Question AMD Phoenix/Zen 4 APU Speculation and Discussion

Page 77 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
22,902
12,971
136
Let's ask Copilot what Win 12 features will run on local inference hardware:
Okay, as far as AI code generation goes, Copilot can already do this with a cloud-based interface, though it's kinda slow and I don't know if it has key limits like ChatGPT. However for the photo editing and file organization I can see it using inference hardware.

I have heard about the TOPs limit for Win12 but I was unsure of exactly why that limit would exist. That sheds some light on the matter. Surely AMD has a reason for making Hawk Point (and advertising it), and perhaps the limit isn't 45 TOPs after all, or perhaps Win12 is a little more flexible than that.
 
  • Like
Reactions: lightmanek

FlameTail

Diamond Member
Dec 15, 2021
4,384
2,761
106
Hypothetical Question:

So we all know how 3D V-cache boosts gaming performance. As it stands though, the V-cache is an extension to the CPU L3.

What if the V-cache existed as an LLC/SLC? Will we see similar performance gains in games if that case?

This might be happen in a future APU where the V-cache acts as an SLC/LLC that is shared across all processing blocks of the APU.
 

SteinFG

Senior member
Dec 29, 2021
732
869
106
Hypothetical Question:

So we all know how 3D V-cache boosts gaming performance. As it stands though, the V-cache is an extension to the CPU L3.

What if the V-cache existed as an LLC/SLC? Will we see similar performance gains in games if that case?

This might be happen in a future APU where the V-cache acts as an SLC/LLC that is shared across all processing blocks of the APU.
The question is, would everyone pay 100 dollars to get that. At this point, just get a real GPU
 

Gideon

Platinum Member
Nov 27, 2007
2,030
5,035
136
It's already there: https://github.com/HoloISO/holoiso

If you want an official source Valve promised ages ago to finally update https://store.steampowered.com/steamos/buildyourown (which is very outdated at this point) but Valve is using Valve time again.

In reality i would suggest ChimeraOS (for both handhelds and DIY consoles):

It's essentially the same thing (initial boot is to the Steam Deck mode) but works much better OOtB and IMHO has more features (you can even run select GoG and Epic games for instance with some work):


They also test it on other handhelds and while it has it quirks, it's pretty usable
 

Timorous

Golden Member
Oct 27, 2008
1,978
3,864
136
The “4” in 8945 means Zen 4. All Zen 5 parts will use 5 as the third number (8955hx, for example). 9xxx will be next year.

People (including Intel) are hating on AMD for launching mixed generations of products, but I see nothing wrong with it.

It tells you what Zen is in the product upfront so from that perspective it is good.

It has weird hiccups though with the 7440U being a far better part for general usage and igp gaming than the 7520U is although I cannot find a 7440U laptop to check the pricing vs 7520U laptops, entirely possible they occupy totally different pricing segments.

Personally if they want year 1st so OEMS have a new shiny to sell each year then AMD should have swapped brand and architecture.

Then Zen2 will always be 72xx, Zen3 would be 73xx and Zen4 would be 74xx. For the class of 2024 8000 range it would be 84xx and 85xx when they launch later. Within that the XX would be the sub tier indicator. Looking at Mendochino the stack would be 7220U, 7230U and 7250U. Then when compared to the 7440U it makes more sense.

I think that would be much cleaner. Sure you might have cases where the 7380 8c part is faster in MT than the 7440 4c part but for single threaded apps the 7440 should be faster. Also given the combination of architecture, core count and TDP there are always going to be naming quirks IMO.
 

FlameTail

Diamond Member
Dec 15, 2021
4,384
2,761
106
It tells you what Zen is in the product upfront so from that perspective it is good.

It has weird hiccups though with the 7440U being a far better part for general usage and igp gaming than the 7520U is although I cannot find a 7440U laptop to check the pricing vs 7520U laptops, entirely possible they occupy totally different pricing segments.

Personally if they want year 1st so OEMS have a new shiny to sell each year then AMD should have swapped brand and architecture.

