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 My advice to everyone: treat the old APU rumours as being out of date.

 

[TESKATLIPOKA]

If 8C/16T 7700X clocks around 5.55 GHz at 65W TDP - at 35W TDP I think we can expect even 4.4-4.5 GHz all core boost.

Well Rembrandt 45W parts, like 6900HX, were actually 55W parts based on the power drawn for whole package.

P.S. With such temperatures on the CPU, for 65W TDP I think those CPUs are easily able to be passively cooled by something like Noctua NH-P1, or BeQuiet! Shadow Rock 3.

Let alone even lower wattage APU chips.

I am more interested in average clockspeed than some peak clocks no longer than 1-2s.
Average of 4.8GHz a 65W(88W) is still very good.
I think we won't see TDPs over >90W simply because It can't be easily cooled.

 

[Glo.]

I am more interested in average clockspeed than some peak clocks no longer than 1-2s.
Average of 4.8GHz a 65W(88W) is still very good.
I think we won't see TDPs over >90W simply because It can't be easily cooled.

RMB maximum power state is 80W for whole package.

 

[TESKATLIPOKA]

RMB maximum power state is 80W for whole package.

You can find Ryzen 9 6900HX with higher limit.
98W sustained.

 

[adamge]

Notebookcheck.net tested Ryzen 9 7950X and Ryzen 7 7700X.

View attachment 72329


View attachment 72330

I think 8C16T Phoenix could hit 4GHz all core in CB R15 with 45W TDP or 35W?

P.S. How is mobile vs desktop TDP calculated? Is It the same or mobile TDP is basically desktop PPT?

I wholeheartedly agree with the use of reduced power budget configuration on these factory-overclocked products. It shows just how silly the manufacturers have become in their marketing, to the point where they sacrifice the efficiency of their product to extreme degrees.

I do not like to see that percentage representation of the core temperature difference. That is a nonsensical number.

 

[TESKATLIPOKA]

I wholeheartedly agree with the use of reduced power budget configuration on these factory-overclocked products. It shows just how silly the manufacturers have become in their marketing, to the point where they sacrifice the efficiency of their product to extreme degrees.

I think this is not only manufacturer's fault. People want performance first, efficiency is second at best. If there is not enough generational uplift, then you have to increase TDP.

 

[Kaluan]

The latter. Only desktop goes through the weird TDP = 1.35 * PPT calculation. Mobile and server go TDP = PPT.

Isn't it PPT = 1.2x TDP for non-DTR mobile Ryzen chips?
PL1 = 35W, PL2 = 42W on HS
PL1 = 45W, PL2 = 54W on H/HX
U class seems different tho and at any rate, laptop ODMs cTDP up or down these chips as much they want based on the designs they want or have.

 

[Mopetar]

I think 8C16T Phoenix could hit 4GHz all core in CB R15 with 45W TDP or 35W?

P.S. How is mobile vs desktop TDP calculated? Is It the same or mobile TDP is basically desktop PPT?

Should be able to do that without issue. Cezanne could already boost above 4 GHz on the 35W TDP parts, so there little doubt that Phoenix will be able to considering it's on a better node.

AMD gave AM5 a lot of extra headroom, but I don't think Zen 4 is actually all that good at taking advantage of it. As you can see from the images you linked, the chips don't lose much performance from dropping down the different power buckets.

Even though some changes were made to allow Zen 4 to clock higher, it's pretty clear that AMD still designed it around the kind of power targets that the server market will use and that it doesn't scale all that well outside of that. Perhaps Zen 5 will be slightly different, but Zen 4 still shines in lower power settings.

 

[eek2121]

You didn't provide any reason why Steam couldn't use PHX2 instead of a custom chip. BTW the custom one Steam used is pretty much Van Gogh and that one was meant for Microsoft originally If I remember correctly, so It wasn't like they commissioned AMD to make them one, they used an existing one.

What I meant was a Steam Deck 2 Lite based on PHX2, which would replace Steam Deck at the $399 price point. Steam deck 2 would have a more powerful chip with more storage or even more RAM and higher resolution screen. I don't dare to say they will do It, but could do It, and that looks to me more realistic than a new Atari console with PHX2.

