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 My advice to everyone: treat the old APU rumours as being out of date.

 

[SpudLobby]

What are people on about with "but same number of CUa as Rembrandt"... Hello? They're not the same architecture. 😅

I for one don't find it unrealistic to expect a similar IGP jump with Phoneix as from Cezanne to Rembrandt. Hell I think the RX 570 projection puts it at less than that. We'll know in a month anyway.

What are people on about with "but same number of CUa as Rembrandt"... Hello? They're not the same architecture. 😅

I for one don't find it unrealistic to expect a similar IGP jump with Phoneix as from Cezanne to Rembrandt. Hell I think the RX 570 projection puts it at less than that. We'll know in a month anyway.

Agreed. I think RDNA3 in Phoenix will be a big jump even absent crazy clocks. It’s for laptops, so.

 

[TESKATLIPOKA]

What are people on about with "but same number of CUa as Rembrandt"... Hello? They're not the same architecture. 😅

I for one don't find it unrealistic to expect a similar IGP jump with Phoneix as from Cezanne to Rembrandt. Hell I think the RX 570 projection puts it at less than that. We'll know in a month anyway.

Agreed. I think RDNA3 in Phoenix will be a big jump even absent crazy clocks. It’s for laptops, so.

Cezanne has 8CU Vega IGP at 2.1GHz.
Rembrandt has 12CU RDNA2 IGP at 2.4GHz.
Rembrandt IGP improved everything -> specs, clocks and architecture.

Phoenix has the same specs as Rembrandt. Improvement in architecture is so far only ~9% in N31 although AMD claimed 17.4%.
Phoenix with less bugs could do better, but I don't expect better improvement than RDNA2 over Vega, which was 25-40% just from architecture.

If frequency is not a lot higher, then I don't see how Phoenix IGP could manage to perform comparably to a 63% faster RX 570.
Just to be on par you would need for example 20% improvement from architecture and 3300MHz clockspeed.
From leaks Phoenix ES IGP is only at ~2.6GHz, so you would need 50% improvement from architecture!

BTW even thinking about comparable improvement as Rembrandt had over Vega is totally unrealistic. Rembrandt is 2x faster than Cezanne!

 

[insertcarehere]

Cezanne has 8CU Vega IGP at 2.1GHz.
Rembrandt has 12CU RDNA2 IGP at 2.4GHz.
Rembrandt IGP improved everything -> specs, clocks and architecture.

Phoenix has the same specs as Rembrandt. Improvement in architecture is so far only ~9% in N31 although AMD claimed 17.4%.
Phoenix with less bugs could do better, but I don't expect better improvement than RDNA2 over Vega, which was 25-40% just from architecture.

If frequency is not a lot higher, then I don't see how Phoenix IGP could manage to perform comparably to a 63% faster RX 570.
Just to be on par you would need for example 20% improvement from architecture and 3300MHz clockspeed.
From leaks Phoenix ES IGP is only at ~2.6GHz, so you would need 50% improvement from architecture!

BTW even thinking about comparable improvement as Rembrandt had over Vega is totally unrealistic. Rembrandt is 2x faster than Cezanne!
View attachment 72840

Agreed, after Navi 31 launch it's hard to see how 12CU RDNA3 get above ~40% performance (i.e above GTX 1650) vs 12 CU RDNA2 without either massive power budget increases or significant changes to the actual architecture..

 

[SpudLobby]

Cezanne has 8CU Vega IGP at 2.1GHz.
Rembrandt has 12CU RDNA2 IGP at 2.4GHz.
Rembrandt IGP improved everything -> specs, clocks and architecture.

Phoenix has the same specs as Rembrandt. Improvement in architecture is so far only ~9% in N31 although AMD claimed 17.4%.
Phoenix with less bugs could do better, but I don't expect better improvement than RDNA2 over Vega, which was 25-40% just from architecture.

If frequency is not a lot higher, then I don't see how Phoenix IGP could manage to perform comparably to a 63% faster RX 570.
Just to be on par you would need for example 20% improvement from architecture and 3300MHz clockspeed.
From leaks Phoenix ES IGP is only at ~2.6GHz, so you would need 50% improvement from architecture!

BTW even thinking about comparable improvement as Rembrandt had over Vega is totally unrealistic. Rembrandt is 2x faster than Cezanne!
View attachment 72840

Well, shit.

