AMD is Supposedly Launching More Zen 3 Ryzen 5000X3D CPUs with 3D V-Cache on Socket AM4

[BiG StroOnZ]

Summary

AMD is looking to expand its Socket AM4 Ryzen 5000X3D processor lineup, according to Greymon55, a reliable source with AMD rumors. The 3D V-Cache lineup could see an expansion, with announcements expected in July.

 

Quotes

Quote

AMD is not entirely finished with its current AM4 socket platform; as something of a last hurrah, AMD will release one final series of processors to give the AM4 socket the send-off it so thoroughly deserves.

 

Well-known leaker Greymon55 tweeted that we'll get information on "several new" Zen3D products next month, along with details of some new lower-end parts for Socket AM4.

 

We could see one final series of AM4-based processors including the introduction of SKUs such as the Ryzen 5 5600X3D, Ryzen 9 5900X3D, or perhaps even the 5950X3D; with the latter two featuring a mind-boggling 200MB of Total Cache (L2+L3).

 

Arguably more interesting would be something like a 5600X3D. As the largest gains from 3D V-cache were found in gaming, and as gaming generally doesn't care about cores past six or eight, a reasonably-priced Ryzen 5 5600X3D could be a price-performance winner for gaming systems.

 

My thoughts

AMD said themselves at Computex that the AM4 platform wasn't going anywhere, saying AM4 "will continue for many years to come". This makes this alleged claim that they plan to release more 3D V-Cache products on AM4 as quite sensible. It allows people who are on AM4 an upgrade path that aren't ready to jump ship to AM5, DDR5, and 7000 Series. Dr. Lisa Su confirmed that AMD will continue supporting its AM4 platform for years to come, so it makes sense to offer a range of 3D V-Cache products like the: 5600X3D, 5900X3D and 5950X3D to help expand the Zen 3 5000 series lineup. Many people wondered when the 5800X3D launched why there wasn't a 5900X3D or 5950X3D, because those two chips would allow the gaming performance increases seen with the 3D V-Cache, as well as the productivity prowess of the original 5900X and 5950X. I think the only outlier here is that if the 5900X3D and 5950X3D performs too good, it could hinder Zen 4 sales. This is why I personally see a 5600X3D chip as the more likely scenario (for now), since it can make use of failed 5800X3D dies without crippling Zen 4 sales; since those customers weren't probably planning on going that route anyway (simply looking for the best price to performance upgrade option). This alleged reveal is going to occur sometime in July, so hopefully we will find out soon whether or not this claim is accurate.  

 

Sources

https://hothardware.com/news/amd-allegedly-launching-more-am4-cpus-with-3d-v-cache

https://www.techpowerup.com/296392/amd-readies-more-ryzen-5000x3d-processors

https://www.eteknix.com/amd-to-expand-ryzen-5000x3d-series-new-models

[WereCat]

I disagree that 5600X3D makes sense. The price would compete with 5700X/5800X and the whole point of picking up 6-core is to save some money and this would make it the most expensive 6-core which would now have to compete with 12th gen i5 that are much cheaper.

 

I actually look forward to 5950X3D as someone with 3900X. I don't want 5800X3D because I will lose multi-thread performance but I want that single-threaded boost. 5950X3D is a perfect upgrade path for me.

 

edit:

 

Do we know for sure that AMD Zen4 IPC incrase is based on comparison vs their non-3D Vcache CPUs?

[rotexen]

i'm not going to upgrade my 5900X in 3 years

[Levent]

How does this X3Dfying work? They take wafers of regular Zen3 CPUs, pick the binned ones and slap extra cache on them? Or is this a completely different wafer and manufacturing method?

 

I wonder, if we will ever see quad core AM4 cpus with extra cache. Those would make superb low power server cpus or heck even gaming laptop cpus.

[Vishera]

AMD said that it doesn't rule out releasing more stuff for AM4 in the future and have AM4 and AM5 co-exist.

[TrigrH]

32 minutes ago, Levent said:

How does this X3Dfying work? They take wafers of regular Zen3 CPUs, pick the binned ones and slap extra cache on them?

Yes, so unless cores die after the cache is stacked we wont be seeing 5600x3d or 5900x3d SKUs. 99% of the volume of this product is for Milan - X.

 

I call BS on this rumor:

 

[SimplyChunk]

10 minutes ago, TrigrH said:

Yes, so unless cores die after the cache is stacked we wont be seeing 5600x3d or 5900x3d SKUs. 99% of the volume of this product is for Milan - X.

