Ryzen 2 slide in KitGuruTech video

[Topweasel]

I don't see that happening on the 12nm node. On the 7? You bet.

Yeah considering the hard wall at 4GHz and only talking about a very slight density increase (and not an actual pathway size decrease), on a faster version of the process. It's going to be faster and maybe possibly people can consistently overclock to 5GHz+. But there isn't going to be 5GHz chips from AMD here. What I would expect is possibly mid to high 1-2 core turbo's and an all core turbo's capping at 4.3GHz maybe upwards of 4.5GHz on very cherry chips.

 

[Yotsugi]

Yeah considering the hard wall at 4GHz and only talking about a very slight density increase (and not an actual pathway size decrease), on a faster version of the process. It's going to be faster and maybe possibly people can consistently overclock to 5GHz+. But there isn't going to be 5GHz chips from AMD here. What I would expect is possibly mid to high 1-2 core turbo's and an all core turbo's capping at 4.3GHz maybe upwards of 4.5GHz on very cherry chips.

You're comparing density-focused server die to client-only something. Your suggested bump is too small for AMD to waste time and money that could be put into making Zen2 better.
Since 12LP delivers ~10% more performance with 7.5T HD libs, they might as well sacrifice desnity and go for 9T HP/UHP ones.
They will also optimize for clocks, because why not?
Basically they will try everything to delete Intel's clock advantage.

 

[Topweasel]

You're comparing density-focused server die to client-only something. Your suggested bump is too small for AMD to waste time and money that could be put into making Zen2 better.
Since 12LP delivers ~10% more performance with 7.5T HD libs, they might as well sacrifice desnity and go for 9T HP/UHP ones.
They will also optimize for clocks, because why not?
Basically they will try everything to delete Intel's clock advantage.

I think you are over focusing on this as AMD putting in a lot of work. They started on a LPP process for several reasons including EPYC. They always had plans on letting Ryzen get a bump for a new model this year by moving to LP. I don't think they ever had dreams about really taking it to Intel in 2018. 2018 by all accounts seems to be about treading water till 7nm, where they would be in position to really put the screws to Intel in lots of ways. The move to 12nm has more to with GF basically saying that you wait a couple more months and we will have something better and it won't take much work from you. That's the only reason for that move. Not because AMD is tossing money at the pure idea of using this process change to assume a leadership role.

2017 Return to competitiveness
2018 Maintain competitiveness
2019 Use die shrink to assume performance lead.

There is a reason it's called Zen+
There is a reason why AMD called Pinnacle Ridge, Summit Ridge architecture with "performance uplift"
There is a reason why AMD hasn't been all over the place talking about the switch to 12nm. Because it isn't real node shrink.

I think it's really important to understand that AMD is probably going to be in a good spot in 2018. They will close enough to Intel in clocks to not really matter. They can do some tweaks help eak out a little more IPC. In the worst case scenario Intel will have core parity with AMD. AMD will be able to maintain decent ASP even if they have to be a bit more of a value CPU than they spent 2017. But the full market availability next year of RR will make up for a drop in Ryzen value. AMD will get through 2018 better than 2017 probably. But that doesn't mean they will have the best chips ever.

Trying to make Pinnacle Ridge into something it's not. Is just going to feed into expectations. That are going to fall flat. Then it will be a disappointment. Then everyone will talk about how AMD screwed the pooch. AMD is better off making Zen 2 as amazing as we felt Zen was than trying to make up too much ground with Zen+. AMD can't do everything even if they wanted to and years ago they had to make a choice regarding what to do after Zen what to put the money into and Zen 2 is that answer not Zen+.

 

[CatMerc]

I think you are over focusing on this as AMD putting in a lot of work. They started on a LPP process for several reasons including EPYC. They always had plans on letting Ryzen get a bump for a new model this year by moving to LP. I don't think they ever had dreams about really taking it to Intel in 2018. 2018 by all accounts seems to be about treading water till 7nm, where they would be in position to really put the screws to Intel in lots of ways. The move to 12nm has more to with GF basically saying that you wait a couple more months and we will have something better and it won't take much work from you. That's the only reason for that move. Not because AMD is tossing money at the pure idea of using this process change to assume a leadership role.

2017 Return to competitiveness
2018 Maintain competitiveness
2019 Use die shrink to assume performance lead.

There is a reason it's called Zen+
There is a reason why AMD called Pinnacle Ridge, Summit Ridge architecture with "performance uplift"
There is a reason why AMD hasn't been all over the place talking about the switch to 12nm. Because it isn't real node shrink.

