AMD Ryzen Gen 2 Set For Q2 2018

[french toast]

People here keep saying that Zen+ is going to be ~10% faster.

Was this ever officially said anywhere?

Or is this: Somebody uncle's 5th cousin-in-law said so?

Yes, if you look a couple of pages back there is a link (abwx?)..article claims and told them internal tests show around 10% performance improvement.
This closely aligned with sisandra leak, which shows 6% frequency boost with QS sample of 2600...with about 3-4% in IPC...although this could be precision boost?.

I expect ~10%.

 

[The Stilt]

People here keep saying that Zen+ is going to be ~10% faster.

Was this ever officially said anywhere?

Or is this: Somebody uncle's 5th cousin-in-law said so?

I believe this is the source:

Doesn't say at ISO power, while GF's own marketing releases do.

 

[neblogai]

But we'll see if the R5 2600 indeed has 3.4GHz base and 3.8GHz maximum single core turbo (non-XFR for both counts) or not soon enough

I'd argue, that stock clocks of R5 2600 are not really important, they do not tell much. The SKU is made for segmentation, is aimed at 65W TDP, and it's only requirement is to be somewhat faster than previous product, R5 1600. AMD can be limiting it however they want- and we can only expect it's efficiency to be similar, and, if that Sandra benchmark loads all cores and reports correct clocks (3.4GHz)- new model to be ~8% faster at Sandras tests while at same clocks.
More interesting to me is the new Precision Boost 2, and how it will work with XFR. With Raven Ridge, and, presumably, 12nm Zen+ chips- cores do not stay at the same clocks, but can downclock a lot while other cores race. So if those racing cores are to have really free reign on both X and non- X CPUs- the performance in games could be about the same with overclocking or not. If not segmented artificially- it will depend only on having good cooling and the best die.
Here is a video of Raven Ridge cores downclocking all the way to 1.1GHz (and ~70% load), while most used cores clock 2.5-2.8Ghz and are also loaded ~70%. So I'm obviously interested in how it will work at highest clock spectrum for R5 2600 when chip is not power/temperature limited:
https://youtu.be/gc6T59or21Y?t=508

 

[Mockingbird]

I believe this is the source:

Doesn't say at ISO power, while GF's own marketing releases do.

GlobalFoundaries, I have ZERO faith in.

If GlobalFoundaries said that there's a 10% performance improvement, there's probably a 3% improvement (from 14nm --> 14nm+/12nm).

Now, AMD can make some changes (for example Precision Boost 2) that would give additional performance improvement, but I highly doubt that there would be a 10% performance improvement just from 14nm+/12nm.

 

[french toast]

GlobalFoundaries, I have ZERO faith in.

If GlobalFoundaries said that there's a 10% performance improvement, there's probably a 3% improvement (from 14nm --> 14nm+/12nm).

Now, AMD can make some changes (for example Precision Boost 2) that would give additional performance improvement, but I highly doubt that there would be a 10% performance improvement just from 14nm+/12nm.

Indeed, the 2600 leaks show ~6%...with some tests showing ~10% in subtests, alongside that article abwx? Posted a couple of pages back saying amd tested it at 10% also (or maybe they are just reference the same sisandra tests).

That's where the 10% is coming from..

 

[cytg111]

Indeed, the 2600 leaks show ~6%...with some tests showing ~10% in subtests, alongside that article abwx? Posted a couple of pages back saying amd tested it at 10% also (or maybe they are just reference the same sisandra tests).

That's where the 10% is coming from..

In this climate Id say that 6-10% is decent ...

 

[Abwx]

Estimating perfs improvements with the Sandra submission is accurate for the very reason that PN has the same ISA as SR.

FI if AVX512 would had been added to PN this would had completely blurred the numbers and rendered any comparison innacurate...