ATARI VCS:
AMD R1606G is basically Ryzen 3 3200U a 14nm Zen+ based 2C4T 2.6GHz(3.5GHz) with 3CU 1.2GHz Vega IGP
4-8GB RAM depending on model
32GB eMMC
Introduction price:
$299 (Onyx Base system without controllers)
$399(All-In system with 2 different controllers)
Those specs are so pathetically cheap, yet they dared to ask $299-399 for this!

Performance comparison:
3200U IGP -> 3CU*64SP*1.2GHz*2FLOPs => 461 GFLOPs
PHX2 IGP -> 4CU*64SP*2.4GHz*2FLOPs => 1229 GFLOPs
If I include the better architecture, then PHX2 IGP could be ~4-4.5x faster.

	R15 ST 64bit	R15 MT 64bit	R20 ST	R20 MT
Ryzen 3 3200U	130 (100%)	353 (100%)	318 (100%)	727 (100%)
Ryzen 5 PRO 6650U	234 (180%)	1539(+436%)	574 (181%)	3564 (490%)

PHX2 CPU should be faster than 6650U.

4nm PHX2 based machine with more and costlier DDR5 RAM, bigger SSD would naturally cost more, so $299 for base without controllers is already pretty questionable and who would buy this console for that price without a controller?
You have better options like a more powerful Xbox with a Xbox controller for the base price or simply buy a Steam Deck for the price of console + controller with a way better software support.

You also didn't explain why Atari would even need to use PHX2 in Atari VCS2, when the games they produce simply don't need so much horsepower and can be played on Windows. Atari would need first those visually better games and only after them could have a reason for this console to exist or for buyers to want It.
Possibility of installing windows on It is not such a great selling point.

By far the most important part of a mobile gaming system is the GPU. Look at desktop for a moment. Compare a 3600 to a 5600x with the same GPU and look at the numbers.

Valve's goal with the Steam Deck is to give gamers an affordable, yet compatible/accessible goal for portable gaming. Moving from Zen 2 to Zen 3 or Zen 4 will not fix that.

I strongly suspect their next gen APU will be on 6nm and a (lower end) Zen 3 based processor so that they can easily target the < $400 demographic.

Note that for the steam deck, I purchased the most expensive version with the improved screen and 512gb storage, I most certainly wasn't their target.

It is possible that Valve has a higher end option, of course, but it likely won't be the default. Even on the current Steam Deck, the GPU sucks almost all the power.

 

[Kaluan]

In case someone missed it (I did), here's a alleged die diagram (edited by OP to mock a Apple SOC design) of PHX2/"Little Phoenix":

https://twitter.com/x/status/1597289898112823296

If it is based on something real... no idea if it's to scale, but holy smokes, the Zen4C cores look ridiculously tiny.

 

[BorisTheBlade82]

In case someone missed it (I did), here's a alleged die diagram (edited by OP to mock a Apple SOC design) of PHX2/"Little Phoenix":

https://twitter.com/x/status/1597289898112823296

If it is based on something real... no idea if it's to scale, but holy smokes, the Zen4C cores look ridiculously tiny.

This goes against AMD's own communication that Zen4c is roughly half the size.

 

[moinmoin]

laptop ODMs cTDP up or down these chips as much they want based on the designs they want or have.

Exactly.

Have you seen those TDP divert from PPT? Under Linux I haven't, but on my Renoir laptop I don't even have the option to set the cTDP in the BIOS so I'm not sure at what cTDP it's officially set and sold. STAPM limit is 18W on mine. (On Geizhals the model is listed as 15W TDP, 10-28W cTDP, but that looks like standard values for 4500U or even the whole 4000U series.) The more important FAST and SLOW limits are 30W and 25W respectively btw. though. The chip stays well under 10W in normal use and around 3W at idle though so all those values are kinda whatever outside of iGPU heavy use like gaming. The iGPU most certainly is the far more demanding power eater whenever in serious use.