Then Zen2 will always be 72xx, Zen3 would be 73xx and Zen4 would be 74xx. For the class of 2024 8000 range it would be 84xx and 85xx when they launch later. Within that the XX would be the sub tier indicator. Looking at Mendochino the stack would be 7220U, 7230U and 7250U. Then when compared to the 7440U it makes more sense.

I think that would be much cleaner. Sure you might have cases where the 7380 8c part is faster in MT than the 7440 4c part but for single threaded apps the 7440 should be faster. Also given the combination of architecture, core count and TDP there are always going to be naming quirks IMO.
You should be hired as AMD's Product Naming Officer
 
  • Like
Reactions: alexruiz

Thibsie

Golden Member
Apr 25, 2017
1,127
1,334
136
The “4” in 8945 means Zen 4. All Zen 5 parts will use 5 as the third number (8955hx, for example). 9xxx will be next year.

People (including Intel) are hating on AMD for launching mixed generations of products, but I see nothing wrong with it.

Yeah, 'cos Intel never does that...
 

qmech

Member
Jan 29, 2022
82
179
66
It has weird hiccups though with the 7440U being a far better part for general usage and igp gaming than the 7520U is although I cannot find a 7440U laptop to check the pricing vs 7520U laptops, entirely possible they occupy totally different pricing segments.

The 7520U might be easier to find as the (exact same) 7520C, with the "C" denoting Chromebook, although BestBuy is absolutely willing to screw you over with an overpriced HP Windows laptop offering:


$630 is a steal for that. No question as to who is doing the stealing and who is getting stolen from. And for that you obviously also get Windows 10 in "S" mode.

Or if you want the "incredible performance" of something "designed to provide an ultrafast and smooth experience", you could go with an ASUS Vivobook Go 15 for a list price of $550, currently on sale for the amazing low, low price of just $459:


The Vivobook 15 Go offers you the choice between one of the pure e-core i3 chips and the 7520. At a pricepoint of $400+, neither laptop should probably exist. I honestly don't understand why laptop OEMs feel the need for so many different models, nor do I understand why the pricing is all over the place.

In any event, the 7520U/C just should not be called a Ryzen 5. The processor should have been simply the 7325U, which fairly denotes that it is a slightly faster variant of the 7320 (15% higher base, 5% higher boost).

To me, this is perfectly consistent with AMD's stated definition of the least significant digit, but in fairness to AMD, it should be noted that their usage so far has always been one where the -0 and -5 used completely different silicon, although with roughly similar performance.
 

Timorous

Golden Member
Oct 27, 2008
1,978
3,864
136
The 7520U might be easier to find as the (exact same) 7520C, with the "C" denoting Chromebook, although BestBuy is absolutely willing to screw you over with an overpriced HP Windows laptop offering:


$630 is a steal for that. No question as to who is doing the stealing and who is getting stolen from. And for that you obviously also get Windows 10 in "S" mode.

Or if you want the "incredible performance" of something "designed to provide an ultrafast and smooth experience", you could go with an ASUS Vivobook Go 15 for a list price of $550, currently on sale for the amazing low, low price of just $459:


The Vivobook 15 Go offers you the choice between one of the pure e-core i3 chips and the 7520. At a pricepoint of $400+, neither laptop should probably exist. I honestly don't understand why laptop OEMs feel the need for so many different models, nor do I understand why the pricing is all over the place.

In any event, the 7520U/C just should not be called a Ryzen 5. The processor should have been simply the 7325U, which fairly denotes that it is a slightly faster variant of the 7320 (15% higher base, 5% higher boost).

To me, this is perfectly consistent with AMD's stated definition of the least significant digit, but in fairness to AMD, it should be noted that their usage so far has always been one where the -0 and -5 used completely different silicon, although with roughly similar performance.

The 7520 is 4c8t where as the 7320 is 4c4t. There is a bit more than just higher clocks.

I agree the 7520 should not be Ryzen 5 though, That should be 6c12t CPUs, R3 7420 would be a better name to reflect the spec bump vs the 7320.

You could also do as you suggest. Have the 2c2t and 2c4t be Athlon 7120/7125 and the 4c4t/4c8t be Ryzen 3 7320/7325. Gives you differentiation without contaminating the R5 branding.

So yes. AMD naming could be better in a number of ways. I still think the fact you know what iteration of Zen you are getting is valuable and regardless of if the naming is sub optimal for consumers the fact that information is transparent is a strong benefit in my book.
 
  • Like
Reactions: Thibsie

Thibsie

Golden Member
Apr 25, 2017
1,127
1,334
136
Ryzen 5 should be 6 core minimum as on desktop, sometimes without SMT if "needed" but still 6 cores.
 

qmech

Member
Jan 29, 2022
82
179
66
The 7520 is 4c8t where as the 7320 is 4c4t. There is a bit more than just higher clocks.

According to AMD, both are 4 core, 8 thread parts:


I agree the 7520 should not be Ryzen 5 though, That should be 6c12t CPUs, R3 7420 would be a better name to reflect the spec bump vs the 7320.

AMD doesn't have a Ryzen 4 series and making this the sole member of that set seems odd.

AMD naming could be better in a number of ways. I still think the fact you know what iteration of Zen you are getting is valuable and regardless of if the naming is sub optimal for consumers the fact that information is transparent is a strong benefit in my book.

I don't even really dislike AMD's naming scheme. Even as an "enthusiast", I have to continually look up both Intel and AMD model numbers to figure out what the heck is hiding behind it. What is the chance that a "regular" consumer is going to have any idea without just doing a nanoreview/notebookcheck comparison, which makes any naming scheme moot? I do think that *one* chip is misplaced in the hierarchy, but whatever.

The best way to reduce complexity would probably be to reduce the number of SKUs. Do we *really* need 27 different 7000-series processors? (not counting 7040S variants and handheld Z variants). I understand that some kind of binning and market segmentation is needed, but 27 is almost certainly overkill. It adds consumer confusion, it adds support complexity, it adds production costs, it adds development costs (for OEMs), etc.
 

Timorous

Golden Member
Oct 27, 2008
1,978
3,864
136
According to AMD, both are 4 core, 8 thread parts:




AMD doesn't have a Ryzen 4 series and making this the sole member of that set seems odd.



I don't even really dislike AMD's naming scheme. Even as an "enthusiast", I have to continually look up both Intel and AMD model numbers to figure out what the heck is hiding behind it. What is the chance that a "regular" consumer is going to have any idea without just doing a nanoreview/notebookcheck comparison, which makes any naming scheme moot? I do think that *one* chip is misplaced in the hierarchy, but whatever.

The best way to reduce complexity would probably be to reduce the number of SKUs. Do we *really* need 27 different 7000-series processors? (not counting 7040S variants and handheld Z variants). I understand that some kind of binning and market segmentation is needed, but 27 is almost certainly overkill. It adds consumer confusion, it adds support complexity, it adds production costs, it adds development costs (for OEMs), etc.

huh, I read the spec sheets and kept thinking I saw 4c/4t. Really odd.

In the naming wheel 3/4 are both Ryzen 3 brandings.

13uiunlm5fca1.png


This is why there is a 7440U and a 7430U.

Since both 7320 and 7520 are the same c/t count then yea, 7320 and 7325 would be far better.
 

GunsMadeAmericaFree

Golden Member
Jan 23, 2007
1,392
379
136
Are production line yields for AMD and Intel processors pretty similar right now? A lot of folks have posted about power efficiency and cpu capability, but I find myself wondering about % yield, defects on the line, etc.
 

eek2121

Diamond Member
Aug 2, 2005
3,410
5,049
136
Have any 8020 parts been confirmed at this point?

8030 parts make for great chips to go in mini PCs. I use Cezanne for a plex Jellyfin server since the pi5 currently isn’t supported. Rock solid and handles my large household without so much as a fan spinup. An 8935 would probably be even better!

Are production line yields for AMD and Intel processors pretty similar right now? A lot of folks have posted about power efficiency and cpu capability, but I find myself wondering about % yield, defects on the line, etc.
AMD has excellent yields for all of their parts.
 

moinmoin

Diamond Member
Jun 1, 2017
5,242
8,456
136
In reality i would suggest ChimeraOS (for both handhelds and DIY consoles):

It's essentially the same thing (initial boot is to the Steam Deck mode) but works much better OOtB and IMHO has more features (you can even run select GoG and Epic games for instance with some work):


They also test it on other handhelds and while it has it quirks, it's pretty usable
Neat.

Is it also based on Arch Linux just like Steam OS? They unnecessarily seem to completely obfuscate what software base they actually use which is frowned upon among Linux distributions.

Also I'm running most Epic and GoG games just fine on my Steam Deck using Heroic Games Launcher installed through already set up Flathub. And EmuDeck for all things emulation.

So right now I'm kind of confused what ChimeraOS is actually doing, just put together existing efforts on Steam Deck as well as HoloISO and offer a polished distribution? The obfuscation seems to want to pretend it's more than yet another distribution.

It's not on distrowatch.com (great place for getting more info about any distribution out there). The only piece of (very old) text about ChimeraOS there is https://distrowatch.com/weekly.php?issue=20220606&mode=68 which is full of red flags to me.
 
  • Like
Reactions: Bigos

jpiniero

Lifer
Oct 1, 2010
16,810
7,253
136
The best way to reduce complexity would probably be to reduce the number of SKUs. Do we *really* need 27 different 7000-series processors? (not counting 7040S variants and handheld Z variants). I understand that some kind of binning and market segmentation is needed, but 27 is almost certainly overkill. It adds consumer confusion, it adds support complexity, it adds production costs, it adds development costs (for OEMs), etc.

The problem is that MLID. That's why AMD has been selling the old stuff mostly... and why Little Phoenix was needed to make Big Phoenix viable.
 

qmech

Member
Jan 29, 2022
82
179
66
huh, I read the spec sheets and kept thinking I saw 4c/4t. Really odd.

In the naming wheel 3/4 are both Ryzen 3 brandings.

Apparently, neither one of us can read. I never noticed that the 4 and 6 segment digits were part of Ryzen 3 and Ryzen 5, respectively. Oh well.
 
  • Like
Reactions: Timorous

Timorous

Golden Member
Oct 27, 2008
1,978
3,864
136
Have any 8020 parts been confirmed at this point?

8030 parts make for great chips to go in mini PCs. I use Cezanne for a plex Jellyfin server since the pi5 currently isn’t supported. Rock solid and handles my large household without so much as a fan spinup. An 8935 would probably be even better!


AMD has excellent yields for all of their parts.

Just 8040 and 8045 at the moment.
 

DrMrLordX

Lifer
Apr 27, 2000
22,902
12,971
136
Are production line yields for AMD and Intel processors pretty similar right now?
Depends on the node. Plus Intel doesn't like to publish yields anymore (if they ever did).

Intel 7 "super" (13900k, 14900k, similar) should be pretty good. Intel 4? Ehhhhhhhhh.

However all of the nodes AMD uses right now (in particular, TSMC N6 and N5P) are yielding extremely well, and N4P (Zen5) should also be doing very well.
 

Gideon

Platinum Member
Nov 27, 2007
2,030
5,035
136
Neat.

Is it also based on Arch Linux just like Steam OS? They unnecessarily seem to completely obfuscate what software base they actually use which is frowned upon among Linux distributions.

Also I'm running most Epic and GoG games just fine on my Steam Deck using Heroic Games Launcher installed through already set up Flathub. And EmuDeck for all things emulation.

So right now I'm kind of confused what ChimeraOS is actually doing, just put together existing efforts on Steam Deck as well as HoloISO and offer a polished distribution? The obfuscation seems to want to pretend it's more than yet another distribution.

It's not on distrowatch.com (great place for getting more info about any distribution out there). The only piece of (very old) text about ChimeraOS there is https://distrowatch.com/weekly.php?issue=20220606&mode=68 which is full of red flags to me.

Yeah it's arch based and an immutable distro, meaning the kernel, mesa and packages are locked down by default - managed by the os makers just like SteamOS itself (you can still install flatpak stuff). It actually implemented the idea before SteamOS 3 did it.

And it isn't using HoloISO at all (it predates it) and is quite different (immutable , gnome desktop vs kde, etc). Here is a very good review (though some things have improved since):


For starters, in case you’re not aware of what ChimeraOS is, it’s an Arch-based Linux distribution that’s designed to run Steam’s Big Picture interface on startup. The idea is you can transform your desktop PC into a console by having a controller-driven interface, without ever needing to connect a keyboard or mouse. Basically, it was the much-needed spiritual successor to the old SteamOS 1.0/2.0, that got rid of a lot of the headache that came with SteamOS prior to version 3.0. It also provided an enhanced experience with things like the Chimera web app, which allows the user to add ROMs and certain Flatpak applications with another device connected to the same network.


ChimeraOS also comes with enhanced security by keeping the core filesystem immutable. Updates to ChimeraOS – things like the kernel, the graphics driver, etc. – are handled entirely by the ChimeraOS team and not the end user. This allows for curated packages, ensuring they’re properly tested and working before they are shipped to the end user. If something goes wrong, it’s usually because of the ChimeraOS team, not the user.


That being said, users can still make use of their own software through the use of Flatpaks. These Flatpaks can be added through the aforementioned Chimera app, or by using the GNOME Software Center in Desktop Mode. Users can also unlock the file system via a terminal command, and install packages with pacman.


Sound familiar? That’s because that is pretty much what SteamOS 3.x does. Like ChimeraOS, SteamOS 3 is based on Arch, uses an immutable file system, and makes use of software outside of Steam via Flatpaks. So here’s an interesting tidbit that I’d like you to know: the ChimeraOS team were the first ones to implement this idea. Valve quietly copied these ideas into SteamOS 3, without ever acknowledging the ChimeraOS team. It seriously makes me wonder what SteamOS 3 would have been like if it weren’t for ChimeraOS. ChimeraOS founder, Alesh Slovak, even hinted at this when I interviewed him back in April.

All-in-all the developers actually take the time to run it on other handhelds (e.g. Rog Ally) and make plenty of fixes to those.


It also isn't a black box. All of their code is on github. The image building repo is here, where you can see what set of kernel / drivers, etc they are using:
 

Rigg

Senior member
May 6, 2020
710
1,805
136
Neat.

Is it also based on Arch Linux just like Steam OS? They unnecessarily seem to completely obfuscate what software base they actually use which is frowned upon among Linux distributions.

Also I'm running most Epic and GoG games just fine on my Steam Deck using Heroic Games Launcher installed through already set up Flathub. And EmuDeck for all things emulation.

So right now I'm kind of confused what ChimeraOS is actually doing, just put together existing efforts on Steam Deck as well as HoloISO and offer a polished distribution? The obfuscation seems to want to pretend it's more than yet another distribution.

It's not on distrowatch.com (great place for getting more info about any distribution out there). The only piece of (very old) text about ChimeraOS there is https://distrowatch.com/weekly.php?issue=20220606&mode=68 which is full of red flags to me.
Bazzite is also an option. Its an OCI image based on Fedora Silverblue/uBlue (immutable base system) and uses a "cloud native" approach to install all of the gaming stuff in an arch container. It's in interesting approach and works well. I run the GNOME version on an old 960 pro i had laying around as an alternate boot drive. If I'm running a game from my steam library I boot it into that. Game Pass is about all that keeps me from ditching windows for my gaming rig at this point. Windows is becoming unbearible to me as it gets more bloated and invasive every time it gets feature updates.