As for said 10% it was stated by AMD in a press event that was distinct from the CES and reported by a french journalist who was there :

C’est en marge du CES qu’AMD nous a conviés à un événement destiné à nous livrer ses plans pour 2018

Le gain de performances à attendre entre les deux séries se situe autour de 10 % sur les tests pratiques, si l'on en croit les données fournies par AMD

For those who missed the link :

https://www.lesnumeriques.com/cpu-p...-livre-plans-ryzen-2000-vega-7-nm-n69981.html

 

[eek2121]

People here keep saying that Zen+ is going to be ~10% faster.

Was this ever officially said anywhere?

Or is this: Somebody uncle's 5th cousin-in-law said so?

I go back to something I read a year ago. To paraphrase, AMD stated that Ryzen was a new architecture and when it comes to performance improvements there was plenty of low hanging fruit to tackle. I wish I could remember where I read it. Barring that, people are:

A) reading WAY too much into this single benchmark
B) most don't realize that the summit ridge benchmarks were flawed to begin with. Slow RAM, an inefficient Windows scheduler and power profile, and a new architecture painted a far more dire picture of Ryzen than is actually the case.

What I have learned thus far about Pinnacle Ridge is:

a) the memory controller has received improvements to increase compatibility and combat latency.

b) Clock speeds are receiving a slight boost.

c) A reworked boost algorithm means much stronger performance regardless of the core count used.

My bet is that the 2800x will either be 3.8-4.2 or 4.0-4.4 (the latter frequency is what I believe it will be, I have been saying this for months) and we will see double digit per clock performance improvements in things like games, etc. I have had my Threadripper at 4.4Ghz at 4-16 cores and I assure you, it already competes with the 8700k quite well, even with 4 12 cores disabled. Ryzen+ will simply seal the deal.

Don't take my word for it, wait for the reviews.

I will be sharing Gen1 benches pretty soon, and I hope to get a Gen+ platform at launch.

 

[CatMerc]

@Mockingbird In private talks with AMD they told multiple different people and outlets to expect 10% bump in general, depending on workload. I'm not going to dig up tweets now, but David Schor said so and I believe (but not certain) that Ian Cutress said he was told the same.

It's definitely something AMD said.

 

[eek2121]

@Mockingbird In private talks with AMD they told multiple different people and outlets to expect 10% bump in general, depending on workload. I'm not going to dig up tweets now, but David Schor said so and I believe (but not certain) that Ian Cutress said he was told the same.

It's definitely something AMD said.

You don't need to, what I read happened back at the original Ryzen launch. Despite the negativity, Ryzen was only behind in IPC compared to Kaby Lake by around 3% or so in gaming provided it was given a proper setup...and most of that was latency specific stuff (which supposedly PR fixes) except for the occasional rare app built on an Intel compiler. On my own (custom written) benchmarks (written in unity C#, though I remember enough OpenGL/C++ and can easily learn Vulkan should certain fanboys want to show up...not that they would believe me anyway, but even as someone who remains objective, it's fun to take idiots to school). Yeah, the 2xxx series is going to come out, and I'm going to grab the latest Intel chip at launch. i'll run about 100 benchmarks (already have on my 1950x) and provide enough details for users to make their own conclusions. They'll fight, they'll argue, they'll kick and scream...and they'll look like morons just like they did when Intel had it's first actual chip recall...then they'll make excuses, etc. Those that are old enough will remember. This is something like the 4th time ti has happened.

 

[Gideon]

Considering the incoming 6-core mobile Coffee-Lakes and that the desktop Ryzen 1700 has already been used in a laptop. I started wondering whether it would make sense to also do 12nm 8-core mobile-workstation versions of the chip?

I know It seems a bit far fetched, but when clocked right (targeting power rather than performance) a 12nm 2700 should be able to hit a 45W TDP and still be an excellent processor (pretty much bar-none) for heavily threaded apps.

 

[Abwx]

With 12LP the R7 2700 could be within 45W at 2.8GHz and 35W@2.5GHz, this amount to 1300pts and 1160pts respectively in CB R15...

Not sure that these will be used in laptops for this node, but at 7nm LP it is a certainity given that power will be cut by 60% in respect of the current Ryzen, a 7nm LP R7 1700 will be within 25W@3GHz..

