Question AMD Phoenix/Zen 4 APU Speculation and Discussion

[uzzi38]

I can finally make this thread.

https://twitter.com/x/status/1519669375283957760

Phoenix is indeed RDNA3. My advice to everyone: treat the old APU rumours as being out of date.

 

[Abwx]

Thread usage is binary - thread is at given time either busy or free. So one threaded core can't newer be only partially loaded. Core monitors that report those partial load just calculate load/idle ratio for given core.

This change nothing, if all threads are at 100% during 2.5% of the time we end at 2.5% CPU usage and power is 2.5% of the 25W max power available, what matters is the duty cycle over time, not that a thread is binary on an instantaneous fashion that may last 10ms every second.

Say a 8C/16T CPU that run at 2GHz/25W when all threads are fully loaded as basis for 100% CPU usage.

Excluding 25% SMT it will be at 80% when running at 2GHz with only 8T, power will go down to 20W.

At 4T/2GHz it will be at 40% and 10W, and at 4T/1.4GHz it will be at 28.5% and about 4W, wich is exactly what is occuring in the review i linked with Far Crime test.

 

[TESKATLIPOKA]

It is, because i use actual numbers of an APU used for gaming, while you re lelying on CB at 3x the game play frequency, and on a 7nm node that is half as efficient as 4nm..

You posted either 1165G7 vs 5800U video or ultrabookreview's 680M review, but they are not using 4nm process.
So where did you link to that Phoenix data, can't find It?

% of CPU usage ≠ % of power budget
20% CPU usage in the case of 5800U means 3.2 threads are loaded, that's a fact.
But this doesn't necessarily mean the cores consume only 20% of 25W. It all depends on the actual clockspeed.
If It's clocked at only 1.9GHz or below then It's <=5W, but If It's at 3GHz, 4GHz or more, then It's a lot more than just 5W.

It is clocked at 1.4GHz when set at 25W TDP and playing Far Crime, that s what is displayed in the review, at 1.4GHz a Zen 4 core use less than 1W, so it could be up to 5 cores running at this frequency and the whole thing would be at less than 5W.

Edit : At 30W and 40W the CPU still run at 1.4GHz, wich mean that almost all the added TDP go entirely to the GPU.

Ok, and how does this contradict what I wrote and you quoted?

All this time, the only thing I tried to point out is that CPU % usage is not necessarily equal to % of power budget, even though you wrote It like It does.
It doesn't matter If It's Zen3 or Zen4 core, If you for example start CB R23 single threat which loads only a single core, then power consumption will be >15W for both of them, because frequency is set very high. This of course wouldn't match with CPU % usage, but this was my point from the beginning.
Of course If the clockspeed was low then power consumption will go drastically down, I never disputed this.

 

[Abwx]

You posted either 1165G7 vs 5800U video or ultrabookreview's 680M review, but they are not using 4nm process.
So where did you link to that Phoenix data, can't find It?

If a Zen 3+ use 5W for X FPS with a 680M it s logical that a 4nm Zen 4 with 10% higher FPS from the GPU will use even less power, you are grasping at straws here..

Ok, and how does this contradict what I wrote and you quoted?

All this time, the only thing I tried to point out is that CPU % usage is not necessarily equal to % of power budget, even though you wrote It like It does.
It doesn't matter If It's Zen3 or Zen4 core, If you for example start CB R23 single threat which loads only a single core, then power consumption will be >15W for both of them, because frequency is set very high. This of course wouldn't match with CPU % usage, but this was my point from the beginning.
Of course If the clockspeed was low then power consumption will go drastically down, I never disputed this.

Your 15W are at 4.4GHz with a 7nm based Zen 3, here we re talking of a CPU that run at 1.4GHz with a 4nm process.

Beside you re completely wrong even for 15W in CB, at 4.4GHz a Zen 4 will use less power despite higher IPC than a 5800U core

Think about a DT 7800X/7950X, if each core was to use 15W/4.4Ghz it would use 20W at this same frequency when running SMT, that would make 160W for 8 cores at 4.4GHz, and something like 320W for a 7950X 16 cores, yet they run at 5Ghz, wich would add 40% to those 160 and 320W...

Hope that you understand at wich point you are completly off the tracks.

 

[eek2121]

Performance yes, battery life hell no.

