Discussion Future ARM Cortex + Neoverse µArchs Discussion

Page 22 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

soresu

Diamond Member
Dec 19, 2014
3,273
2,549
136
When will ARM cores go beyond the standard 128b vector length?
Standard off the shelf IP?

Who knows, maybe V4 will have it.

As for non off the shelf ARM cores Fujitsu's A64FX core designed for a supercomputer system has 2x 512 bit SVE units.

Its successor Monaka due in 2027 is likely to follow a similar design, possibly up to 4x 512 bit SVE2 units going by their 2x perf projection.
 

SarahKerrigan

Senior member
Oct 12, 2014
735
2,035
136
Standard off the shelf IP?

Who knows, maybe V4 will have it.

As for non off the shelf ARM cores Fujitsu's A64FX core designed for a supercomputer system has 2x 512 bit SVE units.

Its successor Monaka due in 2027 is likely to follow a similar design, possibly up to 4x 512 bit SVE2 units going by their 2x perf projection.

2x is compared to "2027 competitor", not A64FX. Monaka is less specifically HPC-focused than A64FX is, and I wouldn't be surprised if Fujitsu does not increase per-core vector throughput - and I could even see them reducing it.
 

Nothingness

Diamond Member
Jul 3, 2013
3,134
2,145
136
2x is compared to "2027 competitor", not A64FX. Monaka is less specifically HPC-focused than A64FX is, and I wouldn't be surprised if Fujitsu does not increase per-core vector throughput - and I could even see them reducing it.
During SC22 Fujitsu showed this: https://www.fujitsu.com/global/products/computing/servers/supercomputer/topics/sc22/

One of documents talks about Monaka. Another one talks about "Next-generation Supercomputing Infrastructures". The latter doesn't mention Monaka but talks about "Fugaku-Next" around 2030.

So I agree it's unlikely Monaka will have wider vectors and I bet as you they will reduce the width to put more cores on a single die.
 

SarahKerrigan

Senior member
Oct 12, 2014
735
2,035
136
During SC22 Fujitsu showed this: https://www.fujitsu.com/global/products/computing/servers/supercomputer/topics/sc22/

One of documents talks about Monaka. Another one talks about "Next-generation Supercomputing Infrastructures". The latter doesn't mention Monaka but talks about "Fugaku-Next" around 2030.

So I agree it's unlikely Monaka will have wider vectors and I bet as you they will reduce the width to put more cores on a single die.

Monaka is a stacked package with all 144 cores on one die and a large cache and I/O on another. It's also going to arrive earlier than Fugaku-Next (which afaik is not even confirmed to have a Fujitsu ARM in it!) and is explicitly targeting a datacenter and cloud role, not just an HPC role.

EDIT: I reviewed the docs and it's ten dice, not two - four for cores, four for SRAM, one for I/O, one for interposer. Mea culpa.
 
Last edited:

soresu

Diamond Member
Dec 19, 2014
3,273
2,549
136
2x is compared to "2027 competitor", not A64FX. Monaka is less specifically HPC-focused than A64FX is, and I wouldn't be surprised if Fujitsu does not increase per-core vector throughput - and I could even see them reducing it.
Reducing to less than 2x 512 bit?

Intel is already there or beyond that for some time now in the server SKUs.

Anyone who wants less can just go for off the shelf Neoverse IP.

While I agree it's unlikely that Fujitsu will go to 1,024 bit units I don't see them reducing to 256 bit units when the competition at the lower end is so well established, and I don't see Neoverse V staying at 128 bit indefinitely if ARM have any intentions of eventually snagging more supercomputer contracts, so Fujitsu will need something to differentiate themselves from the OTS IP.
 

SarahKerrigan

Senior member
Oct 12, 2014
735
2,035
136
Reducing to less than 2x 512 bit?

Intel is already there or beyond that for some time now in the server SKUs.

Anyone who wants less can just go for off the shelf Neoverse IP.

While I agree it's unlikely that Fujitsu will go to 1,024 bit units I don't see them reducing to 256 bit units when the competition at the lower end is so well established, and I don't see Neoverse V staying at 128 bit indefinitely if ARM have any intentions of eventually snagging more supercomputer contracts, so Fujitsu will need something to differentiate themselves from the OTS IP.

Fujitsu seems to think they have differentiators that aren't related to vector width. They've been upfront about Monaka being a general-purpose server CPU. The fact it takes a large step back in terms of memory bandwidth compared to A64FX is a strong hint to how it's being positioned.
 

SpudLobby

Senior member
May 18, 2022
991
682
106
Hoho, now that I have the numbers (PPC), I think I can interpret what he said:

  • Qualcomm's custom core is basically the same as Apple A14 in terms of IPC, which is slightly better than Cortex-X4: Yes, my number aligns with what he said, see the table above with A14's IPC
Yeah A14 IPC is right. A17 and M3 have about 12-14% more IPC even at their higher frequencies.
  • Cortex-X5's IPC will surpass Apple, according to the leaked score of 1930 GB5 and 2700 GB6 @ 3.35GHz: Yes, D9400's IPC/PPC is 5% higher than A17 Pro.
If that’s true then I wonder what power looks like.

