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?

 

[VirtualLarry]

I don't know. I mean, if there were legitimately a way for AMD to get TSMC to fab Zen2-based CPUs, that would ACTUALLY clock at 5Ghz, part of me would be leaping for that chance, just to one-up Intel in yet another product dimension (clock freq.). It might potentially really shut-up Intel fans / AMD detractors for good, about the so called "frequency wars" (now that AMD has brought the "core wars" smackdown to Intel). Plus, the performance could be really good, thought they might want to implement an "AVX2 offset" like Intel does, in that case, just to keep power limits in check for "heavy" FPU workloads. (Like BOINC / PrimeGrid, as I am a fan of running.) A 125W TDP on those chips wouldn't be a bad thing, I don't think. It wasn't so very long ago, that AM2/AM2+ CPUs, like the Athlon X2 5600+ and 6000+, and the Phenom II X4/X6 CPUs were rated at 125W TDP. None of them died, to my knowledge.

 

[Vattila]

they might want to implement an "AVX2 offset" like Intel does, in that case,

All they need to do, as far as I understand, is to set a sufficiently low base clock. The Precision Boost algorithm will handle the rest, adjusting frequency to keep within TDP and temperature limits.

 

[maddie]

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 offer 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?

View attachment 11358

Zen3, Ryzen 4xxx be damned?

 

[NostaSeronx]

imho, it would be better to jump to N6 EUV re-using N7 w/ N7P. Mostly a re-tapeout, overall higher yields. From 550 to 710 good dies, etc(from numbers I have seen in EUV over DUV). HPC/High Frequency does get a performance boost in Fmax/Vmax capability with EUV. From lower BEOL & FEOL/MOL parasitics.

Zen3, Ryzen 4xxx be damned?

Zen3 might not reach the height of Zen2, with it going the power efficiency route.

Zen1 -> Zen3
Zen2 -> Zen4

 

[maddie]

imho, it would be better to jump to N6 EUV re-using N7 w/ N7P. Mostly a re-tapeout, overall higher yields. From 550 to 710 good dies, etc(from numbers I have seen in EUV over DUV). HPC/High Frequency does get a performance boost in Fmax/Vmax capability with EUV. From lower BEOL & FEOL/MOL parasitics.Zen3 might not reach the height of Zen2, with it going the power efficiency route.

Zen1 -> Zen3
Zen2 -> Zen4

Are you saying/implying that there will be no replacement for the Ryzen 3xxx series in 2020?

 

[Vattila]

Zen3, Ryzen 4xxx be damned?

I expected a yearly cadence from Ryzen 2000 to 3000, i.e. a launch in April. However, AMD needed another quarter, and Ryzen 3000 launched in July. In light of the delay, I have adjusted my expectations to a cadence of 5 quarters, which would mean Ryzen 4000 launching early October next year. If so, a Ryzen 3000 refresh early April would give the series a 6 months sales boost.

Intel's desktop refresh, Comet Lake-S, is expected in 2020-Q1 with 10 cores and further refinements, while retaining leading performance in lightly threaded workloads, due to high frequency and IPC. AMD's desktop range could use a refresh in the same time frame, to maintain momentum and drive further gains in market share.

 

[soresu]

Zen3, Ryzen 4xxx be damned?

This.

Zen3 is already design complete, straight from the horses mouth, I'd expect engineering samples in their hands between christmas 2019 and early Q2 2020 at the latest.

 

[soresu]

Zen3 might not reach the height of Zen2, with it going the power efficiency route.

So you are thinking a Cortex A72 -> A73 type situation, slight IPC loss with significant perf/watt gain?

 

[maddie]

I expected a yearly cadence from Ryzen 2000 to 3000, i.e. a launch in April. However, AMD needed another quarter, and Ryzen 3000 launched in July. In light of the delay, I have adjusted my expectations to a cadence of 5 quarters, which would mean Ryzen 4000 launching early October next year. If so, a Ryzen 3000 refresh early April would give the series a 6 months sales boost.