 

[deasd]

In case someone missed it (I did), here's a alleged die diagram (edited by OP to mock a Apple SOC design) of PHX2/"Little Phoenix":

https://twitter.com/x/status/1597289898112823296

If it is based on something real... no idea if it's to scale, but holy smokes, the Zen4C cores look ridiculously tiny.

He said he made it looked like Apple diagram by himself, and he never said this was an AMD product.

https://twitter.com/x/status/1597390614815215617

IMO the 'new co-processors' , 'new 4 GPU' , 'new AI engine' just literally make it more like a smartphone SoC.

This goes against AMD's own communication that Zen4c is roughly half the size.

cuz this image is made by him and exactly not from AMD.

 

[yuri69]

https://twitter.com/x/status/1597289898112823296

Next-generation memory? Nope. It's still Rembrandt's LPDDR5.

 

[TESKATLIPOKA]

He said he made it looked like Apple diagram by himself, and he never said this was an AMD product.

https://twitter.com/x/status/1597390614815215617

IMO the 'new co-processors' , 'new 4 GPU' , 'new AI engine' just literally make it more like a smartphone SoC.


cuz this image is made by him and exactly not from AMD.

He didn't say If It's AMD, true, but All The Watts!! said It's actually PHX2.
As you said, we shouldn't speculate about dimensions because he made that image, so we don't know how accurate it is.

 

[TESKATLIPOKA]

By far the most important part of a mobile gaming system is the GPU. Look at desktop for a moment. Compare a 3600 to a 5600x with the same GPU and look at the numbers.

Valve's goal with the Steam Deck is to give gamers an affordable, yet compatible/accessible goal for portable gaming. Moving from Zen 2 to Zen 3 or Zen 4 will not fix that.

I strongly suspect their next gen APU will be on 6nm and a (lower end) Zen 3 based processor so that they can easily target the < $400 demographic.

Note that for the steam deck, I purchased the most expensive version with the improved screen and 512gb storage, I most certainly wasn't their target.

It is possible that Valve has a higher end option, of course, but it likely won't be the default. Even on the current Steam Deck, the GPU sucks almost all the power.

You are correct that GPU is the most important, but I don't see why It should be based on Zen3 when Zen4 is faster for gaming and more efficient, which would help Steam Deck even If GPU is the main battery eater. They need the most efficient chip, so 6nm is also not a good option, my bet is at least 5nm process.
If this PHX2 had 3WGPs and not just 2, I would say It would be a good option for <$400 Steam Deck.

 

[deasd]

He didn't say If It's AMD, true, but All The Watts!! said It's actually PHX2.
As you said, we shouldn't speculate about dimensions because he made that image, so we don't know how accurate it is.

yeah. It is the account that already being suspected days before. That's why I take a truck load of salt.
This guy just cited every rumors floated around, like 'Zen4 X3D has 3 models and same TDP' from Wccftech, he immediately added comment that every X3D skus has 170w TDP. quite funny.

 

[TESKATLIPOKA]

Should be able to do that without issue. Cezanne could already boost above 4 GHz on the 35W TDP parts, so there little doubt that Phoenix will be able to considering it's on a better node.

Does a 35W Cezanne sustain 4GHz during CB R23 MT 10min run?

yeah. It is the account that already being suspected days before. That's why I take a truck load of salt.
This guy just cited every rumors floated around, like 'Zen4 X3D has 3 models and same TDP' from Wccftech, he immediately added comment that every X3D skus has 170w TDP. quite funny.

You are right about being suspicious, I am starting to suspect he is actually Greymon.

 

[uzzi38]

You are right about being suspicious, I am starting to suspect he is actually Greymon.

I think he's not Greymon but someone else I know. But I really am guessing here based off of the things they have been posting.

The person I'm thinking of is one of the usual Twitter gang though.

 

[Thibsie]

I think he's not Greymon but someone else I know. But I really am guessing here based off of the things they have been posting.

The person I'm thinking of is one of the usual Twitter gang though.

Twitter Silicon Gang ? Must be flip-flop xD

 

[scineram]

So many allegedly credible leaks, yet nobody is willing to stake $10?

