Speculation: The CCX in Zen 2

[maddie]

Do we assume that Zen was designed primarily for the server market in AMD's priorities?
If we assume that the server space is where the big profits exist and that running medium stressed VMs is a large part of that market, then higher core count Zen2 is very much a rational product.
If we also assume that Zen2 is a drop-in replacement for Zen, then this becomes an even bigger selling point for server farms.

Ignore the above at your own risk, in my opinion.

 

[eek2121]

You see that is the problem. Those "early" "leaked" "unconfirmed" slides are contradicting to the newest "leaked" "unconfirmed" slides. And we all take it as a fact.

I am not aware of AMD publicly stating there will be a 48 Core Starship Server CPU. And therefore the only reason why we are polling for 6 Core CCX is simply because of this illusionary 48 Core CPU.

Purely speculating, those earlier slides make much more sense. 7nm in early 2018 ( From those 2nd leaked slides ) is simply too early. It make much more sense to have 14nm++ along side with improved IPC Zen 2.

And 4 Core make so much more sense for consumers. That is up to 8 Thread. For the past decades we haven't seen another killer apps that requires a leap of computer resources. A Core 2 Duo 10 years ago, with enough RAM and SSD will still be very speedy. I believe the 4 Core Entry, 8 / 16 Core for Middle and Semi Pro. And those who want infinite resources I think there will be 8 Core CCX with up to 32 Core per CPU.

It looks like you quoted the wrong person.

 

[eek2121]

Do we assume that Zen was designed primarily for the server market in AMD's priorities?
If we assume that the server space is where the big profits exist and that running medium stressed VMs is a large part of that market, then higher core count Zen2 is very much a rational product.
If we also assume that Zen2 is a drop-in replacement for Zen, then this becomes an even bigger selling point for server farms.

Ignore the above at your own risk, in my opinion.

Ryzen was built as a one size fits all solution. AMD's goals are to increase margins to beyond 40% by 2020 and they won't get there by raising the cost to produce the chips. 7nm is already going to be more expensive. Unless Intel changes course, you won't see a larger CCX size out of AMD. The model that AMD went with has worked pretty well for them, so I expect them to stay exactly where they are. We might see higher core count CPUs at some point, but I expect the CCX will remain the same size.

 

[moinmoin]

You see that is the problem. Those "early" "leaked" "unconfirmed" slides are contradicting to the newest "leaked" "unconfirmed" slides. And we all take it as a fact.

I am not aware of AMD publicly stating there will be a 48 Core Starship Server CPU. And therefore the only reason why we are polling for 6 Core CCX is simply because of this illusionary 48 Core CPU.

Purely speculating, those earlier slides make much more sense. 7nm in early 2018 ( From those 2nd leaked slides ) is simply too early. It make much more sense to have 14nm++ along side with improved IPC Zen 2.

And 4 Core make so much more sense for consumers. That is up to 8 Thread. For the past decades we haven't seen another killer apps that requires a leap of computer resources. A Core 2 Duo 10 years ago, with enough RAM and SSD will still be very speedy. I believe the 4 Core Entry, 8 / 16 Core for Middle and Semi Pro. And those who want infinite resources I think there will be 8 Core CCX with up to 32 Core per CPU.

Which slides are you referring to? Zen 2 is on 7nm and is around late 2018/early 2019, all officially:

Do we assume that Zen was designed primarily for the server market in AMD's priorities?
If we assume that the server space is where the big profits exist and that running medium stressed VMs is a large part of that market, then higher core count Zen2 is very much a rational product.
If we also assume that Zen2 is a drop-in replacement for Zen, then this becomes an even bigger selling point for server farms.

Ignore the above at your own risk, in my opinion.

The fact that the same Zeppelin die is capable of the two sockets, 8 dies Epyc systems makes it clear that the die was primarily designed for the server market. There is a lot of unused uncore on the Ryzen chips.