I think it's really important to understand that AMD is probably going to be in a good spot in 2018. They will close enough to Intel in clocks to not really matter. They can do some tweaks help eak out a little more IPC. In the worst case scenario Intel will have core parity with AMD. AMD will be able to maintain decent ASP even if they have to be a bit more of a value CPU than they spent 2017. But the full market availability next year of RR will make up for a drop in Ryzen value. AMD will get through 2018 better than 2017 probably. But that doesn't mean they will have the best chips ever.

Trying to make Pinnacle Ridge into something it's not. Is just going to feed into expectations. That are going to fall flat. Then it will be a disappointment. Then everyone will talk about how AMD screwed the pooch. AMD is better off making Zen 2 as amazing as we felt Zen was than trying to make up too much ground with Zen+. AMD can't do everything even if they wanted to and years ago they had to make a choice regarding what to do after Zen what to put the money into and Zen 2 is that answer not Zen+.

I don't really want to take a stance of agreement or disagreement with the above, but I'd like to point out that AMD has a different, very good reason, not to talk about Pinnacle Ridge.

Simply, Summit Ridge is still selling, and they don't want to Osborne the stock in flight by talking about Pinnacle Ridge. Before Ryzen released, AMD wasn't an actual option, so leaking information about Zen was only beneficial. Now it could actively hurt them.

 

[Yotsugi]

I don't really want to take a stance of agreement or disagreement with the above, but I'd like to point out that AMD has a different, very good reason, not to talk about Pinnacle Ridge.

Simply, Summit Ridge is still selling, and they don't want to Osborne the stock in flight by talking about Pinnacle Ridge. Before Ryzen released, AMD wasn't an actual option, so leaking information about Zen was only beneficial. Now it could actively hurt them.

Absolutely.
AMD would not repeat Intel's mistake of not clearing the inventory of last gen products.
CFL-S as is left mobo vendors particularly angry.

 

[IRobot23]

That image is fake from my perspective.
There will be no 12C for desktop, reaching over 4.5GHz on all cores is just insane to think.


Here are my thoughts and predictions:

AMD said pinnacle ridge is still based on (slides) "summit ridge architecture", but they can still improve IPC by few percents. With 12nm LLP they could achieve 4.5-4.6GHz is possible on single core.

AMD clock speed:
1. Fmax ~ 15% @ ~4GHz ZEN 14nm = 4.6GHz, I would say that all core turbo might reach around 4GHz.
2. IPC even with very small architecture optimizations they can get some IPC improvements.

So even if they do that there is still a problem with Ryzen DDR4 latency for gaming. As we know when DDR4 came out with those speeds @ 2133MHz/2400MHz CL15 was pretty much bad against DDR3 which could hit exact frequencies with much superior latencies.

3. If improved Infinity fabric for lower latencies. DDR4 latency is main problem for gamers. With new node and optimization we can also hope for higher ddr speeds. Maybe with that IMC and design we can get over 60GB/s on dual channel.


.... yet again, AMD is quiet as they were before Ryzen launch. Nobody expected ZEN to be this good and this cheap.

 

[Yotsugi]

reaching over 4.5GHz on all cores is just insane to think.

The madness will become reality.

 

[Topweasel]

I don't really want to take a stance of agreement or disagreement with the above, but I'd like to point out that AMD has a different, very good reason, not to talk about Pinnacle Ridge.

Simply, Summit Ridge is still selling, and they don't want to Osborne the stock in flight by talking about Pinnacle Ridge. Before Ryzen released, AMD wasn't an actual option, so leaking information about Zen was only beneficial. Now it could actively hurt them.

I was talking about the transition in process and not the CPU's themselves. Even post Ryzen release AMD was talking up a storm about where they were and how much work was going in to make 7nm right for Zen 2. When GF announced the 12nm Process AMD was there to say that "Hey this is awesome, and makes sense for some of our product lines". We assume that Ryzen 2 is going to be on 12nm because it will be a new die expected around the time 12nm leaves risk production, but AMD hasn't updated any roadmaps to include the move to 12nm. Which means they don't feel that this gives as big a difference as people are expecting otherwise I would expect that AMD in a conference call would talk about "Lower material costs, with major uplift in expected clock rates, which will increase our competitiveness and allow us to sell at higher ASPs". That you would expect. Still nothing. AMD has really not said much outside GF's announcement regarding 12nm at all.

