Discussion Future ARM Cortex + Neoverse µArchs Discussion

Page 32 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

LightningZ71

Golden Member
Mar 10, 2017
1,658
1,940
136
It's a far cry from what it was decades ago when I was getting my Comp Eng degree... Granted, these FPGAs are FAR more capable than what I was fooling with...
 

NTMBK

Lifer
Nov 14, 2011
10,269
5,134
136
Idk man. Mongoose cores were using more die area than Apple's P cores, while having similar performance to ARM's A7x cores and worse efficiency. In fact, Mongoose M6, the 2021 core (which was cancelled), was rumoured to add SMT.

SMT in phones!

Edit: corrected to M6
Sure, but now they're stuck with jacked up licensing costs, and they're up against Qualcomm who are shipping custom cores in laptops. If Samsung had kept improving their design then that could be them (and shipping chips fabbed in their own factories, in their own laptops).

A design team given funding and time can turn things around. Just look at the jump from Bulldozer to Zen.
 
Reactions: Tlh97 and hemedans

NostaSeronx

Diamond Member
Sep 18, 2011
3,688
1,222
136
Sure, but now they're stuck with jacked up licensing costs, and they're up against Qualcomm who are shipping custom cores in laptops. If Samsung had kept improving their design then that could be them (and shipping chips fabbed in their own factories, in their own laptops).

A design team given funding and time can turn things around. Just look at the jump from Bulldozer to Zen.
The full development of Bulldozer started in like 1997 till its tapeout image release in 2010. Zen only took what two~three years to get near its launch design.

Pre-Bulldozer/K8 Project 2 = 1997-2002 (Single-threaded Clustered heavily related to K5/K6 cores: two K5/K6 cores in a single processor is quite literally the patent design)
Pre-Bulldozer/K10 = 2002-2004 (Multi-cluster Multithreading, CMP-like, single monolithic fp)
Bulldozer/K10 Project 2 = 2004-2007 (Cluster-based Multithreading, SMT-like :: Core is related to Bobcat, basically being Dual-cluster (Int and FP clusters) with a single High Perf AGU/LD/ST :: 4/3/4, was basically JK's/MA's K8p2 and K9p2 low-power/ultra-mobile projects)
Bulldozer/K10 Project 3 = 2007-2009 (Reverted back to CMP-like, single monolithic fp, with mobile/low power optimization not as big as K10p2)

DW's design was dropped for FW's high-frequency design. RM's design was dropped for AW's/MB's design by DM's/HR's push for high-frequency over mobile IPC from PH's direction. Hence, why DM replace PH and DM replaced HR, why DM left as he miscalled where the market headed.
 
Last edited:

hemedans

Senior member
Jan 31, 2015
207
102
116
Did you forget that Cortex X series exists.

Samsung ditched Mongoose M, the same year ARM unveiled Cortex X. They simply couldn't keep up.

And Samsung's implementation with their custom cores was rather funky. They still used ARM's Cortex cores as the mid and littles, despite having their own custom Mongoose M cores- which were used as Primes.
View attachment 99979
The Mongoose cores had a seperate L3 cache from that of the Cortex cores.

Edit: Below is the diagram for the Exynos 990. The above diagram is that of it's predecessor, the Exynos 9820.

View attachment 99980
Can you believe that the A55 cores in the 9820 didn't have an L1 cache?😭

Cortex X1 and X2 was underwhelming, they used lot of power while perfomance was so and so,



A78 which was one generation less than X2 was still more efficient, but for some reasons all A78 successor barely had any improvement. If Samsung improved on A78 by delivering competent sub 2W they would be relevant even today, Middle core is workhorse of Android cpu and they are not good enough.

Also Mongoose weren't always bad if you look at 9820 it was preety good vs A75
 
Reactions: Tlh97

soresu

Platinum Member
Dec 19, 2014
2,939
2,162
136
Not the Snapdragon X Elite Devkit, well not really.
They don't have much choice there - they need the whole industry to bite their hands off on this to make the Nuvia purchase worth it.

Give it a generation or 2 of success and you won't find them so generous.
 

FlameTail

Diamond Member
Dec 15, 2021
3,137
1,790
106
This leaker is saying that Dimensity 9400's Cortex X925 may reach 3.8 GHz.