 

[Yotsugi]

Not sure that these will be used in laptops for this node

These will probably see wins in DT replacement "laptops".

 

[bsp2020]

With 12LP the R7 2700 could be within 45W at 2.8GHz and 35W@2.5GHz, this amount to 1300pts and 1160pts respectively in CB R15...

I can't stop thinking about MCM package with this and Vega M (Pinnacle Ridge Lake G?). It should crush Kaby Lake G with ease. If AMD connects Pinnacle Ridge with Vega using Infinity Fabric... It will effectively double memory bandwidth to CPU, boosting CPU performance even higher.

 

[Topweasel]

These will probably see wins in DT replacement "laptops".

The biggest issue is going to be packaging. I doubt AMD will want to create a separate laptop package for the 8c Ryzen dies, knowing that they will only go into the handful of systems that will use a dGPU. Which means any implementation won't just be a DT replacement Desktop, but a pre-historic looking DT replacement laptop because it has to support a full AM4 socket. AMD might not even have a laptop socket solution that would work. Chances are that their mobile platform to keep the socket small only uses the pins needed for RR and BR and not SR and PR.

 

[raghu78]

These will probably see wins in DT replacement "laptops".

People keep forgetting that there is a 2019 Raven Ridge refresh called Picasso. Its most definitely going to be built at 12LP. btw Zen 2 based successor for Pinnacle Ridge built at 7LP is already showing up on HWINFO. I think we could see first samples in Q2 2018.

https://videocardz.com/72934/rumor-amd-matisse-picasso-vega-20-and-ryzen-5-pro-mobile

https://videocardz.com/newz/hwinfo-gets-support-for-amd-starship-and-mattise-cpus

AMD could obviously have a Pinnacle Ridge R7 based DT replacement chip running 8C/16T at 2.8 Ghz at 45W TDP. I would not be surprised to see them along with AMD mobile Vega GPU. It would be one real nice combo pack for OEMs. I think AMD was very confident that they will have the widest range of notebook design wins in 2018 based on Zen. I think RR for entry level notebooks and a 45w PR with a 75w Vega mobile GPU for DT replacement gaming notebooks is going to be AMD's notebook strategy for 2018.

 

[neblogai]

People keep forgetting that there is a 2019 Raven Ridge refresh called Picasso. Its most definitely going to be built at 12LP. btw Zen 2 based successor for Pinnacle Ridge built at 7LP is already showing up on HWINFO. I think we could see first samples in Q2 2018.

https://videocardz.com/newz/hwinfo-gets-support-for-amd-starship-and-mattise-cpus

If you look at the same link, but second slide- you can see Raven Ridge refresh (named Grey Hawk for Enterprise products->Picasso for consumer market) is 7nm, 4c8t, 10-35W. It looks a lot like a die shrink of RR level hardware.

 

[raghu78]

If you look at the same link, but second slide- you can see Raven Ridge refresh (named Grey Hawk for Enterprise products->Picasso for consumer market) is 7nm, 4c8t, 10-35W. It looks a lot like a die shrink of RR level hardware.

There is no 7nm APU in 2019. First 7nm Navi GPU and 7nm Zen 2 CPUs will launch by mid-2019. 7nm APUs are expected to launch by late Q4 2019 or early 2020. RR refresh is built on 12LP.

Sent from my SM-G935V using Tapatalk

 

[neblogai]

There is no 7nm APU in 2019. First 7nm Navi GPU and 7nm Zen 2 CPUs will launch by mid-2019. 7nm APUs are expected to launch by late Q4 2019 or early 2020. RR refresh is built on 12LP.

Sent from my SM-G935V using Tapatalk

If you read your post again- you will see you are contradicting yourself. You probably mean- 7nm not coming in 2018. Well- I never said it would. But do you have any source about 'RR refresh is built on 12nm'? Because these old slides do not show such thing. They only show Picasso in 2019 with better efficiency for consumer market, and 'Grey Hawk' APU in same 2019 at 7nm, 4c8t, 10-35W, so again- better efficiency. Judging by these slides- Picasso and Grey Hawk must be the same design, just like RR and Horned Owl, and no mention of 12nm RR refresh.
https://videocardz.com/newz/hwinfo-gets-support-for-amd-starship-and-mattise-cpus

 

[formulav8]

First 7nm Navi GPU

Looks like 7nm Vega is first, not Navi.

