New Zen microarchitecture details

[DrMrLordX]

What exactly would prevent AMD from supporting the chipsets of AM4 platform on, let's say Vista and newer? IMO exactly nothing.

Depends on a lot of factors. Sure they could support Vista on up, but . . .

That would be a real freaking shame, IMHO. If true, that would also mean next to no traction for AM4 platform among business buyers.

Are we talking people buying servers or workplace desktops? I would think AMD would be quickest to offer "legacy" OS support for firms with numerous old Windows Server licenses. For large buyers with major Win7 desktop installations, they might do custom jobs the same way MS offers support for XP on a demand basis.

For everyday schmucks trying to hold on to Win7 just a little bit longer . . . frankly I don't know what incentive AMD has to cater to that audience. MS is probably going to pay them under the table to move primary support to Win10 anyway. Something is pushing Intel in that direction. It wouldn't surprise me if there's a quid pro quo of some kind.

You realize Kaby Lake will have no support for Windows 10 either?

Even Skylake will lose all but critical security support after July 2018

In other words, ALL businesses will have to move to windows 10 very shortly if they intend to purchase anything new in 2017

That's exactly what I was thinking. Intel's going that route now, and AMD will probably follow suit.

The argument that Skylake was purposefully designed to not step on Broadwell's toes seems contradicted by pretty much everything I've read, including the fact that Skylake shipped with an EDRAM controller.

Well, the other thing is . . . if you don't want to undercut Broadwell-C with Skylake-C (or whatever it would be called), why release a new chip that's actually slower than the old one? The only reason I could see had to do with margins. Intel knew they could get away with it due to supply constraints (that you cited in your post), and they probably figured they could reap greater profits selling a non-eDRAM part as their flagship consumer CPU.

Broadwell-C sort of let the cat out of the bag on what Intel could be doing to keep pushing performance. They're deliberately not doing it because they see no reason to outdo themselves by more than %5 per generation. What they don't seem to understand is that their penny-wise, pound-foolish approach has made the desktop profoundly uninteresting to many buyers who just see no need to participate in the desktop market, period. One could argue that Intel's decisions wrt Skylake etc. are driving down industry sales. AMD isn't helping matters either, really. Maybe with Zen, they can change that. Maybe.

 

[dooon]

AIDA64 Extreme Version: 5.75.3916 beta (Jul 14, 2016)
http://www.aida64.com/downloads/YTdiMzQyZGM=

chipset information for AMD K17 (Zen Summit Ridge) FCH
chipset information for AMD K17 (Zen Summit Ridge) IMC
chipset information for AMD K17.1 (Zen Raven Ridge) FCH
chipset information for AMD K17.1 (Zen Raven Ridge) IMC
chipset information for AMD Promontory ASM2016
SMBus support for AMD K17 (Zen Summit Ridge) FCH
SMBus support for AMD K17.1 (Zen Raven Ridge) FCH

 

[sm625]

I predict that the top tier 95W Zen CPU from the initial release group will have the following passmark scores:

13000 Multithread
1800 Single thread

This prediction is based on the fact that Nvidia engineering + TSMC 16nm appears to be more than 50% ahead of AMD engineering + GF14nm in terms of perf per watt. And Intel should at least be able to match Nvidia+TSMC, if not outright beat them.

 

[swilli89]

I predict that the top tier 95W Zen CPU from the initial release group will have the following passmark scores:

13000 Multithread
1800 Single thread

This prediction is based on the fact that Nvidia engineering + TSMC 16nm appears to be more than 50% ahead of AMD engineering + GF14nm in terms of perf per watt. And Intel should at least be able to match Nvidia+TSMC, if not outright beat them.

Not intended to offend, but this is about the worst approximation I think I've ever seen in my life.

 

[sm625]

Not intended to offend, but this is about the worst approximation I think I've ever seen in my life.

Would you care to place a bet on what the passmark scores will be?

 

[frozentundra123456]

That is about equal to stock 5820K, so it doesnt seem unreasonable, especially if they really adhere to the 95 watt TDP.

Zen will have 8 cores, so that is 33% more cores, but if the ipc is 10% less and clockspeed is 10% less, you are left with about 10% faster in multithreaded than 5820K.

 

[Insert_Nickname]

Zen will have 8 cores, so that is 33% more cores, but if the ipc is 10% less and clockspeed is 10% less, you are left with about 10% faster in multithreaded than 5820K.

Wait, wait. AMD competing with a 5820K? I can live with that... :biggrin:

 

[guskline]

I notice that the AIDA64 download dooone linked to is an Engineering beta. What significance does that have? Is Zen starting to be more rigorously tested "out in the wild"?

What info, if any can we derive from these Aida64 listings?

