is how close Arm would be getting to Intel and AMD’s current best desktop systems in terms of performance. If both incumbent x86 vendors weren’t already worried about Arm’s yearly rate of improvement over the last few generations, they should outright panic at these figures if they actually materialize – and I do expect them to materialize. "
Love this part. Considering vast majority of people use browser, Office and some minor programmes, is it pretty much safe to say we reached the point where we can shift from x86 to ARM? Yes, pro software and games would stay on x86 for now, but everyone else could migrate to ARM without any problem.

 

[Doug S]

Don't forget nVidia, the use of A78 on Tegra Orin suggests Carmel didn't make the cut.

Don't bet on it, mobile SoC's have consistently pushed the state of the art on process nodes ahead of other market segments.

Besides TSMC's 3nm will not be nearly as different as Samsung's as it is targeted to remain finFET instead of moving to MBCFET as Samsung will on 3nm.

What he was saying is that we'll be on some sort of 5nm+ in 2021, not that ARM cores will not eventually move to 3nm when it is available (in 2022)

 

[soresu]

What he was saying is that we'll be on some sort of 5nm+ in 2021, not that ARM cores will not eventually move to 3nm when it is available (in 2022)

Well obviously no one can fab on a node that isn't even running yet.

Besides which he was talking next gen, not A78 and X1 which will be out next year. Earliest we will see Matterhorn in silicon might be HiSilicon in early Q2 2021, so it's entirely possible that Matterhorn and the other nu big core could be on 3nm.

 

[Tabalan]

Nvidia Carmel had average performance (Cortex A75 competitor), at least it seemed like it from Anandtech review. Efficiency was laging behind competitors, but it was partially due to worse process node (FFN 12).

TSMC is expected to hit HVM in 2022, so after next gen uarch cores. Link:

Report: TSMC To Start 3nm Volume Production In 2022

TSMC fab plans are as aggressive as ever.

www.tomshardware.com

 

[Tabalan]

Ahh, didn't refresh thread before posting previous message.

I believe A78/X1 will be released this year with latest HiSilicon Kirin 1020. The leaks about +50% CPU perf would indicate Cortex X1.
On the other hand, next gen might be used in 2022 if Huawei problems (China vs USA trade war) will prolong as Huawei is the only manufacturer who releases SoC the same year as ARM unveils new uarch. Samsung and Qualcomm do it next year (well, with Qualcomm we have test devices, but these are far from final product).

 

[Abwx]

Can't wait for Nvidia to launch a Tegra A57 using the Cortex X1.

Well....

https://twitter.com/x/status/1267020683072229379

 

[thigobr]

That's very interesting... I would love a SBC based on this SoC

 

[Gideon]

That's very interesting... I would love a SBC based on this SoC

Well....

https://twitter.com/x/status/1267020683072229379

VideoCardz outright mentioned it was fake. Ian Cutress also strongly hinted (arm hasn't used ARM Cortex with ARM all caps as a tradmark for a loooong time)

 

[Roland00Address]

Well....

Extremely confident this is fake.

That said I wonder if I would want that device running some form of windows (using emulation) like Windows on Snapdragon. Or just give up on that dream and hope for a good Android / Chromebook port.

 

[Gideon]

This fake-leak got me thinking of a really dumb idea (suggested in the past but not quite like this) What If AMD designed a hybrid Arm/x86 design for ~9W fanless designs running only Windows on ARM.

It could have:

• 4x A55 small cores
• 2x A78 big cores
• 2x Zen 3 cores for x86 and x64 apps (instead of emulating as ARM currently does)

MT performance would be really lackluster (though probably not worse than Lakefield) but it would have full compatiblity for x86 apps and inherit Arm's order of magnitude better battery life

Just a stupid idea not to take too seriously, but IMO if Lakefield makes sense to some companies, this wouldn't be totally insane, just 99%

 

[Tabalan]

@Gideon
Not big fan of such solution unless you can somehow run OS on A55 cores in background, while x86 apps would use Zen cores same time.

Btw, is it possible to create x86 ARM hybrid core?


Also, one more thing. ARM announced Neoverse N1 (server core based upon Cortex A76) Feb 2019. Still no signs of N2, such a bummer. X1 would be such fine base for new, more powerful core for servers. Would love to see comparison of such core vs Icelake-SP vs AMD Rome/Milan.

 

[moinmoin]

Zen is already a hybrid ARM/x86 design, the ARM part is just a "Secure Processor" (and likely part of the SCF).

 

[soresu]

Still no signs of N2, such a bummer. X1 would be such fine base for new, more powerful core for servers.

Hera/X1 probably is the base for N2.

Hera being the wife of Zeus (N2) as Enyo (A76) was the wife of Ares (N1). This could be entirely coincidental naming, but I doubt it.