I don't think AMD could hit Nintendo's price needs unless it was fabbed at Samsung. There was actually speculation that Nintendo had some interest in Samsung's RDNA2 chips but that has died down and it seems like they will stick with nVidia.

Apples and oranges.

The switch is a much lower performing system. Keep in mind that the Steam Deck can emulate a switch.

You can’t even compare a title such as the witcher 3, because there are switch specific changes that make the comparison meaningless. Shaders were completely replaced, for example.

The only proper comparison would be rendering the same scene with the same amount of polygons with the same shaders at a fixed framerate. In such a scenario, the Steam Deck would win.

 

[Mopetar]

I don't quite get how anyone could consider the Switch to be more capable hardware. It's using Maxwell era graphics technology on an incredibly old process node.

Does that mean that it can't produce a good looking game? No. Does it make it terribly inefficient? No. Expecting it to perform as well as a modern architecture on a modern node is silly though. There are certainly performance gains to be had from optimizing to a specific hardware configuration, but nothing prevents anyone from doing that for a Steam Deck if they wanted.

I think Nintendo's best reason to stick with NVidia is DLSS. They're already running underpowered hardware so being able to upscale to 4K is a big deal for them. Yeah they could use FSR for this, but NVidia does have a lead in this technology right now.

I think their architecture is a little bit better as well and if Nintendo gets cutting edge NVidia graphics tech paired with a newer process node they'll have something special even if the ARM cores are just stock. If I were Nintendo I'd just insist they rip out any of the RT or tensor cores because the hardware won't be powerful enough and that'd make a much leaner SoC.

I don't know if NVidia would make custom silicon like that for Nintendo though. Maybe they would given how well the Switch did, but they don't have a lot of history of such relationships.

 

[Rigg]

I think Nintendo's best reason to stick with NVidia is DLSS. They're already running underpowered hardware so being able to upscale to 4K is a big deal for them. Yeah they could use FSR for this, but NVidia does have a lead in this technology right now.

I think their architecture is a little bit better as well and if Nintendo gets cutting edge NVidia graphics tech paired with a newer process node they'll have something special even if the ARM cores are just stock. If I were Nintendo I'd just insist they rip out any of the RT or tensor cores because the hardware won't be powerful enough and that'd make a much leaner SoC.

I don't know if NVidia would make custom silicon like that for Nintendo though. Maybe they would given how well the Switch did, but they don't have a lot of history of such relationships.

DLSS requires tensor cores. It'd be pretty hard to lean on DLSS as a crutch if they stripped out the tensor cores required to run it.

 

[LightningZ71]

If Nintendo goes anywhere with update Switch internals, their only real choice is the Orin nano platform. It's dramatically better than the Tegra X1 in every possible way. Going any other direction will give them a significant programming headache if they want any sort of forward compatibility for any of the existing Switch ecosystem. Of course, most any modern phone SoC with their full house gpu section should be about as good as the Maxwell iGPU section of the Tegra X1 by this point, but the gpu side of things may be a bit hard to translate.

 

[Mopetar]

DLSS requires tensor cores. It'd be pretty hard to lean on DLSS as a crutch if they stripped out the tensor cores required to run it.

The original DLSS did, but I don't think the more recent versions need it. Maybe they use it for some stuff, but you can do super sampling without them. I don't believe that FSR needs it on NVidia hardware (hence the ability to use it on Pascal or GTX Turing cards) so it's certainly possible to produce the effect without tensor cores.

The main reason that NVidia cards still have tensor cores is so NVidia can sell them to professional and data center customers where they are actually used. I don't know how many additional transistors they eat up, but they could probably be cut with minimal impact.

 

[Rigg]

The original DLSS did, but I don't think the more recent versions need it. Maybe they use it for some stuff, but you can do super sampling without them. I don't believe that FSR needs it on NVidia hardware (hence the ability to use it on Pascal or GTX Turing cards) so it's certainly possible to produce the effect without tensor cores.

The main reason that NVidia cards still have tensor cores is so NVidia can sell them to professional and data center customers where they are actually used. I don't know how many additional transistors they eat up, but they could probably be cut with minimal impact.

"DLSS 3 is powered by the new fourth-generation Tensor Cores and Optical Flow Accelerator of the NVIDIA Ada Lovelace architecture, which powers GeForce RTX 40 Series graphics cards."