 

[HurleyBird]

If frequency is not a lot higher, then I don't see how Phoenix IGP could manage to perform comparably to a 63% faster RX 570.
Just to be on par you would need for example 20% improvement from architecture and 3300MHz clockspeed.
From leaks Phoenix ES IGP is only at ~2.6GHz, so you would need 50% improvement from architecture!

BTW even thinking about comparable improvement as Rembrandt had over Vega is totally unrealistic. Rembrandt is 2x faster than Cezanne!

I think the way you get there, hypothetically at least, is through a combination of a lot of factors.

I'm skeptical we'll see anything like 3300MHz (at least on mobile parts), but final parts might gain a hundred or two MHz.

20%+ improvement from architecture is probably doable with fixed drivers and/or hardware (assuming Phoenix isn't also plagued with hardware bugs the same as N31 is rumored to be). And integrated RNDA3 / N33 seem a bit different architecturally than N31/32. Even if the former may be less fully featured on a surface level, that might actually mean fewer bugs. There's some precedent here where with Vega where the integrated version could be called a fixed version of discrete Vega (on the other hand, there was a lot more time between discrete and integrated Vega than there will be with discrete and integrated RDNA 3).

Also worth mentioning that scaling at the top end is not necessarily indicative of scaling for IGPUs, where you have a low number of CUs and are more bandwidth constrained. The bottlenecks are bound to be a bit different.

Phoenix almost certainly has memory controller improvements, and AMD may do something more tailored to an SOC with integrated RDNA 3's cache. Zen 4 cores will add a *tiny* bit of performance.

That said, I won't be surprised if Phoenix ends up a bit underwhelming given the way N31 launched. I would be somewhat surprised if it was a Rembrandt -> Cezanne type of jump, but I don't know if I'd bet money against it either.

 

[Glo.]

Cezanne has 8CU Vega IGP at 2.1GHz.
Rembrandt has 12CU RDNA2 IGP at 2.4GHz.
Rembrandt IGP improved everything -> specs, clocks and architecture.

Phoenix has the same specs as Rembrandt. Improvement in architecture is so far only ~9% in N31 although AMD claimed 17.4%.
Phoenix with less bugs could do better, but I don't expect better improvement than RDNA2 over Vega, which was 25-40% just from architecture.

If frequency is not a lot higher, then I don't see how Phoenix IGP could manage to perform comparably to a 63% faster RX 570.
Just to be on par you would need for example 20% improvement from architecture and 3300MHz clockspeed.
From leaks Phoenix ES IGP is only at ~2.6GHz, so you would need 50% improvement from architecture!

BTW even thinking about comparable improvement as Rembrandt had over Vega is totally unrealistic. Rembrandt is 2x faster than Cezanne!
View attachment 72840

Well, according to rumors, 35W, 2600 MHz iGPU with 6WGPs, 12 CUs is around RX 570 in performance.
According to rumors, Phoenix and Navi 32 are the "fixed" Navi GPUs, while N31 and N33 are still botched.

So with 200 MHz clock speed increase, we are looking at 63% performance increase.

I presume, the rumors are based on 3Dmark Time Spy scores, for GPU(4000-4400 pts range).

Is it because of properly working Dual Issue on PHX? Is it because of doubled resources for the ALUs?

Or maybe there is something else in that architecture...?

To be honest, I am myself sceptical about the rumors based on how N31 looks. But I have to still be objective, and look at information at hand without projecting my own personal beliefs.

For AMDs sake - the rumors better turn out to be true, in this case.

 

[HurleyBird]

According to rumors, Phoenix and Navi 32 are the "fixed" Navi GPUs, while N31 and N33 are still botched.

Got a source for N33 being botched?

Bondrewd @ B3D seems to be of the impression that all the Navi chips have some bugs, that 31 is by far the most buggy but fixed N31 is coming, that at least the clockspeed bug is fixed in N32, while N33 is quote "mostly un-(word I can't say here)ed" and should be pretty impressive.

 

[Thunder 57]

Got a source for N33 being botched?

Bondrewd @ B3D seems to be of the impression that all the Navi chips have some bugs, that 31 is by far the most buggy but fixed N31 is coming, that at least the clockspeed bug is fixed in N32, while N33 is quote "mostly un-(word I can't say here)ed" and should be pretty impressive.

