Speculation: Spring refresh for Ryzen

[Vattila]

As I've posted elsewhere, I think AMD fell short of their frequency targets for the Ryzen 3000 series. A later launch than expected, statements relating to frequency challenges, Max Boost redefinition, boost issues, 3900X shortages, and now the 3950X delay from September to November, do all corroborate this. I guess they aimed for 5 GHz, but fell ~8% short.

However, TSMC is firing on all cylinders. Amongst a number of process roadmap announcements, they now offer N7P as a refinement of their N7 process on which the Zen 2 chiplet is built. N7P is compatible with N7 design rules, so it should provide a fairly simple and cost-effective opportunity to optimise existing designs.

With that in mind, is there a chance that AMD will do a spring refresh based on a faster stepping of the CPU chiplet?

 

[uzzi38]

3750 and 3850
Could be ported to N7P for some minor performance bump
Doubt it, but wouldn't surprise me

Afaik N7P doesn't require like a retape or anything, so yeah. Hesitate to call it a 'port' though.

That being said, if AMD are doing this, they they would be doing this to all Matisse CCDs, not just for these few parts.

 

[moinmoin]

Reason is that Zen 3 will be the start of the interposer era for CPU packaging, and it will initially be high-end and expensive, with initial capacity only sufficient to cover server (relatively low unit numbers) and high-end desktop (likewise).

I don't really think this is really feasible. AMD's slightly longer than a year cascade doesn't give enough breathing room for a new packaging approach to be inhibitively expensive at the beginning and still approach mainstream scale at the end within those months.

I don't really trust those talks of a Matisse refresh. Though those being related to the new faster Epyc chips, and all of them making use of N7P without any actual silicon level changes could make sense.

 

[BTRY B 529th FA BN]

AMD Ryzen 9 3900 XT, Ryzen 7 3800 XT, Ryzen 5 3600 XT 'Matisse Refresh' Desktop CPUs Confirmed - Same Core Config, Higher Clocks & Price Cuts For Existing Models

AMD is preparing three new Ryzen CPUs as part of its Matisse Refresh family that include Ryzen 9 3900XT, Ryzen 7 3800XT & Ryzen 5 3600XT.

wccftech.com

 

[NostaSeronx]

So, what is going to be the 3950 refresh? Ryzen 9 3950XTX 5 GHz Edition(5GE)?!!?!?

 

[jpiniero]

So, what is going to be the 3950 refresh? Ryzen 9 3950XTX 5 GHz Edition(5GE)?!!?!?

My guess would be 125 W TDP parts with higher base.

 

[Zepp]

So, what is going to be the 3950 refresh?

another example of how they back themselves into corners with the marketing nomenclature

 

[uzzi38]

So, what is going to be the 3950 refresh? Ryzen 9 3950XTX 5 GHz Edition(5GE)?!!?!?

No need to refresh what is already entirely unreachable by the competition.

 

[Olikan]

Maybe Zen3 yields are so good, thanks to EUV, that selling lower binned models is a waste of silicon and money...

Similar to the Radeon approach (Vega + Polaris), Zen2 chips might become the low end ~4500

 

[uzzi38]

Maybe Zen3 yields are so good, thanks to EUV, that selling lower binned models is a waste of silicon and money...

Similar to the Radeon approach (Vega + Polaris), Zen2 chips might become the low end ~4500

More like AMD know they have an absolute monster on their hands and want a month or two of being able to cash in on leadership performance with Zen 3 with their higher tier chips before releasing the lower cost (and margin) R5 and R7 later down the line.

But I highly doubt you'll see MTS mixed in with Vermeer in the 4000 series lineup. RNR sure, but not MTS.

 

[darkswordsman17]

I don't know that this has much to do with Zen 3 (the only aspect it does there is that Zen 3 is likely far away, like 6 months, which I think current rumors are suggesting, with the highest end 12/16 core models maybe coming out earlier). With enough production that they've likely been able to build up bins that do well enough to sell as slightly premium SKUs that they could then release to nullify Intel's offerings. If they can offer another 10% performance that would basically do that. This way they can both drop prices but keep prices the same (meaning the new higher ones take over the price of previous ones, while they drop prices on the others), and counter Intel.