 

I call BS on this rumor:

 

Where is this guy in the video from?  Is it an accent he has?  He kinda octive hops while he talks like a bad Christopher Walken impersonator..

 

 

[KaitouX]

3 hours ago, WereCat said:

Do we know for sure that AMD Zen4 IPC incrase is based on comparison vs their non-3D Vcache CPUs?

Yes, the footnotes mentioned the 5950X. And it probably wouldn't change anything, as the applications used don't really benefit from the bigger cache.

[cmndr]

6 hours ago, WereCat said:

I disagree that 5600X3D makes sense. The price would compete with 5700X/5800X and the whole point of picking up 6-core is to save some money and this would make it the most expensive 6-core which would now have to compete with 12th gen i5 that are much cheaper.

It would also be competing against a 12600k.

There can certainly be a place for a 5600x3D but it'd be kind of niche. Think Factorio builds.
 

[porina]

The bigger question is how does AMD intend to split AM4/AM5 in the near future? Will Zen 4 be an AM5 exclusive? With an appropriate IOD I don't see why that couldn't go to AM4 also, or even Zen 4 APUs at a later date. A transition generation on both platforms would be nice to have, but by the time Zen 5 comes around then I'd expect it to be 100% AM5.

 

If they co-exist, to me it seems logical to focus on AM4 as the value option going forwards, and AM5 is the premium platform. Following on from that, having lower end 3D models wouldn't necessarily make sense, unless they're targeting a specific niche.

 

As a wild idea, does the vcache have to be 64MB per CCD? For example, what if they make a 32MB vcache module? It may potentially halve silicon cost, although the manufacturing cost is probably not affected. This could give them more granularity in optimising price/performance points.

[leadeater]

18 minutes ago, porina said:

The bigger question is how does AMD intend to split AM4/AM5 in the near future? Will Zen 4 be an AM5 exclusive? With an appropriate IOD I don't see why that couldn't go to AM4 also, or even Zen 4 APUs at a later date. A transition generation on both platforms would be nice to have, but by the time Zen 5 comes around then I'd expect it to be 100% AM5.

Can't see it happening, AMD said Zen 4 would be DDR5 only. They could design a different IOD and APUs that actually have DDR4 controllers but that's a lot of effort and cost and then also a ton of validation on AM4 chipsets to actually allow these to run, possibly even requiring new chipset revisions.

 

Zen 4 on AM4 is an extreme long shot.

[porina]

5 minutes ago, leadeater said:

AMD said Zen 4 would be DDR5 only.

I missed that. It was more "could they?", not "would they?". To expand on that, if it were done I wouldn't expect it at launch, but may be an option for much later down the line.

 

My main thinking is that we don't have the details on how Zen 4 is constructed. Specifically the connection between CCD and IOD. One scenario is they use the same IF as Zen 3, and if so, even the existing AM4 IOD could suffice. The big problem with that is it wouldn't really make good use of DDR5 BW on AM5, so they may use a different-enough connection in AM5 and require a new IOD. I'd agree a new AM4 IOD is unlikely outside of possibly cost saving if they expect to make enough for long enough.

[mr moose]

AM4 and AM5 are starting to echo AM2+ and AM3 days.  One only supports one ram type while the other only supports certain CPU types.   

[WereCat]

1 hour ago, porina said:

 

 

If they co-exist, to me it seems logical to focus on AM4 as the value option going forwards, and AM5 is the premium platform.

Agree. As of right now it makes little sense to release Ryzen 3 and maybe even Ryzen 5 on AM5 since the price of DDR5 is so high. Anyone willing to spend that much for RAM is also likely to spend at least enough to get an 8-core and more.

 

59 minutes ago, leadeater said:

Can't see it happening, AMD said Zen 4 would be DDR5 only. They could design a different IOD and APUs that actually have DDR4 controllers but that's a lot of effort and cost and then also a ton of validation on AM4 chipsets to actually allow these to run, possibly even requiring new chipset revisions.

 

Zen 4 on AM4 is an extreme long shot.

They managed to take Zen2 CPU and slap LPDDR5 on it (Steam Deck) so they can likely do it the other way around.