I sure hope that is true so AMD can stick it to NVIDIA with its 4060 Ti 128 bit bus. If AMD can't gain some serious ground this generation (if the NVIDIA rumors are true) than it may be too late.

 

[Mopetar]

I sure hope that is true so AMD can stick it to NVIDIA with its 4060 Ti 128 bit bus. If AMD can't gain some serious ground this generation (if the NVIDIA rumors are true) than it may be too late.

Isn't Navi 33 going to have a 128-bit bus as well? It'll probably wind up with a ballpark similar amount of infinity cache as NVidia has L2 so the bus size or the memory bandwidth limits should hit them both about the same.

Guess it depends on the extent to which it has the same issues that Navi 31 had. The later it launches the more likely that those were fixed or mitigated, but who knows.

 

[Glo.]

I sure hope that is true so AMD can stick it to NVIDIA with its 4060 Ti 128 bit bus. If AMD can't gain some serious ground this generation (if the NVIDIA rumors are true) than it may be too late.

Based on my understanding, 4060 Ti should perform around 3070 Ti levels.

And that is within performance levels of Navi 33, easily, but much more efficient.

 

[Kaluan]

Got a source for N33 being botched?

Bondrewd @ B3D seems to be of the impression that all the Navi chips have some bugs, that 31 is by far the most buggy but fixed N31 is coming, that at least the clockspeed bug is fixed in N32, while N33 is quote "mostly un-(word I can't say here)ed" and should be pretty impressive.

Everyone's new favorite RDNA leaker ("All The Watts!!" on Twitter) seems to say N33 is as broken as current N31 (may have misread tho).
However, they're also saying Phoenix IGP is mostly fine.
They're also saying RDNA3+ designs (unclear if present in Strix Point only) will be when we'll see RDNA3 fully fleshed out and all the kinks worked out. Improvements to the core RDNA3 IP it may bring aside.

So quite a bit of conflicting reports, but all in all it seems AMD is progressively cleaning the RDNA3 mess up with each new design/implementation.

 

[TESKATLIPOKA]

Bondrewd @ B3D was also originally saying 5120SP for N33 and up to 6900XT level of performance.

I more believe in 6750xt level of performance on average and max comparable to RX 6800 with clockspeed comparable to AIB 6650xt that's probably ~2600-2700MHz.
This info is from All The Watts!!

The question is at what resolution this is.
When TPU used Zen 3 5800X It looked like this N22 vs N23:
1080p: +18%
1440p: +26%
2160p: +38%
If It is only at 1080p then that's not very good performance considering gaming frequency increased by ~10-15% so architecture barely helped.
6650M XT has 2162 MHz as gaming frequency for comparison.
TDP didn't change so at worst the efficiency improvement is only 18% thanks to higher clocks.

 

[insertcarehere]

Got a source for N33 being botched?

Bondrewd @ B3D seems to be of the impression that all the Navi chips have some bugs, that 31 is by far the most buggy but fixed N31 is coming, that at least the clockspeed bug is fixed in N32, while N33 is quote "mostly un-(word I can't say here)ed" and should be pretty impressive.

Bondrewd also had this to say about Navi 31 a month ago:

Good news, it won't.
AD102 barely edges ahead at 33% more SMs so...

He does not warrant any more publicity.

Based on my understanding, 4060 Ti should perform around 3070 Ti levels.

And that is within performance levels of Navi 33, easily, but much more efficient.

The 3070 is ~40% faster than a 6650XT at QHD, or about/slightly above the performance uplift from the 6950XT to the 7900XTX.

If 96CU Navi 31 on 5nm with over double the transistor count barely manages ~40% uplift vs 80CU Navi 21. I am extremely doubtful 32 CU Navi 33 with similar transistor count to Navi 23 (200mm^2 N6) can get close to the same vs 32 CU Navi 23, especially given how fast the latter clocks already.

 

[Kaluan]

Bondrewd also had this to say about Navi 31 a month ago:


He does not warrant any more publicity.