 

[maddie]

Ryzen was built as a one size fits all solution. AMD's goals are to increase margins to beyond 40% by 2020 and they won't get there by raising the cost to produce the chips. 7nm is already going to be more expensive. Unless Intel changes course, you won't see a larger CCX size out of AMD. The model that AMD went with has worked pretty well for them, so I expect them to stay exactly where they are. We might see higher core count CPUs at some point, but I expect the CCX will remain the same size.

Which slides are you referring to? Zen 2 is on 7nm and is around late 2018/early 2019, all officially:

The fact that the same Zeppelin die is capable of the two sockets, 8 dies Epyc systems makes it clear that the die was primarily designed for the server market. There is a lot of unused uncore on the Ryzen chips.

It might be a one size fits all, but the following is still true.

I agree. This bolded part pretty much says Server first and still be used effectively in Client.

ksec might be talking about this slide. It seems legit and the fact that it was not made "public" is hardly a reason to reject it outright. Arguing for faster and same # of cores is fine, but anyone refusing to incorporate this slide indicating a 48/96 core thread CPU is ignoring contrary information.

 

[moinmoin]

ksec might be talking about this slide. It seems legit and the fact that it was not made "public" is hardly a reason to reject it outright. Arguing for faster and same # of cores is fine, but anyone refusing to incorporate this slide indicating a 48/96 core thread CPU is ignoring contrary information.

I know where Starship originates from, it's the main reason we have this whole discussion thread after all. I wanted to know which slides he was looking at to come to the following conclusion: "It make much more sense to have 14nm++ along side with improved IPC Zen 2."

 

[Mechanical Man]

If AMD is going to change process mid ryzen like that slide indicates. It dont think it is reasonable to expect it to be more than zen v1.3.

 

[moinmoin]

Again where are you getting that from? By all official indications there's only Zen (14nm), Zen+ (14nm+), Zen 2 (7nm) and Zen 3 (7nm+). And Zen+ doesn't even appear to apply to the whole product lineup but only Raven Ridge (which is a die different from Zeppelin anyway).

 

[Topweasel]

Again where are you getting that from? By all official indications there's only Zen (14nm), Zen+ (14nm+), Zen 2 (7nm) and Zen 3 (7nm+). And Zen+ doesn't even appear to apply to the whole product lineup but only Raven Ridge (which is a die different from Zeppelin anyway).

Don't forget Pinnacle Ridge. EPYC and TR probably won't see an update till Zen 2.

 

[french toast]

Pinnacle ridge is 14+ apparently, rumours of raven ridge also but would hang my hat on that last one.

 

[eek2121]

Don't forget Pinnacle Ridge. EPYC and TR probably won't see an update till Zen 2.

Pinnacle Ridge will likely be a refresh of the entire product stack. I expect that the R7, R5, and R3 2xxx series will all be Pinnacle Ridge with the rollout starting around April of next year based on statements made by AMD.

 

[Topweasel]

Pinnacle Ridge will likely be a refresh of the entire product stack. I expect that the R7, R5, and R3 2xxx series will all be Pinnacle Ridge with the rollout starting around April of next year based on statements made by AMD.

The Ryzen lineup would be refreshed. Epyc's lineup doesn't really have room for a 14nm+ refresh. If EPYC isn't seeing Zen+, I doubt TR is.

 

[moinmoin]

Don't forget Pinnacle Ridge. EPYC and TR probably won't see an update till Zen 2.

Good call. Though Pinnacle Ridge hasn't been officially mentioned since March (while Raven Ridge matches with Ryzen Mobile that was mentioned since), and back then it seemed Pinnacle Ridge would arrive before Raven Ridge in early 2018 and be Zen 2 based which is all clearly no longer the case. AMD may do a Ryzen-specific Zen+ optimization that cuts out all the unused uncore (that's not needed when not targeting TR and Epyc), but as of now this feels superfluous to me unless they have to delay Zen 2. Maybe it's an MCM with two Raven Ridge (so 1x CCX) dies instead?

 

[formulav8]

Pinnacle ridge

I'm thinking Zen+ on Samsung's 14nm LPU based process. Possibly RR as well.