Intel's desktop refresh, Comet Lake-S, is expected in 2020-Q1 with 10 cores and further refinements, while retaining leading performance in lightly threaded workloads, due to high frequency and IPC. AMD's desktop range could use a refresh in the same time frame, to maintain momentum and drive further gains in market share.

Seems like too much trouble for too few benefits. It's not only AMD here, but the entire distribution chain that will have to do extra work. I really don't see all that happening for a few months sales. If Comet Lake is an issue, just adjust prices as we've seen them do repeatedly.

 

[Yotsugi]

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

Why would they waste their time on that given their yearly core cadence?

So you are thinking a Cortex A72 -> A73 type situation, slight IPC loss with significant perf/watt gain?

He's deadass wrong.

 

[soresu]

He's deadass wrong.

I do find it singularly unlikely that they would allow a generational IPC drop after Bulldozer, though the A72 -> A73 situation was nowhere near that bad, in return for significant mhz per watt sustainability improvements.

 

[Yotsugi]

I do find it singularly unlikely that they would allow a generational IPC drop after Bulldozer, though the A72 -> A73 situation was nowhere near that bad, in return for significant mhz per watt sustainability improvements.

Zen3 isn't a mobile-first core where perf/W in vacuum is the biggest consideration.

 

[soresu]

Zen3 isn't a mobile-first core where perf/W in vacuum is the biggest consideration.

Though mobile first is indeed unlikely, I think it's too early to say what exactly it is given the wild speculation floating about presently.

I'm sure the Next-er Horizon event will be revealing soon enough, though that is unlikely to hit so soon after the launch of Zen2 - probably between Q1-Q2 next year.

 

[Yotsugi]

I think it's too early to say what exactly it is given the wild speculation floating about presently.

They've already taped out some Zen3 stuff so haha.

 

[soresu]

They've already taped out some Zen3 stuff so haha.

Taped out very likely given the 'design complete' status affirmed on their roadmaps, though they've given basically nothing away beyond 7nm+, higher efficiency, and that Milan will beat Ice Lake in perf/watt.

 

[Vattila]

Why would they waste their time on that given their yearly core cadence?

#7

 

[darkswordsman17]

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 offer 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?

View attachment 11358

I don't see it happening. AMD likely needs N7 cleared for console production and I'd guess GPU as well, and so they'll want to transition to 7+ since their CPUs are by far more profitable (especially for EPYC). Plus they'll need N7 for APU (which I'm guessing they'll be doing a spring monolithic APU and will ditch the chiplet approach for some future time - I won't be surprised if it doesn't start til they have AM5 socket, where they'll have options of either HBM3 or boosting memory channels to 3 or 4 along with DDR5 to boost memory bandwidth as otherwise a sizable GPU is going to be largely wasted).

I'm gonna say no, and that its full steam ahead for AMD.

I don't know. I mean, if there were legitimately a way for AMD to get TSMC to fab Zen2-based CPUs, that would ACTUALLY clock at 5Ghz, part of me would be leaping for that chance, just to one-up Intel in yet another product dimension (clock freq.). It might potentially really shut-up Intel fans / AMD detractors for good, about the so called "frequency wars" (now that AMD has brought the "core wars" smackdown to Intel). Plus, the performance could be really good, thought they might want to implement an "AVX2 offset" like Intel does, in that case, just to keep power limits in check for "heavy" FPU workloads. (Like BOINC / PrimeGrid, as I am a fan of running.) A 125W TDP on those chips wouldn't be a bad thing, I don't think. It wasn't so very long ago, that AM2/AM2+ CPUs, like the Athlon X2 5600+ and 6000+, and the Phenom II X4/X6 CPUs were rated at 125W TDP. None of them died, to my knowledge.