 

[MajinCry]

That ain't going to be remotely close ta true. Expectin' the wee broke guy ta come out with something that'll compete with the corpulent blue goliath is, err, gonna set ya up ta be let right down.

Zen's gonna be around Sandybridge-Ivybridge performance, if that +40% performance improvement thingie is reliable. There's nae chance of AMD being able ta trade blows with Haswell's finest.

 

[coffeemonster]

I predict that the top tier 95W Zen CPU from the initial release group will have the following passmark scores:

13000 Multithread
1800 Single thread

Excavator in the Athlon 845 is 1774 single thread at 3.5(3.8 turbo)

if ZEN has +40% IPC at the same clocks then it would be more like 2483 single thread

 

[frozentundra123456]

Excavator in the Athlon 845 is 1774 single thread at 3.5(3.8 turbo)

if ZEN has +40% IPC at the same clocks then it would be more like 2483 single thread

You seriously think it will be 20% faster than Haswell (5820K) in single thread??

 

[DrMrLordX]

I'm surprised the 845 gets a 1774 ST in Passmark.

 

[KTE]

I'm surprised the 845 gets a 1774 ST in Passmark.

http://www.anandtech.com/show/10436/amd-carrizo-tested-generational-deep-dive-athlon-x4-845

It would be a BIG achievement for me if Zen topend 8-core 95W at launch could get ~1800ST with <3300MHz.

That would make Zen 1GHz ~ EXC 1.15GHz. So Zen would need to start >3.4GHz to outdo older AMD CPUs in low-threaded apps.

4 Core scaling is usually 3.5-3.75 in good synthetic tests.

While I suspect Zen 1GHz = EXC 1.2GHz on average, frankly, I don't see the above power and frequency requirements happening.

Sent from HTC 10
(Opinions are own)

 

[el etro]

Passmark is not more a relevant benchmark as it was someday... even Geekbench shows better the computer performance these days.

 

[Dresdenboy]

Hiroshige Goto wrote an excellent piece about AMD's power saving tech in Polaris and CZ/BR (voltage droop mitigation, AVFS, etc.):
http://pc.watch.impress.co.jp/docs/column/kaigai/1010459.html (Japanese)
https://translate.google.com/transl...jp/docs/column/kaigai/1010459.html&edit-text= (A Xl8)

I post it here, as it might be found in Zen, too.

 

[AMD Polaris]

Hi all, I've got some info about Zen ES stage. I thought you would like to see it after the many off-topic Polaris posts.

Zen ES is at the moment in revision A0 - it might not be a suprise.

L2/L3 variations: 2/8 MB, 4/16 MB, 8/32 MB, 12/64 MB, 16/64 MB
(512kb L2/core, 8MB/4 cores)

Core counts are: 4c/8t, 8c/16t, 16c/32t, 32c/64t. As it seems now there won't be a 6c/12t at the launch, there will be only complete core complexes. Later AMD might release a 6/12 version, will see.

AMD's working on 2 kind of packages: AM4 and SP3. Later there might be a SP4 package of course.

4 variants of ES Zen are available at the moment:
AM4 8 cores with 95W TDP
AM4 4 cores with 65W TDP
SP3 24 cores with 150W TDP
SP3 32 cores with 180W TDP

The most exciting part is core clock. The 8c/95W variant's base clock is 2.8GHz, all core boost is 3.05GHz and maximum boost is 3.2GHz.
The 4c/65W part's clock is the same. (I would expect 3.5GHz base clock for a retail 4c/95W variant.)
Idle clock is exciting as well. AM4 versions can lower the clock to 550 MHz in idle which is a very nice level from an AMD CPU. Idle wattage is 5W for 8c version and 2.5W for 4c version.
The SP3 versions have even lower idle clock: it's only 400MHz. Regarding the boost clocks the 32c/180W version has a 2.9GHz boost clock and the 24c/150W version has a 2.75GHz boost clock.

AFAIK Intel has no answer for the 32c/64t Zen variant, so it could be a great win for AMD on the server market.

What AMD is doing different in case of Zen is the purpose of the CPU. AMD doesn't build it for the future but for the present. Maybe it won't be strong in AVX and FMA but an average user won't realize it because the average user won't use the AVX and FMA capabilities. Average Joe will see that Zen is as fast as Haswell on Intel's side (if AMD won't screw it up) and it's enough for everything what Average Joe does on his PC.