 

[formulav8]

Raven Ridge in early 2018

RR is 4th Quarter this year for mobile and next year for desktop last I saw.

 

[raghu78]

Don't forget Pinnacle Ridge. EPYC and TR probably won't see an update till Zen 2.

EPYC might not benefit from a higher performance process designed for 4+ GHz speeds but Threadripper surely will. I think Raven Ridge will be the first chip to launch built at 14nm+ . RR launches first in notebooks in late 2017 and in desktops in Q1 2018. Pinnacle Ridge is likely to be the 2nd chip built at 14 nm+ to launch in late Q1 2018 with TR using 14nm+ launching in Q3 2018.

 

[Topweasel]

EPYC might not benefit from a higher performance process designed for 4+ GHz speeds but Threadripper surely will. I think Raven Ridge will be the first chip to launch built at 14nm+ . RR launches first in notebooks in late 2017 and in desktops in Q1 2018. Pinnacle Ridge is likely to launch in late Q1 2018 with TR using 14nm+ launching in Q3 2018.

Higher clocks mean higher power. I don't expect the efficiency to increase. Just it's potential. 180w is still a lot of power usage. I am sure that AMD could release a TR version. I just don't see the need, Intel is unlikely to do another x platform till 2019, and pushing power usage higher might not be that interesting to them. Whereas a 110-120w 5GHz 8/6/4 core Zen would have a lot of value.

 

[french toast]

I'm thinking Zen+ on Samsung's 14nm LPU based process. Possibly RR as well.

Thats the dream I wouldn't count out some AMD own LPP special sauce though, which would be a disaster if we look at Rx 480 vs rx580.

 

[moinmoin]

RR is 4th Quarter this year for mobile and next year for desktop last I saw.

I know, that whole line (derived from slides way before Ryzen's launch) is outdated bunks nowadays which was my whole point.

 

[raghu78]

Higher clocks mean higher power. I don't expect the efficiency to increase. Just it's potential. 180w is still a lot of power usage. I am sure that AMD could release a TR version. I just don't see the need, Intel is unlikely to do another x platform till 2019, and pushing power usage higher might not be that interesting to them. Whereas a 110-120w 5GHz 8/6/4 core Zen would have a lot of value.

Intel has confirmed a 14++ based Cascade Lake line of server and HEDT CPUs.

https://newsroom.intel.com/news/int...pcoming-intel-xeon-processor-scalable-family/
https://www.fool.com/investing/2017/06/05/why-intel-corporations-cascade-lake-is-so-importan.aspx

So AMD would need some improvements to remain competitive. AMD Ryzen and TR is limited by 14LPP 's frequency hard wall at 4 Ghz. If AMD can get to 4.5 Ghz max single core turbo and a max freq limit of 4.5-4.6 Ghz limit they would do well against Cascade Lake.

 

[Topweasel]

Intel has confirmed a 14++ based Cascade Lake line of server and HEDT CPUs.

https://newsroom.intel.com/news/int...pcoming-intel-xeon-processor-scalable-family/
https://www.fool.com/investing/2017/06/05/why-intel-corporations-cascade-lake-is-so-importan.aspx

So AMD would need some improvements to remain competitive. AMD Ryzen and TR is limited by 14LPP 's frequency hard wall at 4 Ghz. If AMD can get to 4.5 Ghz max single core turbo and a max freq limit of 4.5-4.6 Ghz limit they would do well against Cascade Lake.

Yeah I am not so sure I would be worried about the third/fourth attempt at a 14nm CPU from Intel using an arch 10 years old. AMD might fall slightly behind mid 2018, but honestly TR falls behind Cascade Lake no matter what. Why relaunch a new TR middle of next year to be behind anyways. EPYC is mainly fighting in areas that Cascade Lake as an arch won't matter much. Security, cores, and IO make a much bigger difference. Intel needed to plan ahead for that.