 

[Yotsugi]

but AMD hasn't updated any roadmaps to include the move to 12nm.

Pinnacle Ridge existed on roadmaps for ages.

 

[Topweasel]

Pinnacle Ridge existed on roadmaps for ages.

Again not Pinnacle Ridge. I have noted it has been on their ages and in fact made these two statements

There is a reason it's called Zen+
There is a reason why AMD called Pinnacle Ridge, Summit Ridge architecture with "performance uplift"

AMD can't do everything even if they wanted to and years ago they had to make a choice regarding what to do after Zen what to put the money into and Zen 2 is that answer not Zen+

The roadmap update was for 12nm. They were very clear on Pinnacle Ridge and Zen 2 being 14nm+ and 7nm respectively (and Zen 3 being 7nm+). They haven't been clear on what products are going to be on 12nm and that is a process that for AMD outside custom or very specific product lines only has a short 6-9 months of Viability for AMD.

 

[Yotsugi]

being 14nm+

"12nm" is 14+.

 

[IRobot23]

The madness will become reality.

Well 12nm LLP 8C/16T, 95W TDP? Madness. 7nm? Possibility.

 

[Topweasel]

"1210nm" is 14+.

Yes and no. It kind of is, but it isn't and has a different time table for production than was originally mapped for 14nm+. But that is the point, it isn't a major shift from 14nm. Therefore let's not expect too much from AMD basically sliding over from a low power process to a high performance process and realistically that is the only thing of value that is changing between Summit and Pinnacle Ridge.

 

[Yotsugi]

Yes and no. It kind of is, but it isn't and has a different time table for production than was originally mapped for 14nm+. But that is the point, it isn't a major shift from 14nm. Therefore let's not expect too much from AMD basically sliding over from a low power process to a high performance process and realistically that is the only thing of value that is changing between Summit and Pinnacle Ridge.

You do understand that Summit Ridge was fabbed with HD libs?
You do understand what will happen while switching to HP/UHP libs?
Because 14LPP itself has high-performance options.
12LP moreso.

 

[Topweasel]

You do understand that Summit Ridge was fabbed with HD libs?
You do understand what will happen while switching to HP/UHP libs?
Because 14LPP itself has high-performance options.
12LP moreso.

This is getting pointless. I don't think we are going to see eye to eye. If you think like Library/process swap is enough to get AMD to be able to clock a CPU 50% faster then more power to you.

But even if I agreed (I don't). That is basically the end of the changes between Summit and Pinnacle Ridge.

 

[Yotsugi]

clock a CPU 50% faster

Where did I say "50% faster"?

That is basically the end of the changes between Summit and Pinnacle Ridge.

Clocks is literally all that AMD needs to kill CFL-S threat.

 

[Topweasel]

Where did I say "50% faster"?

The General all core clock limit AMD has for it's CPU's are 3.7-8GHz (stock). 5GHz implies a ~1.3GHz upswing in clocks. Which really is about 35-40%% but still. Even if we went with single core turbos. You are talking about still nearly a 1 GHz speed increase for a single core clock a 25% upswing. One I do mention that I think might be able hit with cherry picked options (think 2800x).

Here is what I expect 20% increase in base clocks. 10% increase in all core turbo clocks. 15% in single core turbo clocks.

That is my best case scenario. But realistically we really don't know anything about GF's processes because AMD products are the only thing they have that sits on that kind of performance level. Yeah the Library/process switch should help a lot but by how much? Specially how much on a die as large as Zeppelin? Keeping in mind that AMD isn't going to want to push power usage much higher if at all. They don't need to because they are competitive enough not to need to deal with the mess they did with Bullozer/Piledriver by pushing power usage after initial launch. Specially considering that unlike Piledriver they actually already have OEM adoption of Ryzen and won't want to negatively affect the OEM's.

Clocks is literally all that AMD needs to kill CFL-S threat.

Well You haven't seen the other dozens of suggestions where, it's going more cores already (like the slide) (which side comment we are just 12+ months away from AMD increasing anyways), fixing IF (as if it needed to be fixed/changed), fixing memory latency, and so on.

