Question AMD Rembrandt/Zen 3+ APU Speculation and Discussion

[izaic3]

Alright, so we've had some leaks so far. I don't know if any of it's been confirmed yet, as it's pretty early, but here is what I've surmised so far (massive grain of salt of course):

• Zen 3 "enhanced"
• DDR5 and LPDDR5 support
• PCIE4
• 6nm
• CVML
• Navi 2/RDNA/RDNA 2
• 12 CU
• Socket FP7/AM5
• Possible USB4
• 2021(unlikely) -2022 (more likely) release date

If if turns out to have RDNA 2 and 12 CU, I could see iGPU performance potentially almost doubling over Cezanne.

If I've made any mistakes or gotten anything wrong, please let me know. I'd also love to hear more knowledgeable people weigh in on their expectations.

 

[uzzi38]

@Shivansps Finally, a part you might not complain about the price of

 

[DrMrLordX]

Zen3 on 12nm? Weird. Would like to see benches of that.

 

[uzzi38]

Zen3 on 12nm? Weird. Would like to see benches of that.

12LP+ to be specific.

GlobalFoundries Unveils 12LP+ Technology: Massive Performance & Power Improvements

www.anandtech.com

 

[NostaSeronx]

12LP/12LP+ from what I understand should announce unavailability timeline by end of July. So, I doubt it is GlobalFoundries 12+, but more likely TSMC's 12+; https://n12e.tsmc.com/english/dedicatedFoundry/technology/N12e.htm

This is reflected by AMD's most recent WSA ditching 14nm/12nm FF at GloFo. Currently, AMD-ODC is only operating on 22nm at GlobalFoundries.

AMD project => The project is PNR to signoff ODC PROJECT. The project is carried on 12nm TSMC using Fin FET technology. Pretty much every FinFET project by ODC is at TSMC, so I give a big phat question mark to 12LP+.

 

[moinmoin]

Now introducing: Monet/MNT. Zen 3 () on GloFo's 12LP+.

4c, 2WGP (4CUs) RDNA2. It's the new Raven2 replacement.

If that turns into a real product it's very telling that porting back Zen 3 and RDNA2 to some 12nm node for a Raven/Dali successor is more profitable than the masks and validation just for creating smaller dies on N7/N6. It'd essentially tell us that the lower priced client SKUs AMD still did for Zen 2 (but not Zen 3) may well be gone for good.

 

[Abwx]

12LP/12LP+ from what I understand should announce unavailability timeline by end of July. So, I doubt it is GlobalFoundries 12+, but more likely TSMC's 12+; https://n12e.tsmc.com/english/dedicatedFoundry/technology/N12e.htm

This is reflected by AMD's most recent WSA ditching 14nm/12nm FF at GloFo. Currently, AMD-ODC is only operating on 22nm at GlobalFoundries.

AMD project => The project is PNR to signoff ODC PROJECT. The project is carried on 12nm TSMC using Fin FET technology. Pretty much every FinFET project by ODC is at TSMC, so I give a big phat question mark to 12LP+.

TSMC s 12nm is for low power devices, wich means lower frequencies than what is used in CPUs.

TSMC Launches New N12e Process: FinFET at 0.4V for IoT

www.anandtech.com

 

[NostaSeronx]

TSMC s 12nm is for low power devices, wich means lower frequencies than what is used in CPUs.

TSMC Launches New N12e Process: FinFET at 0.4V for IoT

www.anandtech.com

No, it is for HPC through IoT.

If you think that, then all nodes with FinFETs are low power. Since, their optimal range is low Vdd but not too low and mid Freq but not low or high.

N12e unlike prior TSMC FinFETs is as IP complete as N7. Logic, RF, Analog, etc. Thus, N12e allows anything on N7 to be backported. Since, those nodes are equally complete in IP.

 

[blckgrffn]

Zen3 on 12nm? Weird. Would like to see benches of that.

Ha, and now we can get the cheap 12nm RDNA2 GPUs we never knew we wanted

Seriously, something even to take on the 1030 would be fine even if impractical/never gonna happen.

I like cards to be available.

 

[Abwx]

No, it is for HPC through IoT.

If you think that, then all nodes with FinFETs are low power. Since, their optimal range is low Vdd but not too low and mid Freq but not low or high.

HPC can mean devices like GPUs, low frequencies and massively parallel computations.