Introducing NVIDIA DLSS 3

Multiply frame rates by up to 4X!

www.nvidia.com

 

[Hitman928]

DLSS requires tensor cores. It'd be pretty hard to lean on DLSS as a crutch if they stripped out the tensor cores required to run it.

The original DLSS did, but I don't think the more recent versions need it. Maybe they use it for some stuff, but you can do super sampling without them. I don't believe that FSR needs it on NVidia hardware (hence the ability to use it on Pascal or GTX Turing cards) so it's certainly possible to produce the effect without tensor cores.

The main reason that NVidia cards still have tensor cores is so NVidia can sell them to professional and data center customers where they are actually used. I don't know how many additional transistors they eat up, but they could probably be cut with minimal impact.

All versions (except like 1.8 or something that was a transition version between 1 and 2) require tensor cores, which is why switch games use FSR for upscaling.

Sonic Frontiers on Switch apparently uses AMD’s FSR image upscaling

Sonic Frontiers could be joining some of Nintendo's first party games in taking advantage of AMDs open source FSR image upscaling technique. Several other games on the Nintendo Switch use the…

mynintendonews.com

 

[Mopetar]

"DLSS 3 is powered by the new fourth-generation Tensor Cores and Optical Flow Accelerator of the NVIDIA Ada Lovelace architecture, which powers GeForce RTX 40 Series graphics cards."

Introducing NVIDIA DLSS 3

Multiply frame rates by up to 4X!

www.nvidia.com

Do they need to use them or is NVidia trying to find some reason to use them so customers don't ask why they're paying for transistors that are otherwise useless?

I suspect they could do without them if they had to.

 

[Rigg]

Do they need to use them or is NVidia trying to find some reason to use them so customers don't ask why they're paying for transistors that are otherwise useless?

I suspect they could do without them if they had to.

As @Hitman928 pointed out they were able to adapt DLSS 1 (unofficially named 1.9 ) to use the CUDA shader cores instead of the tensor cores. According to Wikipedia, the only game that used this was Control when it launched and it was later updated to DLSS 2 which does use the tensor cores. So at the very least DLSS 1 didn't necessarily need the tensor cores. Your suspicion could be well founded.

 

[A///]

someone correct me if wrong but someone made a workaround to get dlss3 running on older gen cards and the performance was terrible.

 

[Geven]

You've made some great points, @Mopetar! The Switch does have its limitations due to the older Maxwell-era graphics technology, but it still manages to deliver enjoyable gaming experiences for many users. Nintendo's partnership with Nvidia and the use of DLSS technology is a significant factor in its favor. While FSR is an alternative, Nvidia's expertise and current lead in this area make it a valuable ally for Nintendo. I like your idea of customizing the SoC by removing RT or tensor cores to create a leaner, more efficient system. This could potentially give Nintendo an edge in performance without compromising too much on cost. Whether Nvidia would be willing to create custom silicon for Nintendo is an interesting question. The success of the Switch could be a motivating factor, but as you mentioned, such partnerships aren't very common.

 

[eek2121]

You've made some great points, @Mopetar! The Switch does have its limitations due to the older Maxwell-era graphics technology, but it still manages to deliver enjoyable gaming experiences for many users. Nintendo's partnership with Nvidia and the use of DLSS technology is a significant factor in its favor. While FSR is an alternative, Nvidia's expertise and current lead in this area make it a valuable ally for Nintendo. I like your idea of customizing the SoC by removing RT or tensor cores to create a leaner, more efficient system. This could potentially give Nintendo an edge in performance without compromising too much on cost. Whether Nvidia would be willing to create custom silicon for Nintendo is an interesting question. The success of the Switch could be a motivating factor, but as you mentioned, such partnerships aren't very common.

Are you a bot?

 

[Mopetar]

Are you a bot?

Maybe the G stands for Geven

 

[Geven]

Are you a bot?

What about my message made you think so?

 

[KompuKare]

What about my message made you think so?

Nah, we are not falling for that one: if we correct bot-like talk for all bots, before you know it they bots will learn to not use that style!

 

[Geven]

Well, I'll try to be less bot-like the next time

 

[DrMrLordX]

ChatGPT 5 in the house! woop woop!