The 3070 is ~40% faster than a 6650XT at QHD, or about/slightly above the performance uplift from the 6950XT to the 7900XTX.
View attachment 72925
If 96CU Navi 31 on 5nm with over double the transistor count barely manages ~40% uplift vs 80CU Navi 21. I am extremely doubtful 32 CU Navi 33 with similar transistor count to Navi 23 (200mm^2 N6) can get close to the same vs 32 CU Navi 23, especially given how fast the latter clocks already.

RX 6600M (cut Navi23 ~237mm2, 28CU, N7/N7P) saw a similar uplift from RX 5600M (cut Navi10 ~251mm2, 36CU N7/N7P) as the one alleged from 7600/7700M (Navi33 ~203mm2, 28-32CU, N6). And RDNA2 didn't even have any graphics PPC improvement over RDNA1.

Not sure were you got the transistor count for N33, but please share.

Anyway, not sure why everyone is talking about desktop GPUs in a APU thread. We could at least talk about Phoenix's dGPU companions, pretty sure that's what the N33 leaker was talking about as well.

 

[TESKATLIPOKA]

RX 6600M (cut Navi23 ~237mm2, 28CU, N7/N7P) saw a similar uplift from RX 5600M (cut Navi10 ~251mm2, 36CU N7/N7P) as the one alleged from 7600/7700M (Navi33 ~203mm2, 28-32CU, N6). And RDNA2 didn't even have any graphics PPC improvement over RDNA1.

RDNA2 vs RDNA1 didn't see any PPC improvement, but you are comparing GPUs with 91% difference in clockspeed!
RX5600M = 36CU:2304SP:144TMU:64ROP; boost: 1265 MHz -> 5.83 TFlops
RX6600M = 28CU:1792SP:128TMU:64ROP; boost: 2416 MHz -> 8.66 TFlops
+48.5% TFlops, +69.8% Texture fill rate, +91% Pixel fill rate so It's not a surprise 6600M is a lot faster even If It has 22% less CU.

Mobile N33 will have comparable specs to N23, but boost is ~2700MHz or ~12% higher, the rest of performance has to come from architecture.

Anyway, not sure why everyone is talking about desktop GPUs in a APU thread. We could at least talk about Phoenix's dGPU companions, pretty sure that's what the N33 leaker was talking about as well.

Phoenix can be paired with both N32 and N33 or Nvidia, It has enough performance.
Mobile N33 supposedly performs like desktop RX 6750XT, but we don't know at what resolution. Performance increase should be at least 25%.

 

[Kaluan]

RDNA2 vs RDNA1 didn't see any PPC improvement, but you are comparing GPUs with 91% difference in clockspeed!
RX5600M = 36CU:2304SP:144TMU:64ROP; boost: 1265 MHz -> 5.83 TFlops
RX6600M = 28CU:1792SP:128TMU:64ROP; boost: 2416 MHz -> 8.66 TFlops
+48.5% TFlops, +69.8% Texture fill rate, +91% Pixel fill rate so It's not a surprise 6600M is a lot faster even If It has 22% less CU.

Mobile N33 will have comparable specs to N23, but boost is ~2700MHz or ~12% higher, the rest of performance has to come from architecture.

Phoenix can be paired with both N32 and N33 or Nvidia, It has enough performance.
Mobile N33 supposedly performs like desktop RX 6750XT, but we don't know at what resolution. Performance increase should be at least 25%.

True, but I doubt AMD is preparing only lukewarm upgrades for next gen.

Forgot to mention 5600M, in the few design wins it had, was also geared only up to 85W TGP. 6600M went up to 100W. I suppose 7600M will have a config up 100W as well.

What I found interesting is that 7700M is supposedly up to 120W, while 6700M is up to 135W. Different gen on gen die match up sure, but we can extract some perf/efficiency from it. N33 power characteristics may be strong.

Desktop to mobile N23 SKUs only saw ~10% gain. But N22 SKUs saw much bigger deltas.

I expect something like > 11.500 TS graphics for 7600M, > 13.000 for 7700M, with maybe a +5% for desktop 150W SKUs (Q2 allegedly) and < 4.000 for DDR5 Phoenix "780M" (HS, 35/42W).

 

[jpiniero]

Phoenix can be paired with both N32 and N33 or Nvidia, It has enough performance.

Yeah but Dragon Range is gonna be faster. It might entirely depend on how big Phoenix is whether OEMs would even seriously consider using it for a dGPU gaming laptop compared to Dragon Range or any of the Intels.