 

[Kaluan]

He said he made it looked like Apple diagram by himself, and he never said this was an AMD product.

https://twitter.com/x/status/1597390614815215617

IMO the 'new co-processors' , 'new 4 GPU' , 'new AI engine' just literally make it more like a smartphone SoC.


cuz this image is made by him and exactly not from AMD.

I mistook that as "I erased AMD's and edited in some annotations of my own to make it sound Apple-esque" then.

Next-generation memory? Nope. It's still Rembrandt's LPDDR5.

Phoenix/PHX2 allegedly supports LPDDR5X, Rembrandt doesn't. That's the "next gen" IMHO.

The words are a superfluous made up word play by him, I wouldn't take them seriously anyway.


Well, if it is a Apple SoC diagram and not a AMD one mocked up, anyone know which Ax it is? Or is it 100% made up?

 

[Shivansps]

Not to mention no infinity cache and sumilar DRAM bandwidth. Driver updates in Linux suggest internal cache improvements with respect to RDNA2 and differences vs. desktop RDNA3 that point to a focus on improving capabilities relevant to its APU target, so it may surprise us, however, expecting RX570 level performance out of Phoenix without a significant hardware change vs. Rembrandt is overly optimistic.

Keep in mind, RX570 had 224GB/sec of VRAM bandwidth. Rembrandt was doing well to get 50GB/sec of read bandwidth out of its ddr5-4800 spec. Phoenix is supposedly only getting a modest bump to somewhere in the 5200-5600 range for ram, so maybe another 10% vram throughput is available. I don't care how good your memory efficiency is and your internal caching is, you're not making that up easily.

Remember what happens with the RX550 112GB/s vs the 3400G/5600G with about 26GB/s SHARED. Polaris is very inefficient.

 

[Mopetar]

In case someone missed it (I did), here's a alleged die diagram (edited by OP to mock a Apple SOC design) of PHX2/"Little Phoenix":

https://twitter.com/x/status/1597289898112823296

If it is based on something real... no idea if it's to scale, but holy smokes, the Zen4C cores look ridiculously tiny.

A diagram isn't the same as a die shot. It's not a literal indication of size or layout.

 

[LightningZ71]

Remember what happens with the RX550 112GB/s vs the 3400G/5600G with about 26GB/s SHARED. Polaris is very inefficient.

I agree that Polaris is very inefficient. It is also the case that the RX550 is quite undersized for the amount of bandwidth available. The RX560 is specced for the exact same total bandwidth and is markedly faster at roughly 40% in most cases, depending on which model you are testing.

The RX570 is slightly faster than the gtx 1650 and slightly slower than the 1650 super. If Rembrandt was as fast as the desktop 1650 today, then I would believe that PHX will blow past the RX570. Unfortunately, the desktop 1650 is, on average, 30+% faster than the 680m, with the RX570 being a bit faster. 30+% is a lot to overcome when you start getting to those levels. Is it possible? I suppose, but it's going to take more than bigger caches and 20% more memory bandwidth with a slightly higher iGPU clock to overcome that deficit.

 

[Shivansps]

I agree that Polaris is very inefficient. It is also the case that the RX550 is quite undersized for the amount of bandwidth available. The RX560 is specced for the exact same total bandwidth and is markedly faster at roughly 40% in most cases, depending on which model you are testing.

The RX570 is slightly faster than the gtx 1650 and slightly slower than the 1650 super. If Rembrandt was as fast as the desktop 1650 today, then I would believe that PHX will blow past the RX570. Unfortunately, the desktop 1650 is, on average, 30+% faster than the 680m, with the RX570 being a bit faster. 30+% is a lot to overcome when you start getting to those levels. Is it possible? I suppose, but it's going to take more than bigger caches and 20% more memory bandwidth with a slightly higher iGPU clock to overcome that deficit.

Remember there is a 64 bit version of the RX550 that is actually slower, so it is using that extra bandwidth, maybe not to full extend. The RX560 is petty much a double RX550, incluiding the L2 cache.

I still think there is something wrong with RMB that resulted in lower than expected performance and im not sure if memory bandwidth is the issue here. I think RDNA2 is build around IC and not having it here is creating some issues.