 

[TESKATLIPOKA]

If a Zen 3+ use 5W for X FPS with a 680M it s logical that a 4nm Zen 4 with 10% higher FPS from the GPU will use even less power, you are grasping at straws here..

Here is what you wrote:

It is, because i use actual numbers of an APU used for gaming, while you re lelying on CB at 3x the game play frequency, and on a 7nm node that is half as efficient as 4nm..

You criticized me for using CB R23 ST power consumption for 7nm 5800U, yet you yourself didn't use 4nm Phoenix data but 6nm Rembrandt instead.
And the thing is that I didn't even make a mistake to begin with, because when I wrote that 5800U consumed 15W in CB R23 ST, I was replying to a post of yours which was not about Phoenix APU or even about gaming frequency.
Here is what I replied to:

20% CPU usage is 20%, whatever turbo boost is applied.

100% is 8 cores running at 2.5W each, at a frequency F.

20% is 1.6 cores running at said frequency F and using about 5W.

Beside turbo is a very bad thing, one core/1T at 3GHz use 5W.

2 cores/2T/5W run at 2.3GHz; that s 53% more throughput than 1C/1T/5W.

3C/3T/5W run at 2GHz, and that s 2x the throughput of 1C/1T/5W.

In power limited scenario it is of great advantage for autonomy and perfs to distribute the load on as much cores as possible.

-----------------------------------------------------------------------------------next reply-----------------------------------------------------------------------------------​

Your 15W are at 4.4GHz with a 7nm based Zen 3, here we re talking of a CPU that run at 1.4GHz with a 4nm process.

Beside you re completely wrong even for 15W in CB, at 4.4GHz a Zen 4 will use less power despite higher IPC than a 5800U core

Think about a DT 7800X/7950X, if each core was to use 15W/4.4Ghz it would use 20W at this same frequency when running SMT, that would make 160W for 8 cores at 4.4GHz, and something like 320W for a 7950X 16 cores, yet they run at 5Ghz, wich would add 40% to those 160 and 320W...

Hope that you understand at wich point you are completly off the tracks.

Only you are/were talking about a 4nm CPU that runs at 1.4GHz in games.
I was talking about something else the entire time -> that the CPU usage in % doesn't mean that the CPU part of an APU consumes the same % amount of the whole package, which you pretty much assumed in the post I linked here. I was just reacting to that, that's how our debate started.
The correct answer is that power consumption of the CPU part depends on CPU usage(how many cores/threads are used), frequency and voltage of cores, honestly even the core itself should matter because of the amount of transistors which are the consumers of energy. That's why I was using Cinebench ST all this time as an example, to show you It's not true what you wrote at that time.

I didn't say Zen4 would consume 15W at 4.4GHz in CB ST, that's something you made up.

Those APUs have different turboboost frequency.
7nm 5800U has 4.4GHz boost
6nm 6800U has 4.7GHz boost
4nm 7800U has 5.1GHz boost
So It's not surprising If they are consuming 15W during CB R23 single thread. Even If I was wrong, and power consumption was not 15W, but let's say ~8-9W at 5-5.1GHz, then it still would prove my point mentioned above about CPU usage and power consumption.

I hope you finally understand what I was trying so hard to tell you all this time.
If you are still misunderstanding me, then I am trowing in the towel.

 

[Abwx]

You criticized me for using CB R23 ST power consumption for 7nm 5800U, yet you yourself didn't use 4nm Phoenix data but 6nm Rembrandt instead.
And the thing is that I didn't even make a mistake to begin with, because when I wrote that 5800U consumed 15W in CB R23 ST, I was replying to a post of yours which was not about Phoenix APU or even about gaming frequency.

The mistake is that you dont know what is a Zen 4 power comsumption and that you made no effort to grasp things.

A 7950X wich is 5nm based, so less efficient than a 4nm based Zen 4, use 205W in CB at 4.9GHz with the IOD taking about 20W, that s 11.5W per core at 4.9GHz.

at 4.4Ghz this would be 9W at most and at 2.2GHz something like 1.5W, you just dont realise the efficency of a Zen 4 core, and this number is for a 5nm process, not even 4nm.

From here it is obvious that a few threads running at quite less than 2GHz will use very little power, hence the GPU can be allocated way higher power than the CPU even at a 25W total TDP.