The issue here is that if it's worth the performance penalty. Zen4 is supposed to have much faster DRAM and Infinity Fabric but going backwards to DDR4 and Zen3 Infinity Fabric speeds could result in worse performance than Zen3 if not optimised well. I don't think it's the issue of them not being able to mix and match the IOD dies between different gens of Zen architecture. That's supposed to be the big advantage of Zen, the modular design allows them to experiment as well.

[leadeater]

18 minutes ago, WereCat said:

They managed to take Zen2 CPU and slap LPDDR5 on it (Steam Deck) so they can likely do it the other way around.

Yep but that was AMD Semi-Custom Division, who will basically do anything a customers pays for. It's not really indicative of what AMD would actually do.

 

18 minutes ago, WereCat said:

I don't think it's the issue of them not being able to mix and match the IOD dies between different gens of Zen architecture. That's supposed to be the big advantage of Zen, the modular design allows them to experiment as well.

Modular yes, cross compatible not necessarily. It's not actually implied this is possible. Modularity and compatibility are different things.

 

Zen/Zen+ and Zen2 IFOP (Infinity Fabric On Package) for example are not compatible from what I know.

 

AMD is also adopting UCIe, or at least supporting it, so if their chip interconnects rely on that now then definitely no cross compatibility.

[Taf the Ghost]

My read is we're confusing a few things. We might be getting 12 & 16c 3D parts, along with future APUs in the 6000 series coming to AM4. (Odds are the APUs will be OEM only rather than Retail.) That would make the most business sense, but we'll see.

[Vishera]

2 hours ago, porina said:

I missed that. It was more "could they?", not "would they?". To expand on that, if it were done I wouldn't expect it at launch, but may be an option for much later down the line.

 

My main thinking is that we don't have the details on how Zen 4 is constructed. Specifically the connection between CCD and IOD. One scenario is they use the same IF as Zen 3, and if so, even the existing AM4 IOD could suffice. The big problem with that is it wouldn't really make good use of DDR5 BW on AM5, so they may use a different-enough connection in AM5 and require a new IOD. I'd agree a new AM4 IOD is unlikely outside of possibly cost saving if they expect to make enough for long enough.

According to the interview with Robert Hallock (Director of Technical Marketing - AMD) , AM4 and AM5 will co-exist and probably more CPUs will be released for AM4:

Quote

Antony: Is the 5800X3D the last processor we’ll see launched on Socket AM4?

 

Robert: I would say probably not. I don't really know what we plan to do with Socket AM4 but I think you saw Lisa talk about AM4 will continue, it will live on and it certainly has huge demand both from DIY builders and system customers. Could there be more AM4? Probably? But I don;t have anything specific to say on that.

 

Antony: Are socket AM4 and Ryzen 5000 going to sit alongside Socket AM5 in the same way we’ve seen Ryzen 3000 sit alongside Ryzen 5000?

 

Robert: Yeah you'll absolutely see AM4 live on in parallel carrying lower price points and more mainstream options. That's not to say Zen 3 is slow by any means. There's definitely an opportunity for it going forwards.

The "That's not to say Zen 3 is slow by any means. There's definitely an opportunity for it going forwards." is probably a hint that AM4 will not get Zen 4 CPUs.

[Doobeedoo]

That's... rather odd move though. I mean cool I guess. But, why so close to AM5 you'd think that would cannibalize their new product line. Unless they know they can 3D cache existing chips and not worry much for AM5 users that tend to wait. Also with that that 3D cache AM5 chips will release much sooner.

[StDragon]

6 hours ago, leadeater said:

Can't see it happening, AMD said Zen 4 would be DDR5 only. They could design a different IOD and APUs that actually have DDR4 controllers but that's a lot of effort and cost and then also a ton of validation on AM4 chipsets to actually allow these to run, possibly even requiring new chipset revisions.

 

Zen 4 on AM4 is an extreme long shot.

Well, mobile has always been something of a generation behind. Maybe they will Frankenstein such a product for that market. ¯\_(ツ)_/¯

[cmndr]

9 hours ago, porina said:

I missed that. It was more "could they?", not "would they?". To expand on that, if it were done I wouldn't expect it at launch, but may be an option for much later down the line.

 

The "could they" would be tricky.

AM4 and AM5 are the same height for cooler compatibility but the die height of Zen 4 is different to accomidate and LGA form factor.

[porina]

54 minutes ago, cmndr said:

AM4 and AM5 are the same height for cooler compatibility but the die height of Zen 4 is different to accomidate and LGA form factor.