 

[Yotsugi]

They don't need to because they are competitive enough not to need to deal with the mess they did with Bullozer/Piledriver by pushing power usage after initial launch. Specially considering that unlike Piledriver they actually already have OEM adoption of Ryzen and won't want to negatively affect the OEM's.

Their lineup has enough space for both low power and high performance (read housefire) SKUs.

 

[IRobot23]

@Topweasel
OC : 3.9GHz on AVG for desktop ryzen? 15% boost clock should get ryzen to around 4.4GHz on AVG.

Yet again, overclocking DF to 3GHz of higher would be much better deal than 4.4GHz on Core.

 

[Topweasel]

Their lineup has enough space for both low power and high performance (read housefire) SKUs.

There is a large desire for the OEM's not to have changing platforms. It was a big deal with Intel and one of the things AMD wants to stay away from. The last thing AMD wants to do is go to dell and tell them "Hey Michael, our top end sku's for Ryzen 7 are now 120w CPU's, so now redesign your boards and cooling from the ground up. Don't worry it's worth it." Specially if it means only an extra 100-200MHz.

AMD hoped getting BD clocked high enough and priced low enough would lead to design wins because they had none. That isn't the case now and I am waiting for my Ryzen 7 Optiplex to come in (probably first week of January) to play with.

 

[Topweasel]

@Topweasel
OC : 3.9GHz on AVG for desktop ryzen? 15% boost clock should get ryzen to around 4.4GHz on AVG.

Yet again, overclocking DF to 3GHz of higher would be much better deal than 4.4GHz on Core.

I'd say 3.8. While 1700x and 1800x could go higher (and generally the 1800x did). Those didn't sell nearly as well as the 1700 and even with the 1700x I want to say those took a big hit when measuring how many hit 3.9 vs. 3.8. Also look at power usage. We have to look at the clock limitations from 3 angles. Process Limitation, thermal/power limitations, architecture limitations. We know that Process wise it caps out 4.1ish no matter what the other two factor contribute (outside LN2). Do we know that this arch is comfortable being clocked up to 5GHz? Do we know what the power requirements would be at that level and therefore what kind of cooling it needs? That's why I try resisting from looking at it from an OC potential and look at from a stock settings approach. Because even if there were no limits process and arch wise, power is going to play a big part on Ryzen 2's clocks. In the power envelope they provide I question the viability of a 4.4GHz all core clock.

 

[Yotsugi]

Yet again, overclocking DF to 3GHz of higher would be much better deal than 4.4GHz on Core.

Why the hell do you all hate NUCA and want to clock poor data fabric to nuclear degrees? Its very much diminishing returns.

Do we know that this arch is comfortable being clocked up to 5GHz?

14-19 stages; very much comfortable.

 

[IRobot23]

I'd say 3.8. While 1700x and 1800x could go higher (and generally the 1800x did). Those didn't sell nearly as well as the 1700 and even with the 1700x I want to say those took a big hit when measuring how many hit 3.9 vs. 3.8. Also look at power usage. We have to look at the clock limitations from 3 angles. Process Limitation, thermal/power limitations, architecture limitations. We know that Process wise it caps out 4.1ish no matter what the other two factor contribute (outside LN2). Do we know that this arch is comfortable being clocked up to 5GHz? Do we know what the power requirements would be at that level and therefore what kind of cooling it needs? That's why I try resisting from looking at it from an OC potential and look at from a stock settings approach. Because even if there were no limits process and arch wise, power is going to play a big part on Ryzen 2's clocks. In the power envelope they provide I question the viability of a 4.4GHz all core clock.

Where did you get those numbers?
Based on silicon lottery I would say 3.9-4Ghz

 

[Topweasel]

Where did you get those numbers?

Admittingly early Silicon lottery numbers. I haven't checked as of late.

 

[Topweasel]

Where did you get those numbers?
Based on silicon lottery I would say 3.9-4Ghz

Okay makes sense farther into production enough tweaks to the process and understanding of what to look for in the chips for binning. I thought maybe TR would drag down the average on Ryzen since it gets the really good chips. But lets call it 3.9. The rest still stands, so maybe we have chips we can comfortably OC to mid 4.5, hell I'll be really nice and say AMD and GF pulls off a miracle and we can get near 5GHz on OC's. That doesn't mean much if max 8 cores at 90w for retail is something like 4.2GHz. It will help but won't quell the "they screwed up and didn't catch up to Intel and Intel is going to have a 8core CFL chip and then there won't be any reason to get AMD crowd".