Those APUs have different turboboost frequency.
7nm 5800U has 4.4GHz boost
6nm 6800U has 4.7GHz boost
4nm 7800U has 5.1GHz boost
So It's not surprising If they are consuming 15W during CB R23 single thread. Even If I was wrong, and power consumption was not 15W, but let's say ~8-9W at 5-5.1GHz, then it still would prove my point mentioned above about CPU usage and power consumption.

I hope you finally understand what I was trying so hard to tell you all this time.
If you are still misunderstanding me, then I am trowing in the towel.

There s no misunderstanding, see my response above, when talking of power usage one should know what are the real numbers for the APu when gaming, not in CB.

Beside Chips and Cheese made an article about Van Gogh and wich was posted by Schmide previously, it is said that CPU frequencies and power are very low despite using less performant cores and process, and it is no different for a 7840, this latter has only 4 more CUs than VGogh.

Page 55 - Question - AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 55 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Edit : NBC compiled some numbers for the Asus Handeld when emulating PS3 like games :

Asus ROG Ally runs popular console emulators including Xenia for Xbox 360 and RPCS3 for PS3, God of War 3 emulated at 1080p 60 fps thanks to Ryzen Z1's AVX-512 support

The Asus ROG Ally is capable of handling a host of popular console emulators thanks to AVX-512 support in its Ryzen Z1 APU. The Ally was shown to effectively run popular console emulators — from light ones such as the PPSPP for the PSP to the more complex ones including Xenia for the Xbox 360...

www.notebookcheck.net

 

[TESKATLIPOKA]

The mistake is that you dont know what is a Zen 4 power comsumption and that you made no effort to grasp things.

A 7950X wich is 5nm based, so less efficient than a 4nm based Zen 4, use 205W in CB at 4.9GHz with the IOD taking about 20W, that s 11.5W per core at 4.9GHz.

at 4.4Ghz this would be 9W at most and at 2.2GHz something like 1.5W, you just dont realise the efficency of a Zen 4 core, and this number is for a 5nm process, not even 4nm.

From here it is obvious that a few threads running at quite less than 2GHz will use very little power, hence the GPU can be allocated way higher power than the CPU even at a 25W total TDP.




There s no misunderstanding, see my response above, when talking of power usage one should know what are the real numbers for the APu when gaming, not in CB.

Beside Chips and Cheese made an article about Van Gogh and wich was posted by Schmide previously, it is said that CPU frequencies and power are very low despite using less performant cores and process, and it is no different for a 7840, this latter has only 4 more CUs than VGogh.

Page 55 - Question - AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 55 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Edit : NBC compiled some numbers for the Asus Handeld when emulating PS3 like games :

Asus ROG Ally runs popular console emulators including Xenia for Xbox 360 and RPCS3 for PS3, God of War 3 emulated at 1080p 60 fps thanks to Ryzen Z1's AVX-512 support

The Asus ROG Ally is capable of handling a host of popular console emulators thanks to AVX-512 support in its Ryzen Z1 APU. The Ally was shown to effectively run popular console emulators — from light ones such as the PPSPP for the PSP to the more complex ones including Xenia for the Xbox 360...

www.notebookcheck.net

You are right, there was no misunderstanding on your part, I was too naive thinking otherwise.
You just didn't want to understand what was my point, your latest post just confirmed It.
I won't bother with this discussion any longer, It's just pointless and eats up my free time.
I am trowing in the towel.

 

[jpiniero]

Ooh... actual Phoenix laptops. Choose between a 7640HS+3050 6 GB or 7840HS+4050. Don't go rushing all at once.

Lenovo LOQ 15APH8 | Ryzen™ powered gaming laptop

The Lenovo LOQ 15APH gaming laptop is your gateway to PC gaming. Assemble your squad and start crushing AAA titles with AMD Ryzen™ 7000 Series processors and NVIDIA® GeForce RTX™ GPUs, backed up by abundant DDR5 memory and SSD storage options.

www.lenovo.com

 

[moinmoin]

Well, that's at least better than in Germany where there are still only ASUS TUF and ROG models available including Phoenix HS, all of which had been added to the database between February and April:

Notebooks mit Codename AMD: Phoenix Preisvergleich Geizhals Deutschland

Preisvergleich und Bewertungen für Notebooks mit Codename AMD: Phoenix

geizhals.de

Call me again when there finally are any Phoenix U laptops. Framework is still the only one, though backordered.