Usually wider, smarter cores (not always but assuming they were done correctly) are more efficient for the same performance but it’s always possible Apple has them matched on power — the X5 in the D9400 rumor was not good on that note — albeit for more performance (2900 GB6 can be found). Would be a rare case but it absolutely is possible especially given Apple’s memory hierarchy and L2/SLC sizes.
  • If X5 manages power well, Nuvia can only explode to surpass: With lowest clock speed, I am not too worried about thermal power. And Phoenix core is really behind Cortex-X5 in ST, even you OC, it is still the same IPC, Qualcomm has to hope for Pegasus core to increase IPC...
Yeah I mean if Arm pulls something off here where they match Apple IPC AND get their power under control, even if it’s a huge area hit which is worthwhile, then it would be almost comical timing RE: Phoenix core
 
Reactions: Tigerick

Shivansps

Diamond Member
Sep 11, 2013
3,875
1,530
136
Looks like we have the first A78 SBC

Too bad it is kind of a bad option with the Mali G57 and no PCIE on the board.
 

Tigerick

Senior member
Apr 1, 2022
700
615
106
Yeah A14 IPC is right. A17 and M3 have about 12-14% more IPC even at their higher frequencies.

What's your IPC numbers come from? SPECint?

If that’s true then I wonder what power looks like.

Usually wider, smarter cores (not always but assuming they were done correctly) are more efficient for the same performance but it’s always possible Apple has them matched on power — the X5 in the D9400 rumor was not good on that note — albeit for more performance (2900 GB6 can be found). Would be a rare case but it absolutely is possible especially given Apple’s memory hierarchy and L2/SLC sizes.

Yeah I mean if Arm pulls something off here where they match Apple IPC AND get their power under control, even if it’s a huge area hit which is worthwhile, then it would be almost comical timing RE: Phoenix core
D9300's Cortex-X4 comes with high power because they are made by N4X process based on Wiki. Now we know why Mediatek insists of 4P+4E with HPC process. Mediatek is treating D9300 as PC SoC and it is all because of WoA....

That's why I said D9400 with N3E would be more power efficient than D9300...
 
Last edited:

FlameTail

Diamond Member
Dec 15, 2021
4,094
2,465
106
D9300's Cortex-X4 comes with high power because they are made by N4X process based on Wiki.
N4X? Plausible. That explains Mediatek's claim that D9300 is fabbed on '3rd generation 4nm process'.

Still, that's not only reason for Cortex X4's high power consumption.

X4 in 8G3, which is fabbed on N4P, also has high power consumption.

 

SpudLobby

Senior member
May 18, 2022
991
682
106
What's your IPC numbers come from? SPECint?


D9300's Cortex-X4 comes with high power because they are made by N4X process based on Wiki. Now we know why Mediatek insists of 4P+4E with HPC process. Mediatek is treating D9300 as PC SoC and it is all because of WoA....

That's why I said D9400 with N3E would be more power efficient than D9300...
It is not N4x. N4X is extremely leaky. It’s N4P.

And the X4 power isn’t terrible relative to like AMD and Intel PC stuff, still a good mobile core it’s just not great since even the A16 beats it nontrivially on spec with 10-15% less power.
 

FlameTail

Diamond Member
Dec 15, 2021
4,094
2,465
106
It is not N4x. N4X is extremely leaky. It’s N4P.
Then what on earth is "3rd generation 4nm process" that Mediatek says is used for D9300?

If D9300 is indeed N4P, the alternative theory is this:
D9000 : N4E
D9200 : N4
D9300 : N4P
 

Hitman928

Diamond Member
Apr 15, 2012
6,328
11,123
136
Then what on earth is "3rd generation 4nm process" that Mediatek says is used for D9300?

If D9300 is indeed N4P, the alternative theory is this:
D9000 : N4E
D9200 : N4
D9300 : N4P

Probably counting N5 since they are all in the same node family. It doesn't really matter though, it's just a useless marketing line.
 

FlameTail

Diamond Member
Dec 15, 2021
4,094
2,465
106
Will we ever see an ARM powered console anytime soon?

As in normal consoles (xbox/ps form factor) or handhelds (steam deck-like).
 

Doug S

Platinum Member
Feb 8, 2020
2,833
4,819
136
Expect to see one from Apple using their unsold base 8GB/256GB M1/M2 chips if they don't bury them in a landfill...

Which unsold chips would those be? Apple keeps older products on the price list for a while, so they are never going to end up with unsold chips. Worst case if they did they could stick them in the next update of Apple TV. Since it supports games I guess if you want to call that a "console" you could.
 
Reactions: Tlh97 and SpudLobby

FlameTail

Diamond Member
Dec 15, 2021
4,094
2,465
106
They need to make a console with the M Ultra chip...

[Realisation]

Oh wait, the Mac Studio exists...
 

SpudLobby

Senior member
May 18, 2022
991
682
106
Apple really just doesn’t care much for games, and while the recent MacOS update with the porting toolchain and up scaling would seem to prove otherwise I think that’s a very low bar relative to Apple’s warchest.
 

Doug S

Platinum Member
Feb 8, 2020
2,833
4,819
136
Apple really just doesn’t care much for games, and while the recent MacOS update with the porting toolchain and up scaling would seem to prove otherwise I think that’s a very low bar relative to Apple’s warchest.

They care for games just fine, in fact they are the biggest gaming company on the planet. But people here don't consider mobile gaming to be "real" gaming so they don't count that revenue, they only want to count PC and console and since Mac gaming is tiny and Apple doesn't have a console this mistaken belief that "Apple doesn't care much for games" persists.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |