News AMD Confirms Ryzen 8000 AM5 Processors With Zen 5 CPU and Navi 3.5 Graphics

D

Deleted member 2838871

Guest
Sounds great, especially confirmation that AM5 will extend into 2026.

Amen. That's why I just built with AM5... a few years of future proofing with amazing thermals. My first AMD build since 2001.

PC is a beast... but if there's anything I'll upgrade in the next 4 years it will be the7950x3D CPU.
 
The RDNA 3.5 graphics architecture for the integrated GPU is actually the same RDNA GPU that will be powering the successor to Phoenix APUs codenamed "Strix Point".

And we are looking at at least four AM5 processor families for the new socket including the existing Zen 4-based Ryzen 7000 CPUs. Noice !

But it appears that AM5 platform will be receiving a new CPU core and graphics upgrade on an annual cadence, so yeah , we should be getting something new to try out each year. :)
 

salgado18

Distinguished
Feb 12, 2007
925
363
19,370
Still two CCDs for 16C? I imagined that by the time Zen hit 3nm, 8C/32MB CCDs would be so small that AMD would need to aggregate the thing into 16C/64MB.
Need and want are two separate things. They probably could already, but do they want to? 8-core chiplets are good enough for the current market, and if they want more, then it's easy to combine them. When the market/software landscape asks for more, they will probably do it, but it's more cost-effective to keep making 8 cores than going up just because. (I still support your idea, though)
 
  • Like
Reactions: artk2219

g-unit1111

Titan
Moderator
Amen. That's why I just built with AM5... a few years of future proofing with amazing thermals. My first AMD build since 2001.

PC is a beast... but if there's anything I'll upgrade in the next 4 years it will be the7950x3D CPU.

That's one thing I do love about AMD is that they are made to last. My B450 rig is still going strong after upgrading from a 2600 to a 5800X.

The only reason I didn't go with AMD for my new PC that I'm building is that they only had one X670 ITX board available, and it wasn't good. Hopefully they remedy this problem with the next gen and start making more SFF motherboards available.
 

InvalidError

Titan
Moderator
Need and want are two separate things. They probably could already, but do they want to? 8-core chiplets are good enough for the current market, and if they want more, then it's easy to combine them.
The recurring performance issues in mainstream software when two CCDs are involved disagree. Combining two CCDs may be a cheap way to increase core count but not an efficient one for software that has thread-to-thread dependencies (ex.: games) and aren't architecture-aware enough to manage multi-CCDs correctly.
 
  • Like
Reactions: artk2219

rluker5

Distinguished
Jun 23, 2014
598
366
19,260
This story seems pretty expected.
If AMD stopped making a new arch every other year it would be a surprise.

Where AMD claims a new core and graphics upgrade on an annual cadence is disingenuous. Adding a new core every other year then waiting a year to add 3d cache isn't a new core arch every year. Zen3d is still Zen3, Zen4x3d is still Zen4, just with the 3d cache added. I imagine most won't see year one or the other as an upgrade, depending on their use case.

Their illustration of Zen4 coming out in 2023 (in support of a new core every year maybe?) even though the same slide says 2022 is SO sloppy. People can count to two and remember the number of a year stated on the same slide.

Maybe I'm reading it wrong? It seems like AM5 will get Zen4 in 2022, Zen4x3d in 2023, Zen5 in 2024, Zen5x3d in 2025, Zen6 in 2026 and no Zen6x3d because AM5 is only going into 2026 per the slide and Zen6x3d is coming out in 2027. What a bunch of mess. If AMD does that much milking they can keep their longevity. It offends me not only that AMD would do it, but that people would fall for it.
 

mickrc3

Distinguished
Mar 20, 2016
86
21
19,865
Glad I read this article. It saved me from buying a 64GB upgrade for my system. I will get by for another year on 32GB. I may go ahead and upgrade from my Zen 3 system in 2024 or wait for 2025. Either way, I'm glad I chose to skip the Ryzen 7xxx family. It's the first time I showed any restraint on upgrades; even after I retired from computer programming I kept buying into the next Zen family.

