Discussion AMD's Future APU Gone ARM?

Page 7 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

FlameTail

Platinum Member
Dec 15, 2021
2,239
1,217
106
Yes. Even if say for example, AMD made a smartphone SoC using the Cortex X4, A720, A520 cores, on the same N4P process, identical core count and clock speed as Snapdragon 8G3.

That AMD SoC wouldn't be as efficient as the Snapdragon 8G3. Aside from the core design, there are other things that go into an SoC. The likes of Apple and Qualcomm have mastered these other things, including power saving measures which provide them an efficiency advantage.
 
Last edited:

Thibsie

Senior member
Apr 25, 2017
749
800
136
AMD doesn't care about smartphone market.
I don't think it would be that hard for them to match QCnif they give it time and resource.
 

Doug S

Platinum Member
Feb 8, 2020
2,254
3,492
136
AMD doesn't care about smartphone market.
I don't think it would be that hard for them to match QCnif they give it time and resource.

They don't CURRENTLY care about the smartphone market, but why should they ignore it IF they plan to do custom ARM designs for the PC? Wouldn't it make sense to look into providing solutions for the smartphone?

Their big hurdle for trying to sell into the smartphone market would be the lack of a modem, something that doesn't matter for the PC market.
 
  • Like
Reactions: Tlh97

poke01

Senior member
Mar 8, 2022
725
699
106
They don't CURRENTLY care about the smartphone market, but why should they ignore it IF they plan to do custom ARM designs for the PC? Wouldn't it make sense to look into providing solutions for the smartphone?

Their big hurdle for trying to sell into the smartphone market would be the lack of a modem, something that doesn't matter for the PC market.
Not to mention the ISP and video encoders and decoders.
 

Doug S

Platinum Member
Feb 8, 2020
2,254
3,492
136
Not to mention the ISP and video encoders and decoders.

Pretty sure AMD already has video encoders and decoders as good as anyone's, and they could probably license an ISP from someone like Google or Samsung. Apple was able to develop all that stuff internally with no trouble, but a modem has presented more of a problem.
 

moinmoin

Diamond Member
Jun 1, 2017
4,944
7,656
136
AMD collaborated with MediaTek for an AMD branded wifi module, not sure if that way just a one time thing though. If not that could be expanded. As for ISP, video encoders and decoders Xilinx should be able to offer those if the current Radeon state is not considered sufficient. Xilinx would also likely be the department leading development of an ARM SoC as they have more experience with that already.
 
  • Like
Reactions: Thibsie

Thibsie

Senior member
Apr 25, 2017
749
800
136
They don't CURRENTLY care about the smartphone market, but why should they ignore it IF they plan to do custom ARM designs for the PC? Wouldn't it make sense to look into providing solutions for the smartphone?

Their big hurdle for trying to sell into the smartphone market would be the lack of a modem, something that doesn't matter for the PC market.
They can't get their Ryzen Mobile on the market and they'd want to attack smartphone market where timing and execution are are of first importance ?
I don't think so. Not yet at least.

What about margins in this market ? Going after Qualcomm isn't gonna be simple at all although competition would be nice of course.

If they get well established in the ARM PC (maybe tablet?) market, they might want to stretch it but I doubt it.

Then, maybe they wanna buy Mediatek, lol.
 

Doug S

Platinum Member
Feb 8, 2020
2,254
3,492
136
They can't get their Ryzen Mobile on the market and they'd want to attack smartphone market where timing and execution are are of first importance ?
I don't think so. Not yet at least.

Well I would argue bothering with ARM at all is dumb if they aren't fully exploiting their potential opportunities in the PC/server/GPU/AI market that they already know well. But what I was responding to was the scenario where AMD was going to introduce custom ARM designs to challenge Qualcomm in the "ARM PC" market. IF they do that, jumping into the mobile market "only" requires a modem.
 

Thibsie

Senior member
Apr 25, 2017
749
800
136
Well I would argue bothering with ARM at all is dumb if they aren't fully exploiting their potential opportunities in the PC/server/GPU/AI market that they already know well. But what I was responding to was the scenario where AMD was going to introduce custom ARM designs to challenge Qualcomm in the "ARM PC" market. IF they do that, jumping into the mobile market "only" requires a modem.
Technically, yes.
But they would need to come with a new phone SOC at very regular interval to suit new phones releases. Hard task IMO.
 

eek2121

Platinum Member
Aug 2, 2005
2,930
4,026
136
AmpereOne has been reportedly sampling since May 2022, launched in May 2023, and agot dopted by Oracle in Sep 2023. Yet, AFAIK there are no 3rd party reviews available nor is the Oracle A2 instance.

AMD stated they are ready to adopt ARM at any time it makes sense.