 

[raghu78]

If you read your post again- you will see you are contradicting yourself. You probably mean- 7nm not coming in 2018. Well- I never said it would. But do you have any source about 'RR refresh is built on 12nm'? Because these old slides do not show such thing. They only show Picasso in 2019 with better efficiency for consumer market, and 'Grey Hawk' APU in same 2019 at 7nm, 4c8t, 10-35W, so again- better efficiency. Judging by these slides- Picasso and Grey Hawk must be the same design, just like RR and Horned Owl, and no mention of 12nm RR refresh.
https://videocardz.com/newz/hwinfo-gets-support-for-amd-starship-and-mattise-cpus

Firstly the AMD slide from Feb 2016 which shows GrayHawk 7nm APU is very old and the slide which shows Picasso first leaked only 5 months back. So the Picasso information is likely to be accurate. btw if you follow AMD APUs they typically lag their new CPU and GPU arch by atleast 1-2 quarters. Ryzen and Vega launched first and then came Ryzen mobile with Vega graphics. This is why Zen 2 Matisse and Navi will arrive first and then the 7nm APUs combining Zen 2 and Navi will arrive a couple of quarters late. I expect Zen 2 and Navi to launch mid 2019 and 7nm APUs to launch in late 2019 or early 2020.

2018 - Raven Ridge 14LPP
2019 - Picasso 12LP
2020 - Zen2+Navi APU 7LP

 

[Yotsugi]

If you read your post again- you will see you are contradicting yourself. You probably mean- 7nm not coming in 2018. Well- I never said it would. But do you have any source about 'RR refresh is built on 12nm'? Because these old slides do not show such thing. They only show Picasso in 2019 with better efficiency for consumer market, and 'Grey Hawk' APU in same 2019 at 7nm, 4c8t, 10-35W, so again- better efficiency. Judging by these slides- Picasso and Grey Hawk must be the same design, just like RR and Horned Owl, and no mention of 12nm RR refresh.
https://videocardz.com/newz/hwinfo-gets-support-for-amd-starship-and-mattise-cpus

AMD always goes mobile last compared to Intel's mobile first.
Everything 12LP will end on 12LP-based APU refreshes.

The biggest issue is going to be packaging. I doubt AMD will want to create a separate laptop package for the 8c Ryzen dies, knowing that they will only go into the handful of systems that will use a dGPU. Which means any implementation won't just be a DT replacement Desktop, but a pre-historic looking DT replacement laptop because it has to support a full AM4 socket. AMD might not even have a laptop socket solution that would work. Chances are that their mobile platform to keep the socket small only uses the pins needed for RR and BR and not SR and PR.

I don't really think anybody cares about thickness in DT replacement "laptops", really.

 

[french toast]

There is no 7nm APU in 2019. First 7nm Navi GPU and 7nm Zen 2 CPUs will launch by mid-2019. 7nm APUs are expected to launch by late Q4 2019 or early 2020. RR refresh is built on 12LP.
Sent from my SM-G935V using Tapatalk

I would hope we get picasso Q3/4 2018...then 7nm APU same time 2019.
Vega 7nm will be ready for customer testing Q3 2018.. probably mass production Q1 2019.

Looks like 7nm Vega is first, not Navi.

Correct.

 

[Spartak]

Did anybody think of the possibility the difference between the expected ~6% frequency increase and the ~10% performance increase could be due to higher supported RAM speeds?

The core itself is exactly the same with near 100% certainty, and we all know that between 2666MT/s and 3200MT/s a lot of performance is to be gained. RR already supports 2933MT/s so we can expect at least the same from PR.

 

[Yotsugi]

~6% frequency increase

It's >~6% frequency increase though.

The core itself is exactly the same

Yet the changes to uncore may be not-so-slight.