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 My advice to everyone: treat the old APU rumours as being out of date.

 

[gdansk]

50W for an portable CPU
No wonder Nintendo will stick with ARM

 

[Schmide]

I don't think anyone is debating that these APUs can flexibly allocate power, but being "squished below base clocks" (i.e. <2.8GHz, and surely on only a core or so) is not really viable for modern gaming. It's a theoretical possibility, but I don't see anyone demonstrating such a scenario outside of GPU-only synthetics, as they essentially call out in that review. In actual games, it would be reasonable to expect a much more even (i.e. CPU-heavy) power allocation. There's a baseline level of CPU compute games typically need, and trying to push below that makes for a terrible experience.

Everything now runs to a power budget. Inferring that synthetics are the lower bound for this balance is a bit of hyperbole.

When 13k processors can easily reach 200+ watts under certain loads, yet when gaming they often fall back to reasonable 100w levels. (more or less) The same is true on a smaller scale when the bottleneck is pushed to an APU with 15% the GPU power of a flagship card.

 

[deasd]

50W for an portable CPU
No wonder Nintendo will stick with ARM

I guess AMD wasn't even bother designing new chips for handheld and reuse everything from laptop/ULV, just made battery beefy enough and done.

 

[Exist50]

Inferring that synthetics are the lower bound for this balance is a bit of hyperbole.

The C&C article called that out as a GPU-only workload. There are some real-life examples that may produce similar results (e.g. rendering), but it's certainly not representative of gaming. Anyway, I think best to drop the topic at this point. Everything that could be said, has.

 

[deasd]

It's clear that Z1E CPU perf is scaled way too beyond what it need to be. Don't waste this kind of perf and.....play emulator on it.

I'm impressed by Bayonetta2 and GOW3 playing.

 

[Abwx]

If you didn't notice, In that link they compare Core i7-1165G7 vs R7 5800U
That's 4C8T for Intel vs 8C16T for AMD.
25% usage in the case of AMD would mean 4 threads are fully loaded or 16 threads are loaded at 25%.
50% usage in the case of Intel would mean 4 threads are fully loaded or 8 threads are loaded at 50%.

You are assuming that both CPU can use the same power /core, wich is wrong,
AMD has 25W for 8C and Intel 30W for 4 cores...

25% usage for AMD is 25% of 25W whatever the thread number.
Same for Intel, 50% usage of 30W...

 

[Panino Manino]

50W for an portable CPU
No wonder Nintendo will stick with ARM

Are you aware that the Switch CPU is capped at 1020MHz?

 

[hemedans]

50W for an portable CPU
No wonder Nintendo will stick with ARM

Deck already surpass Nintendo Switch in both perfomance and battery life.

 

[insertcarehere]

NBC review of the ROG Ally has been very kind to include comparisons with the GPD Win 4 with 6800U - another gaming handheld.

So... A node shrink and a new architecture yielded an average improvement of 10% in games. I think we know why AMD never included the 6800U in the GPU comparisons.

 

[naukkis]

You are assuming that both CPU can use the same power /core, wich is wrong,
AMD has 25W for 8C and Intel 30W for 4 cores...

25% usage for AMD is 25% of 25W whatever the thread number.
Same for Intel, 50% usage of 30W...

Newer heard of turbo boost? That main principle is to boost cpu clock when usage is less than 100% to redistribute more power to less threads.

 

[Abwx]

Newer heard of turbo boost? That main principle is to boost cpu clock when usage is less than 100% to redistribute more power to less threads.

Clock boost is such that CPU utilisation is 20% of the max throughput that occur if the CPU part use 25W.

Newer heard of turbo boost? That main principle is to boost cpu clock when usage is less than 100% to redistribute more power to less threads.

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.

 

[naukkis]

Clock boost is such that CPU utilisation is 20% of the max throughput that occur if the CPU part use 25W.


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.

20% cpu usage means 20% of cpu threads are in use. It still 20% cpu usage even if those 20% of threads are turbo-boosted to full tdp. What you calculate is only true with cpu that doesn't have any kind of turbo boost for low-thread count scenarios.

 

[TESKATLIPOKA]

You are assuming that both CPU can use the same power /core, wich is wrong,
AMD has 25W for 8C and Intel 30W for 4 cores...

25% usage for AMD is 25% of 25W whatever the thread number.
Same for Intel, 50% usage of 30W...

I never assumed such a thing, so please stop making things up.
That 25% or 50% represents how much of the CPU is used, not the actual power draw. If we were talking about 100% CPU usage, then It would be close to It, let's say ~85-90% of power budget consumed by cores+SMT at full CPU load. @naukkis already explained It to you.

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.

5800U consumes ~15W during CB R20 ST. Techspot 5800U review

The big question around Ryzen Mobile 5000 though isn't multi-thread performance, but single-thread performance, and how that compares to Tiger Lake. The results are promising: the 5800U is 14% faster than the 4800U in Cinebench R20 single-thread, and it remains a situation where single-thread performance is virtually the same whether
testing at 15W or 25W. That's because a single Zen 3 core operating at up to 4.4 GHz tends to only consume about 15W of power, giving us no gains at the higher power spec.