ZenX branding will surely stop at some point. I mean, the brand will be nearly 10 yo with Zen 6.
Intel has used Core branding even longer. Sometimes you don’t need to needlessly rename.

The Core branding may possibly drop from premium products at some point soon-ish in the future.
ARM or RISC-V doesn't inherently bring you power efficiency.

Apple cpus are superior simple because they are well designed and well made; ISA doesn't make that much of a lift.

I believe AMD will make arm cpus when customers demand it.

Given how modular ZEN architecture is, AMD could reuse most of the parts with minimal manpower and time.
Some don’t understand that x86 can be every bit as efficient as ARM (RISC-V already is).

Apple released a core on a cutting edge process that was performant and efficient. People suddenly think ARM is better. (it isn’t in terms of performance or compatibility) — Note that they haven’t been able to much improve on the design since launch.

Chip design is all about compromise. Apple built a mobile chip and scaled it up.

Intel and AMD long ago began designing a single core design centered around the desktop and server at a time when mobility was not as important (desktops used to be more popular than laptops, and smartphones did not exist, so this approach made sense. There has been no need to change (except mobile) and no desire to do it.

Atom began as a a bit of a redesign for mobile and scaled from there, but targeted density. Note (and this is serious IIRC material folks, I am getting too old to keep all this in my head) that when Atom first launched on mobile, it was FASTER than ARM chips that existed at the time, but had worse efficiency.

AMD/Intel have designs in the pipeline that make the M1 look like a toy, but CPU design takes time and you won’t really see the results until next year or maybe the year after. Note: power consumption will be a constant as long as chip designs scale that high.
Pretty sure AMD already has video encoders and decoders as good as anyone's, and they could probably license an ISP from someone like Google or Samsung. Apple was able to develop all that stuff internally with no trouble, but a modem has presented more of a problem.
AMD has the best gfx IP in the business. They just haven’t bothered licensing it out broadly or building their own SoC.
They can't get their Ryzen Mobile on the market and they'd want to attack smartphone market where timing and execution are are of first importance ?
I don't think so. Not yet at least.

What about margins in this market ? Going after Qualcomm isn't gonna be simple at all although competition would be nice of course.

If they get well established in the ARM PC (maybe tablet?) market, they might want to stretch it but I doubt it.

Then, maybe they wanna buy Mediatek, lol.
I think we will see a mobile SoC from AMD eventually. I spoke to folks (engineers and a marketing person at a couple unrelated conferences) at the company in the past and all expressed the desire. AMD actually uses ARM quite a bit even today, and they have some interesting stuff prototyped as well.

They can only execute so fast and the AI boom has definitely impacted that.
I don't understand Samsung. They keep announcing new processes that never appear in any designs anyone sees until ages later and then in small quantities, but have breathless press releases claiming process superiority all the time.

Now they are going to claim they are first to 2nm, but they can't even ship any first gen "3nm" stuff in quantity.

I guess this is to fool shareholders? Because the general public isn't a customer of Samsung foundry, and no one is going to buy an Android phone because it advertises "fabbed by Samsung". What's their game here, how does all this deception actually pay off for them?
They do have process superiority…in South Korea. 🤣

Samsung is like discount Intel. Somewhat worse performance for a significantly better price. That was why NVIDIA used them for so long. Margins > performance.

Why did NVIDIA switch? There were a number of factors. Here are some…the slide deck is probably incredible.

1) Samsung anything is pretty far behind TSMC N4P at this point.

2) RDNA1 scared NVIDIA, so did the last iteration of Vega. They were worried that the gap from SS vs TSMC would become quite considerable and make it difficult to remain competitive.

3) They got a sweetheart deal from TSMC. They pay a low price compared to other customers (a few thousand per wafer less at one point, still significantly more than Samsung)

4) They had already used TSMC in the server space where margins weren’t as tight. It isn’t cost effective to have 2 different designs fabbed at 2 different places. Also: people were buying GeForce desktop or low cost workstation products instead of premium enterprise offerings.

5) Yields were a problem along with capacity. Samsung also reportedly increased pricing by as much as 40% for some line items between 2019 and 2023. In one noted case it was actually cheaper for one client to switch to TSMC!

Point number 4 is also why GPU prices went up so much. AI is a small part of it. NVIDIA really did not want mid-tier stuff eating their enterprise lunch so they drastically changed up their pricing structure. The 4090 could be sold for as little as $699 and still be decently profitable, for example, at least when i did the breakdown at launch.

Don’t count Samsung out yet, however. All major tech companies are in constant discussions with them:

AMD was said to be designing a new affordable APU and GPU for lower end (using Zen and Radeon IP)

EDIT: The APU in question would not
be part of the Ryzen family unless things changed (my info is old). The designs were to target the sub $500 market. Think PCIE 3.0 (fewer lanes as well), much smaller cache, fewer cores, and low clocks. It is my understanding that GloFo and Samsung were having a conversation and Intel has recently been talking to AMD as well.