 

[TESKATLIPOKA]

True, but I doubt AMD is preparing only lukewarm upgrades for next gen.

Forgot to mention 5600M, in the few design wins it had, was also geared only up to 85W TGP. 6600M went up to 100W. I suppose 7600M will have a config up 100W as well.

What I found interesting is that 7700M is supposedly up to 120W, while 6700M is up to 135W. Different gen on gen die match up sure, but we can extract some perf/efficiency from it. N33 power characteristics may be strong.

Desktop to mobile N23 SKUs only saw ~10% gain. But N22 SKUs saw much bigger deltas.

I don't know about what AMD will show us. N31 didn't leave a good impression for me to have high expectations.
I personally expect more from N32 than N33.

I expect something like > 11.500 TS graphics for 7600M, > 13.000 for 7700M, with maybe a +5% for desktop 150W SKUs (Q2 allegedly) and < 4.000 for DDR5 Phoenix "780M" (HS, 35/42W).

RX 6850M XT managed 11762 points, so I am a bit skeptical 7600M would manage the same score or that 7700 would manage >13,000 points.
The good news is that we only have to wait 3 weeks to find out. At least I hope we will see some benches.

 

[TESKATLIPOKA]

Yeah but Dragon Range is gonna be faster. It might entirely depend on how big Phoenix is whether OEMs would even seriously consider using it for a dGPU gaming laptop compared to Dragon Range or any of the Intels.

12-16C will be faster, but in games shouldn't be noticeable, unless new games will increase the required number of cores.
If 8C Phoenix ends up costing more than 8C Dragon Range then OEM will likely prefer that.
CES 2023 will be exciting.

 

[TESKATLIPOKA]

It is isn't why they couldn't use it, it was about why would they use it. If Valve got the custom Steamdeck processor without spending the silicon development cost good on them. However, if they want to do a Steamdeck 2 it would not be PHX2, but rather a custom solution with newer Zen_x/RDNA_y IP instead. Since, Valve is covered by the S3 Group at AMD. Every dollar saved on the first custom Steamdeck processor can be prepared for the second custom Steamdeck2 processor. Since, they can use customer feedback to make a better SoC -> leading to a better handheld or where ever.People who are going to use it as a personal computer?

Atari VCS is currently $200 (Official Atari store) and to get the PC mode USB thing $40 (Also, Official Atari store). It is probably one of the current cheapest upgrade-able minipc's for Raven2. This is a lot cheaper than getting something from System76, purism, Think Penguin, other MiniPC+Linux vendors. As they can use game revenue to offset PC net profit loss.

It also gives them the benefit of just fitting under Apple, if they went MiniPC as well:
https://browser.geekbench.com/v5/cpu/18302986 <== Atari VCS
https://browser.geekbench.com/v5/cpu/19087977 <== Apple A16, (pointing towards Apple TV products but if they were actually aimed at the MiniPC market: TV has access to Apple Arcade. Which of course has a blend of phone-esque to console-esque games. Also, have to bring your own controller for Apple TV:Apple Arcade)

Atari upgrading to PHX2 and doing Streaming, Gaming, PC, etc stuff at least puts them on-par with Intel's cheapo minipcs and Apple's cheapo minipcs.
Games come after hardware releases. If there is going to be an improvement in figures of complexity with better 3D/2D framerates. A Phoenix2(RDNA3) VCS2 optimization would scale up or down from <-> Steamdeck(RDNA2) <-> XSS(RDNA2) <-> PS5(RDNA+) <-> XSX(RDNA2) <-> Gaming PC(RDNA/RDNA2/RDNA3), a lot easier than a VCS(Vega14-GCN) Raven2 optimization.

Atari VCS 1 (plagued by controversies, prior CEO) => Original launch window was Q4-2019 through Q1-2020, didn't become generally available till Q2-2021.
following industry <-- time between consoles -->
Atari VCS 2 (straight launch, sweeping VCS 1 under the rug, new CEO) => Launch window given +4 years = Q4-2023 through Q1-2024, bonus time till Q2-2025(avoiding this hopefully).
//New CEO -> ??? -> New Director of VCS* -> New Lead Architect of VCS*
* Not in known capacity, probably won't be till software is finished on VCS1. => No Crowdfunding.