The big point is AMD can't roll out as quick as we like and can't spread themselves too thin. They have a lot of releases left to go and a tweener HEDT CPU with a 6 month lifespan doesn't serve a purpose. Specially if I am right and there isn't much power savings and it's all just about allowing a higher clock.

 

[raghu78]

Yeah I am not so sure I would be worried about the third/fourth attempt at a 14nm CPU from Intel using an arch 10 years old. AMD might fall slightly behind mid 2018, but honestly TR falls behind Cascade Lake no matter what. Why relaunch a new TR middle of next year to be behind anyways. EPYC is mainly fighting in areas that Cascade Lake as an arch won't matter much. Security, cores, and IO make a much bigger difference. Intel needed to plan ahead for that.

The big point is AMD can't roll out as quick as we like and can't spread themselves too thin. They have a lot of releases left to go and a tweener HEDT CPU with a 6 month lifespan doesn't serve a purpose. Specially if I am right and there isn't much power savings and it's all just about allowing a higher clock.

AMD cannot afford to let up on the momentum gained with Ryzen 7/5 and Ryzen TR launch. Zen on 14+ is likely to drive the 2018 product stack - Raven Ridge and Pinnacle Ridge. Since TR uses the same dies as Ryzen 7 there is no extra effort required. All thats needed is binning to select the best dies for TR. Moreover 7nm CPUs are not going to be available before H1 2019. The way I see it if GF executes very well (which is unlikely given their current track record) 7LP will go into production by late 2018. So allowing for a quarter or so delay would mean 7nm Zen 2 based Ryzen is likely to launch in Q2 2019 and Rome server CPUs in Q3 2019.

Raven Ridge notebook - late Q4 2017
Raven Ridge desktop - late Q1 2018 or early Q2 2018
Pinnacle Ridge late Q1 2018 or early Q2 2018
TR (Pinnacle Ridge based) - mid-late Q3 2018
7nm Zen 2 - Q2 2019
7nm TR and Rome- Q3 2019

 

[LightningZ71]

Just don't loose sight of the fact that the current Zeppelin cores are produced on what's supposed to be GF's low power process. Switching to the faster "performance" process will not only create higher current draws from higher available clock rates, but the process itself is more power hungry to begin with from what I'm seeing online. Processors pushing into the 4.5Ghz range will be consuming a LOT more power than current processors on that process. Then, map that onto TR and EPYC, and you've got power draw numbers that just aren't survivable in anything with a normal power setup and may draw too much for even the socket to handle (though, to be fair, that socket is massive and should be able to handle a lot of power).

 

[Topweasel]

Just don't loose sight of the fact that the current Zeppelin cores are produced on what's supposed to be GF's low power process. Switching to the faster "performance" process will not only create higher current draws from higher available clock rates, but the process itself is more power hungry to begin with from what I'm seeing online. Processors pushing into the 4.5Ghz range will be consuming a LOT more power than current processors on that process. Then, map that onto TR and EPYC, and you've got power draw numbers that just aren't survivable in anything with a normal power setup and may draw too much for even the socket to handle (though, to be fair, that socket is massive and should be able to handle a lot of power).

That's my point Zen+ only matters if they either lower power usage at speed or improve IPC. But as far as we know it's just a move to a higher clock process. The only place this matters is with the desktop Ryzen market. TR is already straining what AMD wants to use on a socket and EYPC is much farther back right into the efficiency range of Zeppelin. Higher clocks do it no good.

Hell even Ryzen mobile (APU) would do better on LPP. Zen + looks to be a 7700/8700/7820> and i5 competitor and not much else.

 

[LightningZ71]

I have to agree. AMD needs something in that space to better compete with the higher clocked I5 and higher clocked/cored i7/i9 parts. Switching process and some minor speedpath tweaks should get them into the mid 4Ghz range, but, those power numbers are going to be brutal, and even under-clocking won't make a big gain. I suspect that you'll see processors in the 6-8 core range with a 100-200 mhz jump in base frequency, but 400+ mhz jumps in the turbo frequencies of both all core and top core situations. Base power should go up a non-trivial amount and boost clock power draw numbers should be just painful.