I rode the release waves with the Ryzen, starting with a Ryzen Zen 1800X/AsRock X370 Fatality, then a Ryzen Zen+ 2700X/ASUS X470 Crosshair Hero VII, then a Ryzen Zen 2 3800X>3900X>Zen 35900X/AsRock X570 Phantom Gaming Pro. Each time I upgraded I moved my old equipment to other family members if I hadn't already built them a current system. I've been building PCs for more than 20 years. I caught the bug when I replaced my Intel 386 and motherboard with a 486SX bundle back in 1992.

Then we hit the self-imposed lull when the Ryzen 7xxxx chips came out and I had some requests from some family members for new systems (they were nieces and nephews mostly running FM2+ systems with A10 or A8 CPUs I built for them in the early 2010s) but because Ryzen 5xxxx chips were plentiful and DDR4 got cheaper by the day I moved ten family members to 5xxxx chips (5900X/5700X/5600X/5600) and X570/B550 motherboards.

Our niece who lives 1000 miles away had a request. Besides the system we built for her she wanted to help out a friend. So we gifted our niece's babysitter one of the old Ryzen systems left over from the upgrades. Her gamer friend helped unbox the system. He said he wanted to see what kind of junk system is just given away for free. He told the babysitter that he would help her get something working even if he had to raid his spare parts stash. Instead he was floored by what he found we had given to her. The system was a Ryzen 2700X on a ASUS Crosshair Hero VII wifi with 32GB of DDR4 3200, a 1TB WD 770 NVMe boot drive, a 2TB Seagate Barracuda data drive, a MSI Duke GeForce GTX 1080, mid-tower ATX case with five RGB fans, keyboard, mouse, Acer 23in 1080p monitor, and Windows 10 Pro.

Even though we could have given away one of the Ryzen 1xxx systems we decided not to due to the lack of Win 11 support with the first gen Ryzen CPUs.
 
Last edited:
  • Like
Reactions: artk2219

sherhi

Distinguished
Apr 17, 2015
79
51
18,610
This!!

I don't plan on jumping onto AM5 until 2024.
I'm pushing my almost 10 year old system as much as I can and it seems to be 2024 or 2025 for me as well, I need bigger monitor for more efficient home-office, why not upgrade to 1440p gaming while at it and then win10 will die as well. The thing is, in 2024/2025 that possible CPU upgrade up to 2026 on the same mobo is not that appealing to me, it seems too late, I don't really need to upgrade CPU after a year or two and maybe Intel will have compatible mobos for 2 years as well then.
AMD's 4 years look nice at first sight but then 1st year of new architecture? Nah thanks, I don't like to be test subject in general not just AMD. 2nd year is more polished, prices should be better...but do you need new CPU in another 2 years? Probably not. Maybe 6 years would be my cup of tea, buy very good mobo in 2nd year with maybe cheapest CPU and in 6th year go for best CPU possible.
 

NinoPino

Prominent
May 26, 2022
138
79
660
Still two CCDs for 16C? I imagined that by the time Zen hit 3nm, 8C/32MB CCDs would be so small that AMD would need to aggregate the thing into 16C/64MB.
I agree with you. Maybe the reason is that, in case of defects, isn't possible to disable a single core but only a full CCD (if someone have info...). If AMD cannot disable the single cores, probably a 16 core chip become too expensive for the low end segment ?
 

usertests

Distinguished
Mar 8, 2013
387
329
19,060
Their illustration of Zen4 coming out in 2023 (in support of a new core every year maybe?) even though the same slide says 2022 is SO sloppy. People can count to two and remember the number of a year stated on the same slide.
AnandTech speculated that it refers to unreleased Phoenix desktop APUs. Previously rumored for a late 2023 release. That would make sense with the slide saying Zen 4 + Navi 3. If Navi 3 on the slide is a typo, yikes.
 