The change away from mobile is to focus more on:
Nightdive's Remake/Remasters like System Shock Remake
Ziggurat's Remake/Remasters like Deadly Dozen Reloaded
But, for Atari games...

Convert an old game(that might not have been in 3D) into 3D. Then, first port is at least on one console and PC(before VCS2), then it is ported down to Atari VCS2(when it comes out). Visually beyond the scope of Haunted Houses in the Reimagined Series. It will take awhile before we see the actual games the PC/Console focus will actually lead to. As well as if Atari does use the PHX2 to extend their renewed VCS(translated: game console)/ST(translated: personal computer) ambitions.

Steamdeck: February 2022 +4 => February 2026, if Valve is following the industry. If not, they are more likely to follow Sony's rapid refresh cycles: 2020(7nm) -> 2021(7nm/lighter) -> 2022(6nm/lighter). In this, it is highly unlikely for Valve to swap out the Steamdeck processor for PHX2. Since, they would be going for a refresh with same exact specs and reduced power/heat like Oberon+: https://www.angstronomics.com/p/ps5-refresh-oberon-plus
With a February 2026 target it at least gives a window for porting down(Nintendo Sw1:20nm/Sw1+:16nm, same console different process type) or developing new(true SD2) on GAA-nodes.

Atari VCS has significant decay where the phone ports of 3D console games can run at 4K/1080p on Apple TV 4K, while it struggles at 720p~900p. Any extraneous work to make a port work would be mute if the hardware(aka using PHX2) was much closer to more modern hardware(2p+4e+>mid-speed iGPU).

Weighting benefits:
Steamdeck => Basically is already at Orin-Switch spec. Valve/Nintendo is already paired, no reason to switch to another chip of different spec than 4c/4wgp. There is no reason to downgrade to keep costs when valve highlights "great on deck" titles. Thus, anything lost on the hardware is given back by the cut of steam store. Which gives the opposite target, they will want an EVEN larger SoC than Phoenix1/Strix. Which will make them wait for mature GAA-nodes later in 2025+.
Atari VCS => Not paired with any current gen console, an upgrade is definitely desired. The minipc aspect is already way behind, Jasper Lake/Alder Lake-N/Alder Lake-M are pulling their cost-effective added-value weight. Thus, PHX2 has the highest benefit is for Atari not Valve. They also don't have billions of excess like Valve for a custom chip.

Distribution of sales:

Steamdeck1 + 64GB eMMC	$399/$400	Extreme minority of sales / Sold as loss
Steamdeck1 + 256GB NVMe	$529/$530	Minority of sales / Sold as profit
Steamdeck1 + 512GB NVMe	$649/$650	Majority of sales / Sold as profit

Given the above, the desire isn't to sale a worse model or a lite model. Since, the volume of sales by Gabe indicates preference of the best one.

A better SoC is preferred for big chunks of triple AAA:
Early millions $70 => $21 for every game license bought
Late millions $70 => $14 for every game license bought
It is up to developers/publishers selling on steam to have a SteamDeck mode(quality mode), since Valve isn't funding that. So a stronger, better SoC means less work for the developers/publishers who do decide to integrate a SteamDeck mode.

Steamdeck2 instead probably would prefer:
8c Zen4 (8p, 2p+6e, 4p+4e)
>8wgp RDNA3
$599 w/ 512 GB
$799 w/ 1 TB
$999 w/ 2 TB

Atari VCS Base (+ PC USB)	$200 (+$40), less on sales	Majority of sales / Sold as loss
Atari VCS All-in (+ PC USB)	$300 (+$40), ' ' '	Minority of sales / Sold as loss

Given the table above, the desire appears more for a preference of a minipc rather than a game console. This is probably a symptom of launching the hardware with a barebone operating system, etc stuff.

Steamdeck was built off the SteamMachines/SteamOS. Atari VCS was a race track to get the hardware out. The current VCS plan is software(Games and getting the OS basic features: Dashboard, AtariLive(XboxLive clone), etc. features) with a shift to newer hardware(More likely to use PHX2).