N2 was also promised to have at least a 30% boost from N1, which an A77 base will be hard pushed to do.

The IPC jump from an A76 to X1 base is also similar to the jump shown from the "Cosmos" platform based on A72 to the A76 based N1 on their original Neoverse information dump:

 

[Roland00Address]

This fake-leak got me thinking of a really dumb idea (suggested in the past but not quite like this) What If AMD designed a hybrid Arm/x86 design for ~9W fanless designs running only Windows on ARM.

Probably not enough volume.

 

[dark zero]

This fake-leak got me thinking of a really dumb idea (suggested in the past but not quite like this) What If AMD designed a hybrid Arm/x86 design for ~9W fanless designs running only Windows on ARM.

It could have:

• 4x A55 small cores
• 2x A78 big cores
• 2x Zen 3 cores for x86 and x64 apps (instead of emulating as ARM currently does)

MT performance would be really lackluster (though probably not worse than Lakefield) but it would have full compatiblity for x86 apps and inherit Arm's order of magnitude better battery life

Just a stupid idea not to take too seriously, but IMO if Lakefield makes sense to some companies, this wouldn't be totally insane, just 99%

Actually VIA tought on that, but got discarded.

VIA's new Isaiah x86/ARM hybrid CPU outperforms Intel in benchmarks - but will it ever come to market?

VIA is working on a new CPU design with an unveil planned for late August -- could this core be a combined ARM-x86 hybrid?

www.extremetech.com

Returning to the topic, maybe AMD might unveil 2 chips but one with colaboration with Samsung to power the Galaxy S20 (fabbed in Samsung) and another one to test luck on the Gaming phones (fabbed in TSMC)? would be plausible to see that.

I can see Lenovo, Asus, Xiaomi and Nubia joining AMD just before Sony ask AMD to join them to a new edition of Xperia Play with said chip.

Ok, that headcanon from mine would be too much, but just imagine an Xperia Play with that kind of AMD Chip (the C7 one), with Sony optimization it would run FULL PS3 games flawlesly! and even run some PS4 games (that would be too much). Sony would sell it by 700 dollars and well, get a true console experience on a phone.

 

[IntelUser2000]

This fake-leak got me thinking of a really dumb idea (suggested in the past but not quite like this) What If AMD designed a hybrid Arm/x86 design for ~9W fanless designs running only Windows on ARM.

It won't work. Even Lakefield having two x86 core types forgo the higher level instruction set(AVX in SNC) and Hyperthreading just to maintain instruction equivalence.

Doing that with a totally different ISA is absurd. Emulation isn't just about performance loss - it results in compatibility loss. That's why even with Win32 emulation certain applications refuse to run on WoA.

And its besides the ISA why ARM platforms get better battery life. If AMD did that they'll end up with an incompatible CPU with regular battery life. Compared to that, a Bulldozer v2 in a Tablet seems a better option.

 

[CHADBOGA]

Well....

https://twitter.com/x/status/1267020683072229379

Hans is frequently spotted wearing a red nose, wig and clown shoes.


Don't name call.

AT Moderator ElFenix

 

[Gideon]

It won't work. Even Lakefield having two x86 core types forgo the higher level instruction set(AVX in SNC) and Hyperthreading just to maintain instruction equivalence.

Doing that with a totally different ISA is absurd. Emulation isn't just about performance loss - it results in compatibility loss. That's why even with Win32 emulation certain applications refuse to run on WoA.

And its besides the ISA why ARM platforms get better battery life. If AMD did that they'll end up with an incompatible CPU with regular battery life. Compared to that, a Bulldozer v2 in a Tablet seems a better option.

I never thought of it as a serious idea, but you seem to have missed the point a bit. I was thinking of essencially an arm chip running only WoA. I certainly never meant that any process would ever migrate between arm or x86 (thus instruction compatiblity comparison is moot). It's still a pure WoA experience, just that x86 cores are used for "emulation".

E.g. all windows processes will run on A55 small cores and everyrthing that can, will run natively on ARM cores (browsers, windows apps getting ARM support). The x86 cores will only run apps that would otherwise use emulation. That shouldn't really hit life in most average-joe usecases: while browsing, watching video, doing most of light stuff as Office 365, Browsers and electron-based apps (spotify, skype, slack) all have native arm support.

x86 cores will only kick in with apps that would otherwise use emulation (which still sucks and only suports 32-bit). It's far from ideal, MT performance would suck (compared to similar sized all-arm or all-x86 design), x86 cores would probably need their own memory space (adding to memory requirements) etc ...

The only reason it came to mind was, because AMD would be kinda uniquily situated to do such a device as they have x86 license (I guess VIA could do but their x86 is still abysmal) and Intel (when hell freezes over), but realistically it's only AMD.