On server side Haswell IPC and high core count with a reasonable price will convince the customers to give AMD a try. And I hope AMD can grab the chance after all the years of failures. :thumbsup:

 

[KTE]

Hiroshige Goto wrote an excellent piece about AMD's power saving tech in Polaris and CZ/BR (voltage droop mitigation, AVFS, etc.):
http://pc.watch.impress.co.jp/docs/column/kaigai/1010459.html (Japanese)
https://translate.google.com/transl...jp/docs/column/kaigai/1010459.html&edit-text= (A Xl8)

I post it here, as it might be found in Zen, too.

Mr Naffziger is slowly bringing his Itanium expertise to AMD

Hiroshige states 28nm to 14nm FF nets 65% lower power or 55% higher performance, 15% CU gains, 1.9-2.8x performance per watt, at 53% die size (in light of AMD GPU slides).

5% Core gain, 30% lower power at 55% die size would alone be a winner for Zen...







Sent from HTC 10
(Opinions are own)

 

[Roland00Address]

I predict that the top tier 95W Zen CPU from the initial release group will have the following passmark scores:

13000 Multithread
1800 Single thread

This prediction is based on the fact that Nvidia engineering + TSMC 16nm appears to be more than 50% ahead of AMD engineering + GF14nm in terms of perf per watt. And Intel should at least be able to match Nvidia+TSMC, if not outright beat them.

So you are arguing it is worse than the i7 2600 in single thread at a passmark benchmark, but better than the i7 2600 in multithread at a passmark benchmark due to 8 actual cores?


i7 2600
ST 1921
MT 8257

13000/8257=57% higher

 

[mikk]

What AMD is doing different in case of Zen is the purpose of the CPU. AMD doesn't build it for the future but for the present. Maybe it won't be strong in AVX and FMA but an average user won't realize it because the average user won't use the AVX and FMA capabilities. Average Joe will see that Zen is as fast as Haswell on Intel's side (if AMD won't screw it up) and it's enough for everything what Average Joe does on his PC.

On server side Haswell IPC and high core count with a reasonable price will convince the customers to give AMD a try. And I hope AMD can grab the chance after all the years of failures. :thumbsup:

To me it sounds like some wishful thinking written from some AMD fanboy. And given how many fakes we got from AMD stuff in latest years I somehow doubt this is true. It's not even clear if these predictions are based on some official data/tested data or just some stuff made up from your wishful thinking. That's very dangerous. I think this is mixed with wishful thinking.

 

[AMD Polaris]

To me it sounds like some wishful thinking written from some AMD fanboy. And given how many fakes we got from AMD stuff in latest years I somehow doubt this is true. It's not even clear if these predictions are based on some official data/tested data or just some stuff made up from your wishful thinking. That's very dangerous. I think this is mixed with wishful thinking.

Yeah, I understand you. If I were you I wouldn't believe either to a nobody who comes from out of the blue and writes things like that.

Please do not be deceived by my "wishful thingking", focus on the technical info instead.

Regarding my "wishful thinking": I've tried to stand with two feet on the ground when I wrote those thoughts at the end of the post. Please pay attention to this: "if AMD won't screw it up". It's the key of everything what AMD is gonna show us in the next 6 months.

 

[Arachnotronic]

What AMD is doing different in case of Zen is the purpose of the CPU. AMD doesn't build it for the future but for the present. Maybe it won't be strong in AVX and FMA but an average user won't realize it because the average user won't use the AVX and FMA capabilities. Average Joe will see that Zen is as fast as Haswell on Intel's side (if AMD won't screw it up) and it's enough for everything what Average Joe does on his PC.

On server side Haswell IPC and high core count with a reasonable price will convince the customers to give AMD a try. And I hope AMD can grab the chance after all the years of failures. :thumbsup:

Based on what you have written, it seems "average Joe" would be better served with something like a 4790K (4.2GHz base/4.4GHz turbo) with its much higher single-threaded performance than with an 8C/16T CPU.

 

[dark zero]

Based on what you have written, it seems "average Joe" would be better served with something like a 4790K (4.2GHz base/4.4GHz turbo) with its much higher single-threaded performance than with an 8C/16T CPU.

Even a Celeron N is OK for an Average Joe...

 

[kapulek]

The 4c/65W part's clock is the same. (I would expect 3.5GHz base clock for a retail 4c/95W variant.)

I expect 4GHz minimum with turbo for a 4 core part.

 

[AMD Polaris]

Based on what you have written, it seems "average Joe" would be better served with something like a 4790K (4.2GHz base/4.4GHz turbo) with its much higher single-threaded performance than with an 8C/16T CPU.

Those who need high SP will choose Intel. If MP is in the focus they will consider AMD's 8c/16t CPUs.

 

[Arachnotronic]

Those who need high SP will choose Intel. If MP is in the focus they will consider AMD's 8c/16t CPUs.

Or the 8C/16T Intel parts. Or the 10C/20T Intel parts.