With this timeline:
January 2022 => Atari new hardware was referenced
July 2022 => Atari has new hardware set in action
Time continues -> Atari finishes their development of the PC OS(usb stick) and Game OS(eMMC).
Time continues -> PHX2 AMD launch or reveal
Mid-2023/Q3-2023+ => Atari announces PHX2-based VCS and it is to use the hybrid-variant PC/Game OS on NVMe. Straight to release, no crowdfunding shenanigans/controversies/etc.

It is much more believable that Atari would use the PHX2 over Valve's Steamdeck successor using it.

Your dream about PHX2 based Atari looks very unlikely.
Atari Calls it Quits on the Atari VCS

P.S. Next time please post something longer, this post of yours was just too short.

 

[NostaSeronx]

Your dream about PHX2 based Atari looks very unlikely.
Atari Calls it Quits on the Atari VCS

It is actually more likely now, rather than less likely. I was half-guessing timelines with the SSD/RAM swap rumors delaying the switch to PHX2. Which will not occur since the modern VCS1 is dead.

The Timeline:
1. Early 2022 => Hardware review begins, then COO Chief of VCS/Connected Devices got ousted.
// New hardware search starts, change to get an Atari-branded Linux operating system starts here as well.

2. Mid-2022 => Hardware review ongoing, information to VCS-backers that supply issues were getting worse.
// Little Phoenix (June 2022)

3. Early 2H2022 => Hardware supply would be shrunk down to 10~30% for 2023, effectively killing the VCS in its current state.
// Pay the same amount for 10~30% of prior years supply or just cancel the order.

This actually makes it easier for Atari to come out with a newer product with PHX2.

 

[TESKATLIPOKA]

It is actually more likely now, rather than less likely. I was half-guessing timelines with the SSD/RAM swap rumors delaying the switch to PHX2. Which will not occur since the modern VCS1 is dead.

The Timeline:
1. Early 2022 => Hardware review begins, then COO Chief of VCS/Connected Devices got ousted.
// New hardware search starts, change to get an Atari-branded Linux operating system starts here as well.

2. Mid-2022 => Hardware review ongoing, information to VCS-backers that supply issues were getting worse.
// Little Phoenix (June 2022)

3. Early 2H2022 => Hardware supply would be shrunk down to 10~30% for 2023, effectively killing the VCS in its current state.
// Pay the same amount for 10~30% of prior years supply or just cancel the order.

This actually makes it easier for Atari to come out with a newer product with PHX2.

I like your optimism.
I would think about Switch 2, but PHX2 is just too power hungry for that and they will likely use ARM once more.

 

[NostaSeronx]

I like your optimism.
I would think about Switch 2, but PHX2 is just too power hungry for that and they will likely use ARM once more.

Nintendo would be placed in the S3 group. So, the Switch2 would be a custom chip, not a mainstream one.

If they were going back to AMD, they would probably use a custom processor derived from Exynos2200. Or, base it off the "dream" Exynos SoC :: 2x Cortex-Xy/3x Cortex A-7xx(or -1x A-7xx and 4x A-5xx) and 4 RDNAx WGP processors.
Nintendo Route A: NVIDIA GPU on 8nm or shrink, extra cost of Nvidia through Samsung. (Linear relationship: Samsung -> Nvidia -> Nintendo :: ~$299 loss per chip after everything)
Nintendo Route B: AMD GPU on 3nm or shrink, no extra cost since it is direct to Samsung and re-utilizing the collaboration with AMD. (Service/triangular relationship: AMD -> Samsung -> Nintendo :: ~$120-160 loss per chip after everything)

PHX2 won't be used in game consoles/PCs with those that have a secure slice of billions in revenue from games.

 

[TESKATLIPOKA]

Lenovo will announce Phoenix based IdeaPad Pro 5 14 and 16 at CES 2023.
You can choose from R9 7840HS or R5 7640HS
and there is also dGPU option for the bigger one: RTX 3050 or RTX 4050.
They wrote 6GB Vram in official specs for 3050, I think It was meant for 4050.
It ended up like I thought, AMD + Nvidia instead of AMD + AMD.

Available from May 2023 at prices starting from €1,099.

 

[jpiniero]

They wrote 6GB Vram in official specs for 3050, I think It was meant for 4050.

A new 3050 model with 6 GB would not surprise me. The memory bus would have to be cut to 96-bit though. I guess either way that would be the case.