Do we know how thick the silicon is for Zen 4 vs Zen 3? The obvious solution would be to make the IHS different thicknesses over each as necessary, unless the difference was too great to allow that to work sufficiently.

 

I hadn't thought about it before, but what options are there on silicon thickness? Gut feel is that thinner is cheaper as you'd need less base silicon, limited by some practical physical thickness for both electrical and mechanical reasons. Can't see it being that different. Probably cheaper to make substrate thickness different instead.

[cmndr]

5 minutes ago, porina said:

Do we know how thick the silicon is for Zen 4 vs Zen 3? The obvious solution would be to make the IHS different thicknesses over each as necessary, unless the difference was too great to allow that to work sufficiently.

 

I hadn't thought about it before, but what options are there on silicon thickness? Gut feel is that thinner is cheaper as you'd need less base silicon, limited by some practical physical thickness for both electrical and mechanical reasons. Can't see it being that different. Probably cheaper to make substrate thickness different instead.

It'll probably vary by HOW thin it is. My suspicion is that if it's already pretty thin, going thinner means more defects since you're basically using a machine to sand down the die even more. If you're really thin adding on more requires more materials.

 

It would also likely depend on the design of the CPU, it's possible that certain design considerations are already made given an assumption of a certain height.

I suspect that height modifications get more finicky every single generation of manufacturing just due to ALL of the complexities involved. I could be wrong though.

[porina]

10 minutes ago, cmndr said:

It'll probably vary by HOW thin it is. My suspicion is that if it's already pretty thin, going thinner means more defects since you're basically using a machine to sand down the die even more. If you're really thin adding on more requires more materials.

Been digging around a bit on wafer thickness. Can't get a definitive answer but there is a trend of typical thickness going up with wafer size. Leading edge processes has been on 300mm for a long time so I doubt thickness has changed much. We'll have to wait for physical samples to get measured, but my gut feel is the silicon thickness for Zen 3 and Zen 4 aren't going to be that different if at all. Heatsink compatibility only requires the top of the IHS to be the same height above mobo, which includes all of the socket, substrate, silicon and IHS. I really don't think silicon thickness will be the deciding factor in that, and it is trivial to vary IHS thickness.

 

I found a der8auer video where he estimates the Zen 4 IHS to be 3.6mm thick and doesn't seem to have a recess in the middle. From the image I think he's looking at the height of metal in a cutout, which doesn't reach the substrate as there's a small gap. So the height above substrate may be slightly more than that. Searching for similar on AM4 I get values of 3.0 to a bit over reported. Der8auer states 5950X is about 2.5-2.6mm in the middle where there is a recess, saying IHS thickness has grown by over 1mm. He speculates one reason is the Z height as we're talking about now. Another might be to do with higher heat density potential from Zen 4 needing more spreading to the heatsink.

 

Anyway, there may be many reasons why Zen 4 may not come to AM4, but I feel silicon thickness is way down the list of possible reasons.

 

For reference:

 

[cmndr]

Die thickness isn't to be underrated as an issue. On AM5 you can make the IOD and the CCDs the same thickness. With AM4 you're presumably working with older IODs. This means you can get away with a more or less uniform IHS.

 

It's not insurmountable (Zen 3D literally has an extra chip on top) it's just an extra barrier.

[porina]

11 minutes ago, cmndr said:

Die thickness isn't to be underrated as an issue. On AM5 you can make the IOD and the CCDs the same thickness. With AM4 you're presumably working with older IODs. This means you can get away with a more or less uniform IHS.

On AM4 the CCDs and IODs are made on different process nodes from different manufacturers. I'm assuming there is no thickness difference on IHS as I've never consciously looked. Either they come off the production line the same thickness, or one has to be processed to make it the same. The former just seems more likely, and I feel it is likely Zen 4 CCDs will also be practically the same. There's probably a standard thickness that will be used unless there's a good reason not to.

 

11 minutes ago, cmndr said:

It's not insurmountable (Zen 3D literally has an extra chip on top) it's just an extra barrier.

Don't disagree that if there is a height difference, it would be an extra step to consider. But I still think it is arguing over a point that is way down the list and largely unimportant. It's like me complaining I'm not a marathon runner because I don't have the right shoes, when the bigger problem is I never ran more than 300m in one go in my life.

 

I was wondering about the 3D cache CPUs earlier. Did they say if they thinned the dies, or if they're regular thickness dies and implicitly adjusted the IHS to suit?