Anyway that s a welcomed move from AMD, hope it will help getting out of the prices hikes frenzy since even Zen+ APUs are at higher prices than in their prime time...

 

[DisEnchantment]

Currently, AMD-ODC is only operating on 22nm at GlobalFoundries.

AMD project => The project is PNR to signoff ODC PROJECT. The project is carried on 12nm TSMC using Fin FET technology. Pretty much every FinFET project by ODC is at TSMC, so I give a big phat question mark to 12LP+.

Ooof!!! Those details are very specific.
By ODC you mean the Indian design team? They seems to have taken up a number of big implementation assignments lately.
Mind to share where is the source? LinkedIn?

Could be chromebooks and other industrial designs. Going Samsung would probably annoy TSMC.

 

[NostaSeronx]

Ooof!!! Those details are very specific.
By ODC you mean the Indian design team? They seems to have taken up a number of big implementation assignments lately.
Mind to share where is the source? LinkedIn?

Could be chromebooks and other industrial designs. Going Samsung would probably annoy TSMC.

ODC = Offshore Design Center, which is any design center not in the United States and usually uses contractors. It was just an example, that AMD's relationship with GlobalFoundries is at its lowest.

Fab 1 => 28nm products have been EOL'd w/ no continued manufacturing.
Fab 8 => 14nm/12nm products to be EOL'd w/ no continued manufacturing soon but definitely not sooner.

I haven't seen any big GlobalFoundries-FF/12LP+ pushes for AMD yet. Most of the new stuff is TSMC only.

The idea that AMD will go to GlobalFoundries for a backport is unlikely. Especially, since AMD has been aggressively de-ramping production at GlobalFoundries. The WSA does agree with this, indicating that IOD3 will probably be used more extensively.

 

[Shivansps]

@Shivansps Finally, a part you might not complain about the price of

To be fair, it is not possible to complain about any price after this mess, they can do anything now. If they want to charge $150 for that, who im to say anything when other APUs are going to be well over $200, it is going to beat the Raven2/Dali by a large margin, i saw Bristol Ridge coming back from the grave at what used to be 3200G/3400G prices, the compiting Celeron/Pentiums/I3 have a crap IGP and GPU oems are scrambling to produce GT730s for the entry level sector.

Thats a excellent Raven2/Dali replacement (if real), they are doing two backports here both Zen 3 and RDNA2, what is a little strange.

 

[blckgrffn]

Thats a excellent Raven2/Dali replacement (if real), they are doing two backports here both Zen 3 and RDNA2, what is a little strange.

If they are really back porting both I think that means that A)it’s a very long life part to justify the work and B)AMD is looking to add silicon in every way possible.

I guess that seems obvious, but I think it means they also see an opportunity for market share even with a process disadvantage (is it really a big disadvantage against 14nm from Intel that will likely see volume for years yet?)

I just deployed a little Gigabyte Brix with the embedded 25W Zen 1 quad and Vega APU and honestly paired with decent memory and a nvme drive it is… fine. It boosted easily into the mid 3 ghz and it was very acceptable for desktop use and using Adobe products like Illustrator.

If they can get all the performance enhancements of Zen 3 micro architecture and the uplift from Vega to RDNA 2 in efficiency and performance per mm of die size well along with a process improvement - it might be surprisingly performant too. I am still assuming GF because I bet they have a relationship there and GF seems like they might be really happy with a volume customer.

🤷‍♂️

 

[Shivansps]

If they are really back porting both I think that means that A)it’s a very long life part to justify the work and B)AMD is looking to add silicon in every way possible.

A 4C (i assume 4/8) Zen 3 with 4 CU RDNA2 may last A LOT OF TIME. Next year that thing could be priced at around $100 as a 3200G replacement, perfect to counter quad I3s, while 2C and/or cut down IGP version can be priced below $100 around the Pentium and Celeron prices. Same thing applies to Notebooks. Even 5 to 6 years into the future after that they could still use them in the Athlon series as entry level, not to mention the embedded market...

 

[DrMrLordX]

It'd essentially tell us that the lower priced client SKUs AMD still did for Zen 2 (but not Zen 3) may well be gone for good.

They can turn higher profits on every N7-family wafer they take from TSMC than they can from 4c parts and low-end APUs. Until that condition changes, you should expect more of the same. Once the market rejects 2018/2019 node products then they're free to do more budget stuff with those wafers.