It's an interesting story.

Initially Qualcomm's Snapdragon 8 Gen 4 chip was rumoured to do about 2800 points in Geekbench 6, and the Dimensity 9400 also a similar about 2700 points.

But then, last month Qualcomm did a 2nd takeout of 8G4, increasing the clock speed to 4.2 GHz and increasing the score to 3150 points.

Now Mediatek has gotten wind of this, and it seems they are also going to increase the clock speed of the final D9400 product to compete with Qualcomm.
 

gdansk

Platinum Member
Feb 8, 2011
2,478
3,373
136
What if both increased power to get a few hundred MHz more? 🤔
Or the initial rumors were all bupkes?
Or the new rumors are all made up?
 

Doug S

Platinum Member
Feb 8, 2020
2,477
4,033
136
Don't the OEMs control the actual clock speed? Just because Qualcomm/Mediatek announce benchmarks at a certain clock rate that doesn't mean OEMs will all use those clock rates. In the past haven't some OEMs clocked faster than the announced speeds, and others slower, depending on whether they want to push out a narrow performance advantage over the competition at the cost of running hot or want to market more on long battery life?

There's also binning. Both Qualcomm and Mediatek have multiple customers, so they could bin out the ones able to run at a higher clock rate and sell them at a higher price to those aggressive OEMs.
 

SteinFG

Senior member
Dec 29, 2021
519
610
106
Don't the OEMs control the actual clock speed?
They control how much power it gets, but I'm pretty sure clock tables are locked on those chips.
There's also binning. Both Qualcomm and Mediatek have multiple customers, so they could bin out the ones able to run at a higher clock rate and sell them at a higher price to those aggressive OEMs.
That's exactly what "8 Gen 2 for galaxy" was. Did they make a similar thing for 8 Gen 3 or not? I don't keep up with it
 

soresu

Platinum Member
Dec 19, 2014
2,939
2,162
136
The current CPU implementation of the ISA often lags behind the published one by a couple of years or more.

v9.4-A was announced in September 2022, which means it was already announced when the first implementations of v9.2-A was released in the 2023 IP stack.

v9.3-A was announced a year earlier, I can't find the exact date for v9.2-A because the only applicable blog entry on ARM's website doesn't actually mention it by name.
 
Reactions: Tlh97 and FlameTail

FlameTail

Diamond Member
Dec 15, 2021
3,137
1,790
106
The current CPU implementation of the ISA often lags behind the published one by a couple of years or more.

v9.4-A was announced in September 2022, which means it was already announced when the first implementations of v9.2-A was released in the 2023 IP stack.

v9.3-A was announced a year earlier, I can't find the exact date for v9.2-A because the only applicable blog entry on ARM's website doesn't actually mention it by name.
The recently announced 2024 IP is also seemingly sticking to ARMv9.2, which is intriguing.

Is there a possibility that Oryon V2 could use ARM v9.4? I have a thought that with lawsuit between them, ARM won't be willing to give an ARMv9.4 license to QC.
 

soresu

Platinum Member
Dec 19, 2014
2,939
2,162
136
The recently announced 2024 IP is also seemingly sticking to ARMv9.2, which is intriguing.
Less intriguing than expected.

First 2 generations of v9 IP used v9.0-A, and now at least 2 after that use v9.2-A.

It wouldn't surprise me if 2025 IP uses v9.4-A but I wouldn't bet on it.
Is there a possibility that Oryon V2 could use ARM v9.4? I have a thought that with lawsuit between them, ARM won't be willing to give an ARMv9.4 license to QC.
It wouldn't surprise me if they stick to v8-A for as long as they can if µArch license for it costs less than v9-A.
 

FlameTail

Diamond Member
Dec 15, 2021
3,137
1,790
106
It wouldn't surprise me if they stick to v8-A for as long as they can if µArch license for it costs less than v9-A
Sure, but they are missing out on a bunch of ARMv9 features ; SME, SVE, SSVE, MTE, etc...

The addition of SME alone gave the M4 a 10% uplift in Geekbench 6 ST. That's a lot of performance that Nuvia would be leaving on the table.
 
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/    |