Still two CCDs for 16C? I imagined that by the time Zen hit 3nm, 8C/32MB CCDs would be so small that AMD would need to aggregate the thing into 16C/64MB.
SRAM scaling on the newer version of N3, which is cheaper, matches current N5 so there are most certainly limits to how small these will be. The decision could also be related to manufacturing scale and how many different sized CCDs AMD plans on having. Right now they only have the two sizes (I've never seen solid info on mobile so I'm not sure if the lower core counts are harvested or if the monolithic design is just different) so if they were to go to a 16 core CCD they would have to have at least 3.
 
They would have to charge twice the money for a CPU with twice the cores, they can only order that many wafers and those only make so many CCDs, if they double the core count they lose half the money they are making, or at least that potential.

Also unless ZEN 5 uses half the power at the max all core clock of ZEN 4 then a ZEN 5 with twice the cores would run at lower all core clocks which would look bad on benchmarks. Twice the cores would result in much less than twice the performance and that would look like a regression of IPC.
And they can't increase the power anymore since ZEN 4 is already pushing the maximum of the platform.
 

InvalidError

Titan
Moderator
Also unless ZEN 5 uses half the power at the max all core clock of ZEN 4 then a ZEN 5 with twice the cores would run at lower all core clocks which would look bad on benchmarks. Twice the cores would result in much less than twice the performance and that would look like a regression of IPC.
And they can't increase the power anymore since ZEN 4 is already pushing the maximum of the platform.
All pretty erroneous conclusions: twice the cores and cache in one CCD would also be about twice the size, which keeps power density about exactly the same as having two smaller CCDs next to each other but would eliminate the CCD-to-CCD latency when threads that would be on different CCDs need to talk to each other, requiring that people either disable the second CCD or use applications to corral their games and non-architecture-aware software onto a single CCD. With twice as many cores to pick from without having to worry about CCD-to-CCD latency, there are also twice as many cores to bin for more efficient boost. The single-threaded performance should be mostly unchanged while multi-threaded throughput should be better for the same power.

The main reason not to do it is to let mainstream stagnate at 6-8 cores for a few more years. 2008-2017 was the age of "four cores is good enough", now we are in the 2018-2025 age of "eight cores is good enough" not counting Intel shovelling buckets of E-cores at its CPUs to keep them competitive and keep TDP in check.
 
All pretty erroneous conclusions: twice the cores and cache in one CCD would also be about twice the size, which keeps power density about exactly the same as having two smaller CCDs next to each other but would eliminate the CCD-to-CCD latency when threads that would be on different CCDs need to talk to each other,
Yes if there is zero size improvement from going to a smaller node then it would be twice the size which would be even worse for AMD...
The inter CCD latency makes zero difference to things like cinebench or video transcoding since you send the workload to the cores once and get the result, the cores don't talk to each other.
The power density also makes zero difference to how much power the platform can provide, that would be a concern on how to cool it if you could give it more power.
With twice as many cores to pick from without having to worry about CCD-to-CCD latency, there are also twice as many cores to bin for more efficient boost. The single-threaded performance should be mostly unchanged while multi-threaded throughput should be better for the same power.
That would matter to manual extreme overclockers, and to put "matter" into perspective it would be like maybe a 100Mhz difference at insane power draws.
For normal people using default settings everybody would get the same amount of max single threaded clock and it would be the one warranted by AMD.

Yes, multi-threaded throughput would be better for the same power, but how much?! If 100% more cores only get you 20-30% more multi-threaded throughput,because the new cores are 30-40% more efficient(just a random guess, did we get any leaks about this? ) , would that be good enough to get sales?!

The main reason not to do it is to let mainstream stagnate at 6-8 cores for a few more years. 2008-2017 was the age of "four cores is good enough", now we are in the 2018-2025 age of "eight cores is good enough"
Ahh, so now holding back is a good thing because AMD is doing it?!
If anyone could get a CPU out with 100% more performance at the same price they would sell gangbusters ,no matter if people need it or not, they would upgrade just because it exists.