 

[NTMBK]

So happy someone finally talked about this:

Now introducing: Monet/MNT. Zen 3 () on GloFo's 12LP+.

4c, 2WGP (4CUs) RDNA2. It's the new Raven2 replacement.

Is there a written article version of this?

 

[moinmoin]

TSMC s 12nm is for low power devices, wich means lower frequencies than what is used in CPUs.

TSMC Launches New N12e Process: FinFET at 0.4V for IoT

www.anandtech.com

That's actually an argument for the back port imo. Zen has sufficient performance for most use cases even at lower frequencies, and AMD showed by using mobile oriented libraries for its CPUs that that doesn't necessarily mean forced lower frequencies anyway. And higher energy efficiency/longer battery life is a selling point even in low budget computers so using a specialized older node for such an apparently long living product would make perfect sense.

Until that condition changes, you should expect more of the same.

And AMD appears to be of the opinion that that condition doesn't change in a long enough time that back porting is well worth the effort.

 

[Mopetar]

If they are really back porting both I think that means that A)it’s a very long life part to justify the work and B)AMD is looking to add silicon in every way possible.

Aren't the parts this chip would be replacing still nabbed by Global Foundries anyway? I don't know what kind of volume AMD has been producing (though the price increases even in low-end chips like the Athlon line and Intel's equivalent Celeron line suggest there's plenty of demand) but there's probably enough of one that trying to move the manufacturing over TSCM where it would have to compete with their other products for wafers isn't a move they can make right now.

 

[jpiniero]

Aren't the parts this chip would be replacing still nabbed by Global Foundries anyway?

Yes and no. As mentioned it's probably really meant for IoT but it could get into some laptops eventually.

GloFo wafers are probably a lot cheaper than anything TSMC.

 

[maddie]

Yes and no. As mentioned it's probably really meant for IoT but it could get into some laptops eventually.

GloFo wafers are probably a lot cheaper than anything TSMC.

4 Zen3 cores and 256 Navi2 shaders are needed for IoT. What the hell is this IoT device?

This is equivalent to 3400G. Faster CPU and slower GPU.

 

[scineram]

So happy someone finally talked about this:

Now introducing: Monet/MNT. Zen 3 () on GloFo's 12LP+.

4c, 2WGP (4CUs) RDNA2. It's the new Raven2 replacement.

I’m not buying that node. It makes absolutely no sense.

 

[soresu]

Now introducing: Monet/MNT. Zen 3 (

) on GloFo's 12LP+.

The big question is, areal density aside what is the power efficiency difference between 12LP+ and N7P?

 

[blckgrffn]

4 Zen3 cores and 256 Navi2 shaders are needed for IoT. What the hell is this IoT device?

This is equivalent to 3400G. Faster CPU and slower GPU.

My thoughts exactly. This is really a APU first - maybe that does mean soldered/embedded but for NUC style PCs and the $300 laptop crowd. It would extremely surprising to me if this really aimed at sub 5W power envelopes.

Also, vs the current Zen 1 based embedded solutions is probably fully Windows 11 compliant and gives AMD a platform for embedded solutions that might allow them to someday drop support for Vega. In like 2030 maybe but still you have to move on sometime

 

[soresu]

If that turns into a real product it's very telling that porting back Zen 3 and RDNA2 to some 12nm node for a Raven/Dali successor is more profitable than the masks and validation just for creating smaller dies on N7/N6. It'd essentially tell us that the lower priced client SKUs AMD still did for Zen 2 (but not Zen 3) may well be gone for good.

Something to bare in mind for 'porting' is that recent advancements in AI/ML augmented transistor placement have made it vastly faster and even more performant than doing it by human hands alone.

Combined with the lower complexity of 12nm and the capacity crunch this may have helped inform AMD's decision to back port a recent design for value parts.

 

[jpiniero]

4 Zen3 cores and 256 Navi2 shaders are needed for IoT. What the hell is this IoT device?

Edge device that needs GPU compute.

I do think they will make SKUs for laptops but I can't see them spending the effort on this for something that was intended mainly to compete with Atoms and Pentium Gold laptops, etc. If they thought Picasso would have competitiveness issues due to Vega lacking features it seems to me that it would be better to just get out. This is going to be a lot of work.