More like AMD know they have an absolute monster on their hands and want a month or two of being able to cash in on leadership performance with Zen 3 with their higher tier chips before releasing the lower cost (and margin) R5 and R7 later down the line.

But I highly doubt you'll see MTS mixed in with Vermeer in the 4000 series lineup. RNR sure, but not MTS.

I don't think he's saying they'll be 4000 series chips, they'll still be 3000, but that they might keep selling/producing Zen 2, while selling Zen 3 at a premium. I think they could do that, and they did that with Zen/Zen+ then Zen+/Zen2, so it has precedent. It will be interesting to see how things go. But Zen/Zen+ they had the wafer agreement with GF, and then Zen+/Zen2 they were on different processes at different companies, so producing one didn't impact production of the other. Keeping Zen 2 might impact production of Zen 3 (or the consoles). But they might still do it to constrain production that Nvidia (and possibly Intel if they're dong GPU there).

 

[RetroZombie]

All this might end up very confusing.
A new 3000 series refresh while the new 4000 line (renoir) releases and than some time latter zen3 also on the 4000 branding.
Too much in so little time on a company with very few resources.

 

[jpiniero]

All this might end up very confusing.
A new 3000 series refresh while the new 4000 line (renoir) releases and than some time latter zen3 also on the 4000 branding.
Too much in so little time on a company with very few resources.

We don't know when Renoir desktop is actually going to be available. Come to think of it, Matisse has to be cheaper for AMD to manufacture due to how expensive the 7 nm wafers are... and the virus has to of decimated desktop OEM demand. It's almost like why even bother. Selling Renoir with busted IGP would make sense but that might be a tough sell from a marketing perspective plus the clutter it causes.

 

[VirtualLarry]

Selling Renoir with busted IGP

Is there a SKU for a GPU-less Renoir yet?

 

[amrnuke]

We don't know when Renoir desktop is actually going to be available. Come to think of it, Matisse has to be cheaper for AMD to manufacture due to how expensive the 7 nm wafers are... and the virus has to of decimated desktop OEM demand. It's almost like why even bother. Selling Renoir with busted IGP would make sense but that might be a tough sell from a marketing perspective plus the clutter it causes.

I don't think 7nm wafer costs w/r/t 12nm wafer costs are different enough in raw numbers to make a large difference.

7nm wafer 300mm = $7500 per IC Tech and Gartner Inc
0.09 defect density
Renoir is ~11.5 x 14 (rough sketch on my part)
309 good dies per wafer
$24.27 per 8-core Renoir die

Even if the 12nm wafers were free, I think the $25 extra for the incredible performance benefit of moving from Zen/Zen+ -> Zen2 PLUS the process efficiency/performance improvements are well worth it.

 

[jpiniero]

I don't think 7nm wafer costs w/r/t 12nm wafer costs are different enough in raw numbers to make a large difference.

7nm wafer 300mm = $7500 per IC Tech and Gartner Inc
0.09 defect density
Renoir is ~11.5 x 14 (rough sketch on my part)
309 good dies per wafer
$24.27 per 8-core Renoir die

Even if the 12nm wafers were free, I think the $25 extra for the incredible performance benefit of moving from Zen/Zen+ -> Zen2 PLUS the process efficiency/performance improvements are well worth it.

I don't mean push the Ridges, I mean cancel Renoir Desktop in favor of more Matisse. To DIY, Renoir desktop is kind of redundant. You can always release it later if OEM demand picks back up.

 

[amrnuke]

I don't mean push the Ridges, I mean cancel Renoir Desktop in favor of more Matisse. To DIY, Renoir desktop is kind of redundant. You can always release it later if OEM demand picks back up.

Oh, wow. I completely and totally was thinking down a side road there.

Anyway, I don't know that canceling Renoir desktop is prudent, unless they have a wafer shortage. The 3200G and 3400G seemed to sell quite well. A substantial performance boost for OEMs who build those mini-SFF and SFF builds for business slim workstations might be reason enough to push Renoir APU.