Its not worth it, and they'll get bigger uptick in performance from Zen 3 so there's no point. No one cares about GHz any more. Even Intel has faltered there and most people just shrugged at any Intel fans that tried to push the 5GHz arguments.

I don't think the clocking issues with AMD has anything to do with TDP or need for clock offsets, as according to AMD neither is the issue. Its just the overall design and transistors have limits. AMD has done a lot to try and maximize the ability to reach those limits (without needing to overclock or resort to as farcical of TDP nonsense as Intel).

AMD benefits more from just moving to Zen 3. They really can't afford to start pulling an Intel and changing plans and adding obscurity. AMD needs to execute. They also likely have other uses for the N7 production that they need as well. Their CPUs have the margins that enable them to push to newer nodes.

I expected a yearly cadence from Ryzen 2000 to 3000, i.e. a launch in April. However, AMD needed another quarter, and Ryzen 3000 launched in July. In light of the delay, I have adjusted my expectations to a cadence of 5 quarters, which would mean Ryzen 4000 launching early October next year. If so, a Ryzen 3000 refresh early April would give the series a 6 months sales boost.

Intel's desktop refresh, Comet Lake-S, is expected in 2020-Q1 with 10 cores and further refinements, while retaining leading performance in lightly threaded workloads, due to high frequency and IPC. AMD's desktop range could use a refresh in the same time frame, to maintain momentum and drive further gains in market share.

Wasn't the issue the chipsets though and not the CPUs? The transition to the 600 series chipsets shouldn't be as problematic.

 

[Ajay]

This.

Zen3 is already design complete, straight from the horses mouth, I'd expect engineering samples in their hands between christmas 2019 and early Q2 2020 at the latest.

Yup. Besides, Zen+ was a year two refresh (14->12nm), so it’s not needed this year as 7nm EUV is already ramping up (is it in risk production yet?).

 

[Vattila]

Seems like too much trouble for too few benefits.

Yeah, the cost-benefit analysis may not be favourable. A refresh would be a good insurance, though, should Ryzen 4000 stumble and delay.

 

[Yotsugi]

Taped out very likely given the 'design complete' status affirmed on their roadmaps, though they've given basically nothing away beyond 7nm+, higher efficiency, and that Milan will beat Ice Lake in perf/watt.

Well of course they're supposed to be as vague as they can be.
Only the paranoid survive.

Yeah, the cost-benefit analysis may not be favourable. A refresh would be a good insurance, though, should Ryzen 4000 stumble and delay.

Lisa sez stumbles and delays are not a thing.
Believe in Lisa.

 

[soresu]

should Ryzen 4000 stumble and delay

Zen4/Ryzen 5000 is more likely to stumble given its reported focus on 3d packaging/stacking, which introduces execution problems beyond semicon fab process tech alone.

 

[Yotsugi]

Zen4/Ryzen 5000 is more likely to stumble given its reported focus on 3d packaging/stacking, which introduces execution problems beyond semicon fab process tech alone.

We dun know if Genoa/other Zen4 stuff will be the lead SoIC product.
I mean TSMC pkg roadmap is strictly HPC-driven now, but 3D is nontrivial and doing it for something hueg from the get-go can result in less than fun stuff.

 

[soresu]

We dun know if Genoa/other Zen4 stuff will be the lead SoIC product.
I mean TSMC pkg roadmap is strictly HPC-driven now, but 3D is nontrivial and doing it for something hueg from the get-go can result in less than fun stuff.

Agreed, very possibly we may see a '3D logic stacking/PIM pipecleaner' product to use AMD HD 4770 perlance.

Possibly some sort of mobile GPU or APU stacking HBM on the logic.

Either way I would not expect stacks to exceed 2 high with a major logic die in the mix, even at 5nm.

 

[soresu]

Lisa sez stumbles and delays are not a thing.
Believe in Lisa.

We need a meme with a banknote with Lisa's face on it, with 'In Lisa we trust' in the usual place