Intel uses Samsung for chipsets.

Apple uses them.

Everyone that wants to design “affordable” AI chips are talking to Samsung. They can be up to 60% cheaper depending on the node, options, etc…or they were in late 2019.

Samsung’s biggest competitive threat currently is Intel. Intel is offering a significantly better process for less money.
 
Last edited:
  • Like
Reactions: NTMBK and Tlh97

Tup3x

Senior member
Dec 31, 2016
960
943
136
AMD has the best gfx IP in the business. They just haven’t bothered licensing it out broadly or building their own SoC.
Well, that's debatable. So far they are lagging behind the main competition on mobile and desktop. APU's are a bit different story since there's only NVIDIA.
 

Tigerick

Senior member
Apr 1, 2022
650
535
106
Well, we should be hearing Windows Server for ARM pretty soon...
Yeah, Windows Server 2025 for ARM64 has entered beta build 26063. That's mean Microsoft is serious about ARM server platform and preparing to launch Windows Server for ARM64. Windows Server 2025 for x86 should be launching by the end of 2024. No one know about ARM version but by the name of it, ARM edition should launch no later than end of 2025.

Anyone wants to guess who is/are major OEMs to support such a big change? :cool:
 
  • Like
Reactions: SpudLobby

Thibsie

Senior member
Apr 25, 2017
749
800
136
That would require / trigger a lot of compilation work for driver support of specific server hardware.
There might be already ARM version for Linux but I doubt ARM Windows version do exist, yet.
 

Tigerick

Senior member
Apr 1, 2022
650
535
106
That would require / trigger a lot of compilation work for driver support of specific server hardware.
There might be already ARM version for Linux but I doubt ARM Windows version do exist, yet.
Yeah, driver build-up is a massive development. However, Windows 12 ARM and Windows Server 2025 ARM should share lots of code base....
 
Last edited:

Tigerick

Senior member
Apr 1, 2022
650
535
106
Sound Wave was leaked earlier. The name sounds weird; seems unrelated to current Point codenames.

Then I realized Soundwave is one of the characters from the original Transformers. Hmm, could it be the beginning of Transformation ??? :cool:
 

eek2121

Platinum Member
Aug 2, 2005
2,930
4,026
136
That would require / trigger a lot of compilation work for driver support of specific server hardware.
There might be already ARM version for Linux but I doubt ARM Windows version do exist, yet.
I can install Windows on my ARM device TODAY. It already exists.
 

uzzi38

Platinum Member
Oct 16, 2019
2,625
5,913
146
They don't CURRENTLY care about the smartphone market, but why should they ignore it IF they plan to do custom ARM designs for the PC? Wouldn't it make sense to look into providing solutions for the smartphone?

Their big hurdle for trying to sell into the smartphone market would be the lack of a modem, something that doesn't matter for the PC market.
"Custom" is the key word here. They're being contracted to make a product that isn't part of their usual roadmaps by one of their customers. They're even using standard ARM IP for the cores, they don't have any intention on doing much more than necessary. Why would they expand on that idea further without the guarantee of a customer for said product? Like you have things backwards here, a custom product means that things like an actual roadmap are determined by the customer - it's not something AMD will be putting together for ARM SoCs. If a customer wants an ARM SoC from AMD, then they'll be presented with options for IP based off of what they have available, and they'll be paying for anything else that needs significant development resources (or doing it themselves).

AMD are already stretched thin as it is, getting into another market on top that requires development of several new IPs just isn't a sensible business decision. If they ever care about the smartphone market, it would be in many, many years. Current AMD isn't going to be even remotely interested in that idea, not even close.
 

SarahKerrigan

Senior member
Oct 12, 2014
361
519
136
Windows Server for ARM has existed internally for years at this point; Microsoft demoed it publicly back in 2017. There's not a lot of "server specific hardware" in desperate need of difficult ports. NICs and HBAs are relatively easy and generally don't need substantial porting effort from x86.

I don't know why Windows Server for ARM hasn't been released as a full-on official product, but its existence is hardly a secret and I expect it to ship publicly on the very soon side.
 

Tigerick

Senior member
Apr 1, 2022
650
535
106
Cortex-X5: Faster IPC than Apple's M3

As explained here, upcoming D9400 with Cortex-X5 core shown pretty impressive IPC uplifts compared to past generation. Even Cortex-X5 is clocked at 3.35GHz only, D9400 still able to score 2700 points in GB6's 1T. The IPC/PPC is faster than M3, and this is the game changer in IT industry. Intel has been chasing Apple for years in order to make M1 killer, they have to spend billions asking TSMC to use N3B to make Lunar Lake in order to reach ultra-mobile performance. The results are not out yet, but I doubt Lunar Lake will reach M3's IPC, we shall see...