Overall I agree, even if it would theoretically work, it would still be kinda slow and useless, take up a ton of HW and SW effort to make it work and that certainly would be better spent elswhere. It's still afun thing to speculate though.

 

[Tabalan]

Hera/X1 probably is the base for N2.

Hera being the wife of Zeus (N2) as Enyo (A76) was the wife of Ares (N1). This could be entirely coincidental naming, but I doubt it.

N2 was also promised to have at least a 30% boost from N1, which an A77 base will be hard pushed to do.

The IPC jump from an A76 to X1 base is also similar to the jump shown from the "Cosmos" platform based on A72 to the A76 based N1 on their original Neoverse information dump

Good point with wifes, didn't catch that one before. Tho, Cortex A78 should provide this 30% perf advantage over A76. Look at estimated A78 performance vs Kirin 990 (finest A76?). Cortex X1 should give more substantial perf boost, closer to 50-60%.

Spoiler: Chart

@moinmoin
I don't think ARM core used solely for security makes Ryzen hybrid. To me, this ARM core is more of ASIC, built in with single purpose.

 

[moinmoin]

I don't think ARM core used solely for security makes Ryzen hybrid. To me, this ARM core is more of ASIC, built in with single purpose.

An ARM/x86 hybrid won't and can't ever be a seamless hybrid like big.LITTLE or Lakefield. And the ARM core in Zen very likely is not for a single purpose (afaik it's also responsible for POST etc.), just its public access is gated.

 

[Gideon]

An ARM/x86 hybrid won't and can't ever be a seamless hybrid like big.LITTLE or Lakefield. And the ARM core in Zen very likely is not for a single purpose (afaik it's also responsible for POST etc.), just its public access is gated.

Why not, if the chip is a normal AoW ig.LITTLE design that only fires x86 cores up exclusively for tasks normally emulated on Arm. These tasks would remain on zen cores and have their own memoryspace.

I don't think this is a very good idea, I just don't get technical reasons why it couldn't really be done or why would it be so taxing battery life as some claim, if those processes never migrate between different core-architectures.

EDIT:
To further clarify what I meant (as people constantly seem to ignore it)

Don't think of it as a hybrid chip. Think of it more as a totally normal run-of-the-mill Arm big.LITTLE design (A55 + A78) with a dedicated x86 accelerator for emulation.

It's far from technically accurate but perhaps gets the point across.

 

[DrMrLordX]

Why not, if the chip is a normal AoW ig.LITTLE design that only fires x86 cores up exclusively for tasks normally emulated on Arm. These tasks would remain on zen cores and have their own memoryspace.

I don't think this is a very good idea, I just don't get technical reasons why it couldn't really be done or why would it be so taxing battery life as some claim, if those processes never migrate between different core-architectures.

I think a hybrid ARM/x86 SoC would require an entirely seperate bus/mesh and "uncore", as well as its own cache structure. You would be slapping on an x86 CPU via a crossbar or something. Not quite as laggy as putting something like Phi in a PCIe slot, but conceptually the same thing, albeit with the potential to (maybe) share a memory controller with the ARM side of the SoC.

 

[Gideon]

I think a hybrid ARM/x86 SoC would require an entirely seperate bus/mesh and "uncore", as well as its own cache structure. You would be slapping on an x86 CPU via a crossbar or something. Not quite as laggy as putting something like Phi in a PCIe slot, but conceptually the same thing, albeit with the potential to (maybe) share a memory controller with the ARM side of the SoC.

Yeah, the more you actually start to think how it would work, the more it seems to suck

 

[IntelUser2000]

Differentiating between two types - one on emulation and the other not, will be extremely difficult to the point of not being practical. It also introduces another uncertainty on top of emulation.

 

[AkulaMD]

That's very interesting... I would love a SBC based on this SoC

It would be awesome. It can also find its way to thin & light laptops.

 

[soresu]

To me, this ARM core is more of ASIC, built in with single purpose.

It forms part of a system which provides the secure running environment, it likely uses a firmware/BIOS to execute that environment and not an ASIC.

It's still a general purpose microprocessor and can be modified in it's use depending on whatever the firmware requires of it, that it is being used for a specific purpose does not make it an ASIC.

Tho, Cortex A78 should provide this 30% perf advantage over A76. Look at estimated A78 performance vs Kirin 990 (finest A76?). Cortex X1 should give more substantial perf boost, closer to 50-60%.

I already mentioned that 50-60% A76 to X1 boost when I said it was similar to the Cosmos -> N1 IPC uplift.

Under some SIMD heavy loads such as encoding video the X1 core should near double the performance of A77, let alone A76 when you account for the doubling of execution units from 2 to 4.

Considering the sheer amount of video processing that is necessary today in servers that isn't a small thing at all - I'm sure there are various other uses for heavy SIMD code in servers too.