From a business standpoint if I were AMD I'd be printing as many Renoir APUs as possible and shoving them down OEM's throats when they're only costing $25 a pop to make, and provide that much performance. I'd consider selling them at-cost even. Whatever you can to get your performance and security into their computers. Because in the end, once sysadmins become familiar and comfortable with it, that's how you influence bigger purchase decisions.

A 2400GE slim client absolutely hammers the 8100T slim client in my use cases. I can't imagine what a 4400GE would do.

Of course, just my opinion, definitely up to interpretation.

 

[jpiniero]

Oh, wow. I completely and totally was thinking down a side road there.

Anyway, I don't know that canceling Renoir desktop is prudent, unless they have a wafer shortage. The 3200G and 3400G seemed to sell quite well. A substantial performance boost for OEMs who build those mini-SFF and SFF builds for business slim workstations might be reason enough to push Renoir APU.

Because of the virus, OEM demand has to be way down since companies workers are at home and not at work. So even if OEMs cared, they have a large stockpile of Coffee Lake they need to unload. Be easier to get laptop sales now actually and AMD would get a higher price for it.

 

[Vattila]

With the 4000G APUs shaping up to be formidable chips, and with the ramp and widespread availability of Vermeer perhaps a long way out, could AMD be planning to align the APU and CPU model numbering going forward? Perhaps we'll have the Matisse refresh with 4000 model numbers alongside the 4000G APUs this summer, then we'll have to wait for the 5000 series a bit, perhaps launching at CES 2021, with both mobile and desktop APUs, alongside Vermeer desktop CPUs, all sporting the all new Zen 3 architecture. Doesn't sounds so bad, does it?

Matisse is doing great on the desktop, right? Or are you impatient for Vermeer? Would it disappoint you if you had to wait for Vermeer until January?

PS. Vermeer launch at CES 2021 would leave the holiday season all to the consoles, also.

 

[Vattila]

With all these leaks firming up, and it now looks like models 3600XT, 3800XT and 3900XT are to be revealed shortly — where is the 16-core refresh? Is there a possibility that AMD has kept the wraps on a refreshed 3950XT as well, or perhaps a new flagship 3990X, as I speculated here in the original post?

 

[jpiniero]

With all these leaks firming up, and it now looks like models 3600XT, 3800XT and 3900XT are to be revealed shortly — where is the 16-core refresh?

Vermeer...

 

[Hans Gruber]

I assume the Zen2+ offerings are the starting point for Zen3 clocks upon release. I also want someone to find out the rumored clocks before Zen2 was released. The leaks of Zen2+ gets very close to the clocks we were expecting with Zen2. A very impressive advancements by AMD. Perhaps a 5ghz Ryzen CPU may be a reality for Zen3.

 

[Atari2600]

another example of how they back themselves into corners with the marketing nomenclature

Marketing nomenclatures are often utterly stupid.

They could quite easily get around it though.

Current lineup:
R9 3950X
R9 3900X

R7 3800X
R7 3700X

New stepping lineup:
R9 3955X
R9 3905X

R7 3805X
R7 3705X


IMO, if I were running the lines, there would be no R3, R5, R7, R9. No need. Its just needlessly confusing.
Ryzen: ABCD
A - Generation
B - Tier within generation [i.e. number of cores would constitute different tier in some instances]
C - Speed within tier
D - Reserved nomenclature for respins etc

 

[Atari2600]

With all these leaks firming up, and it now looks like models 3600XT, 3800XT and 3900XT are to be revealed shortly — where is the 16-core refresh? Is there a possibility that AMD has kept the wraps on a refreshed 3950XT as well, or perhaps a new flagship 3990X, as I speculated here in the original post?

There already is a 3990X. Having an identical name would be madness

 

[Topweasel]

There already is a 3990X. Having an identical name would be madness

The 10900 would like to have a talk with you. Well I mean technically it has 5 different chips with that name but you have 1 that isn't even using the same Arch.

 

[RetroZombie]

Just a plus in front of the model would be enough:

AMD R9 3900X
AMD R9 3900X+
AMD R7 3800X
AMD R7 3800X+

They already used this in the past.