Cortex-X5 offers chance to leapfrog Apple's M3/M4 IPC. But it is only first step; part of the reason why Intel/AMD won't be able to make low power yet high performance SoC is because Apple is using most advanced process to help maintaining high IPC with low power. And here comes another breakthrough: 64-bit LPDDR6. The combination of Cortex-X5 with 64-bit LPDDR6 will offer solid performance upgrade with low power. If the above-mentioned ARM notebooks are selling for above $1000, they won't help growing the market. NV should be working on their ARM SoC at low cost; thus they are asking Intel Foundry to make it. How about AMD, didn't AMD also getting Surface contract in 2026?
 
Last edited:

Tigerick

Senior member
Apr 1, 2022
650
535
106
AMD's Sound Wave: The Beginning of Transformation ?

AMD-Strix-Sarlak-Kracken-Sound-Wave-APUs-_2.png

The above slide was leaked few weeks ago. Sound Wave was listed as 3nm project...

Exynos2500.jpg

Shown here is upcoming Samsung Exynos 2500's diagram with AMD's RDNA-based Xclipse 950 GPU...

Could these two things are inter-related ? Or should I say AMD will transform Exynos 2500 into Sound wave?

I know it sounds weird, but remember Qualcomm relationship with Samsung Foundry before. Could AMD partner with Samsung Foundry and modify Exynos 2500 like Google Pixel SoC? AMD should know about potential of Cortex-X5 and yet they are tight up with x86 SoC. With Samsung's partnership, AMD will gain foundry process and 5G modem integration....something NV won't be able to get with Intel Foundry...

The IPC uplifts of Cortex-X5 show ARM is getting more powerful and with WoA's exclusivity ends with Qualcomm, I see almost all CPU vendors are going to join ARM platform......except Intel and that's another point :p
 
Last edited:

NTMBK

Lifer
Nov 14, 2011
10,233
5,015
136
AMD's Sound Wave: The Beginning of Transformation ?

View attachment 96595

The above slide was leaked few weeks ago. Sound Wave was listed as 3nm project...

View attachment 96596

Shown here is upcoming Samsung Exynos 2500's diagram with AMD's RDNA-based Xclipse 950 GPU...

Could these two things are inter-related ? Or should I say AMD will transform Exynos 2500 into Sound wave?

I know it sounds weird, but remember Qualcomm relationship with Samsung Foundry before. Could AMD partner with Samsung Foundry and modify Exynos 2500 like Google Pixel SoC? AMD should know about potential of Cortex-X5 and yet they are tight up with x86 SoC. With Samsung's partnership, AMD will gain foundry process and 5G modem integration....something NV won't be able to get with Intel Foundry...

The IPC uplifts of Cortex-X5 should show ARM is getting more powerful and with ARM's exclusivity ends with Qualcomm, I see almost all CPU vendors are going to join ARM platform......except Intel and that's another point :p
AMD have a fantastic architecture in Zen4 and will presumably have an even better one in Zen5. Why would they choose to pay hefty per core license fees to ARM, while also getting worse Windows compatibility? Why put their fate in the hands of their rival?
 
  • Like
Reactions: Tlh97

Shivansps

Diamond Member
Sep 11, 2013
3,851
1,518
136
AMD have a fantastic architecture in Zen4 and will presumably have an even better one in Zen5. Why would they choose to pay hefty per core license fees to ARM, while also getting worse Windows compatibility? Why put their fate in the hands of their rival?
In my mind in the only market that markes sense is in the SBC, network soft routers, low power servers, low end mobile and ultramobile, AMD always failed very badly with their own low power x86 models, right now they only have Mendocino for that and once again looks like a failure. If they are going to try something with ARM cores is for this market.

With this in mind, the license have to be cheap so im guessing maybe A710-A720 cores.
 
Last edited:
  • Like
Reactions: Tlh97

Thibsie

Senior member
Apr 25, 2017
749
800
136
The IPC uplifts of Cortex-X5 show ARM is getting more powerful and with ARM's exclusivity ends with Qualcomm, I see almost all CPU vendors are going to join ARM platform......except Intel and that's another point :p
Once upon a time, they had XScales, so who knows ?
 

soresu

Platinum Member
Dec 19, 2014
2,657
1,860
136
AMD always failed very badly with their own low power x86 models, right now they only have Mendocino for that and once again looks like a failure
I assumed Mendocino was about low cost, high volume rather than low power.

More about competing with lower end Intel SKUs in the Chromebook market and the like.
 
  • Like
Reactions: uzzi38