Less than 10% CPU usage results in 15W consumption for the whole chip.
This happens because of turbo boost!

 

[Abwx]

20% cpu usage means 20% of cpu threads are in use. It still 20% cpu usage even if those 20% of threads are turbo-boosted to full tdp. What you calculate is only true with cpu that doesn't have any kind of turbo boost for low-thread count scenarios.

That s wrong, because there would be no power left for the GPU if the CPU was to boost taht way.

Beside you re wrong when it come to thread utilsation, 8 cores loaded at 2.5% of the maximal loading would yield 20% CPU usage as well.

 

[Abwx]

5800U consumes ~15W during CB R20 ST. Techspot 5800U review

Less than 10% CPU usage results in 15W consumption for the whole chip.
This happens because of turbo boost!

At 4.4GHz, and a 7nm 5800U, that s apples and potatoes.

At 3Ghz a single Zen 4 core would go to 3W at most, beside HV info show the CPU frequency way lower than 3GHz during game play, i linked a review with screenshots, please refer to it before bringing cases that do not exist like this very bad exemple at 4.4GHz..

Third time i post this link, game play CPU frequency is at 1.4GHz when the APU is set at 25W :

AMD Radeon 680M benchmarks, review (RDNA2 AMD iGPU) vs Iris Xe, Intel

The AMD Radeon 680M is the much-awaited top-tier RDNA2-based integrated GPU bundled with the majority of the 2022 AMD Ryzen 6000 Rembrandt processors.

www.ultrabookreview.com

 

[jpiniero]

Deck already surpass Nintendo Switch in both perfomance and battery life.

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.

 

[TESKATLIPOKA]

At 4.4GHz, isnt it, at 3Ghz a single core would go to 5.5W at most, beside HV info show the CPU frqequency way lower than 3GHz suring game play, i linked a reciew with screenshot, please refer to it before bringing cases that do not exist like this exemple at 4.4GHz..

I don't see a problem in using CB R20 ST as an example, because I only wanted to point out that what you wrote below is not necessarily true.

Since CPU usage relate to the total available power (25W for AMD in the test) then if usage is 20% imply that the CPU part use 5W of the available power; rest go the GPU + uncore.

Intel being at 50% FI mean that the CPU part use 15W (out of the 30W available in the test), rest goes to the GPU + uncore.

AMD has 25W for 8C and Intel 30W for 4 cores...

25% usage for AMD is 25% of 25W whatever the thread number.
Same for Intel, 50% usage of 30W...

% 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.

 

[Abwx]

I don't see a problem in using CB R20 ST as an example, because I only wanted to point out that what you wrote below is not necessarily true.

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..

% 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.

 

[hemedans]

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.

Plenty of test out there, deck will only have bad battery life if you run games at much higher quality and 60fps. But if you run games at 30fps and medium Quality (which is still better quality compare to switch) you will get better battery life with steam deck

if we take a game that's available on both platforms, such as The Witcher 3, we can get a better idea of the comparative battery life of the two systems. In my Steam Deck battery life testing(opens in new tab), with The Witcher 3 running locked at 30 fps on medium settings, I was seeing almost four hours of gaming life from the Deck. With the Switch, however, you're looking more likely at the two and a half to three hour mark.

All that said, however, if you were to leave the Steam Deck to its own devices, and run at its limits with every game you throw at it, you'll have great performance, but battery life that can run dry in just 83 minutes.

That's from Pcgamer test

Steam Deck vs Nintendo Switch

It's not as simple as which has the better specs, they're fundamentally different machines.

www.pcgamer.com

 

[Panino Manino]

I can't see this repurposed notebook APU as anything more than a disappointment now after seeing a few reviews. So horribly bottlenecked by memory and power, wow. AMD really needs to develop a specific Die with the necessary optimizations like the one for the Deck if they really want to supply, and dominate, this market.

 

[Mopetar]

AMD doesn't have much competition in this space so there's little reason to spend money developing a chip targeting it in the hope of it taking off.

Also it would be embarrassing for AMD to lose to the Switch. The Tegra X1 wasn't exactly new hardware by the time Nintendo released it.

 

[Panino Manino]

AMD doesn't have much competition in this space so there's little reason to spend money developing a chip targeting it in the hope of it taking off.

Also it would be embarrassing for AMD to lose to the Switch. The Tegra X1 wasn't exactly new hardware by the time Nintendo released it.

Lose in what way?

 

[moinmoin]

I don't think we can compare these new handhelds fairly to Steam Deck. The former in the end are still all off the shelf hard- and software whereas the latter is being continuously optimized, and Valve is completely peerless in its dedication.

 

[naukkis]

Beside you re wrong when it come to thread utilsation, 8 cores loaded at 2.5% of the maximal loading would yield 20% CPU usage as well.

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.

 

[Mopetar]

Lose in what way?

Any way. The Switch uses a Tegra X1 which is 6 years old and was originally made on a 20nm process. The Switch even drops the clock speeds a fair bit as well.