New Zen microarchitecture details

[bjt2]

Except for the part where Jim Kellar talks about taking the DNA from there "high frequency designs" and their "low power designs" and mixing the best of both. he also talks about how amd know how to do high frequency circuit design and leveraging that for the next generation........

https://youtu.be/SOTFE7sJY-Q?t=4m45s

They will say that Keller meant that they will take low IPC from BD and low clock from Jaguar...

Yea, sounds good. What was he going to say, that it would clock low and use alot of power?
may 2014 .. wonder if Keller knew how 14nm FF at glofo was gonna mature? Fab deals. Upcoming Errata.
I want Zen to clock to the skies but I dont see that you have the evidence to support the claim at this time, only conjecture.

He meant low FO4. Obviously he couldn't know at that stage the features of 14nm but they know how to do high frequency designs: low FO4. If they managed to have 4.3GHz on 28nm BULK with HDL, surely he would not expect lower clock with same FO4 on 14nm...

 

[lolfail9001]

3.6GHz Base clock ES still at 95W

Didn't CanardPC tweeted that latest revision of Zen fails to meet 95W target on 8 core? Yep, they did.

retail 3.2GHz base clock at 140W TDP are a reality.

That's the part when you actually stop wearing any pretense at objectivity.

I can't believe that a chip with +400MHz base clock and less TDP will overclock less...

Happens all the time with Intel chips and some bad luck.

 

[cytg111]

They will say that Keller meant that they will take low IPC from BD and low clock from Jaguar...


He meant low FO4. Obviously he couldn't know at that stage the features of 14nm but they know how to do high frequency designs: low FO4. If they managed to have 4.3GHz on 28nm BULK with HDL, surely he would not expect lower clock with same FO4 on 14nm...

Yea, I get all that. And the odds, IMO, is definitely in your arguments favor. Here comes another devils advocate; broadwell, how did that 14nm beastie clock vs. 22nm haswell?
Bottom line, I probably wont believe anything for sure-sure until I read Anandtechs review, Toms, Hards and and... it is what it is.

 

[AtenRa]

Didn't CanardPC tweeted that latest revision of Zen fails to meet 95W target on 8 core? Yep, they did.

Could you please link to that ?? thanks

 

[leoneazzurro]

Yes, it seems they said on 3.6/4.0 ES their source said it slightly exceeeded the 95W mark. It was on a tweet from them, linked in one of the Zen threads, sorry but at the moment I cannot find it. As it is anyway a ES, and it is not understood on which platform this sample was mounted, it is also possible that final voltages are not set in stone, so final power usage may differ (if it differs, probably it would be lower). Anyway, also the 140W TDP on Broadwell-E is quite exaggerated compared to actual power draw. It is also true that there is no 3.6/4 GHz Broadwell-E around, and that also is likely to draw more power at those frequencies. So, all in all, it's quite more likely that Zen power draw will be lower anyway.

 

[coercitiv]

Could you please link to that ?? thanks

This is the tweet.

Les Ryzen 8C/16T actuels (3.6/4.0 GHz) dépassent bien les 95W de TDP. Logique vu les résultats obtenus dans nos tests.

Translation:

Current Ryzen 8C/16T chips (3.6/4.0Ghz) easily surpass 95W TDP. A logical outcome considering the results in our tests.

However, this is not a clear indication of anything except that Ryzen at higher than 3.6Ghz turbo for all cores exceeds 95W. It was hardly news even when CanardPC wrote the tweet.

As it is anyway a ES, and it is not understood on which platform this sample was mounted, it is also possible that final voltages are not set in stone, so final power usage may differ (if it differs, probably it would be lower).

It doesn't even have to be a voltage difference, the chips are unlocked and certain BIOS settings and/or XFR could easily allow the CPU to turbo past the TDP if certain criteria is met. (this may actually be intended behavior)

The only bad news would be Zen @ 3.6Ghz going past 95W, but the tweet made no such clear indication.

 

[bjt2]

Didn't CanardPC tweeted that latest revision of Zen fails to meet 95W target on 8 core? Yep, they did.

It's still an ES. Let's wait retail.

That's the part when you actually stop wearing any pretense at objectivity.

Tell me that what i said is false (TDP=140W), please. Actual consumption is lower (for not AVX2 code), but my statement it's true.
If they could guarantee 95W they would have rated it 95W and not 140W.

Happens all the time with Intel chips and some bad luck.

There will be lucky and unlucky samples, but if base frequency is higher, with lower consumption (and rated TDP), i think that they will overclock more.

 

[bjt2]

Yea, I get all that. And the odds, IMO, is definitely in your arguments favor. Here comes another devils advocate; broadwell, how did that 14nm beastie clock vs. 22nm haswell?
Bottom line, I probably wont believe anything for sure-sure until I read Anandtechs review, Toms, Hards and and... it is what it is.

22nm gained a lot respect 32nm thanks to FinFet. We are in a zone of diminished return, because of limits of the technology: the transistors are few atoms thick yet. You can't expect huge gains.

 

[Atari2600]

Didn't CanardPC tweeted that latest revision of Zen fails to meet 95W target on 8 core? Yep, they did.

While it galls me to be responding to someone who has "lol" and "fail" in their username (kinda indicating their balls are yet to drop), CPC did not say anything of the sort.

They said given the power consumption of their ES, they say that using a reasonable extrapolation, that the retail version will exceed 95W TDP at 3.6/4.0 GHz.

CPC do not have the latest revision (or at least, did not at the time of writing that). Given AMD demonstrated with a later revision at New Horizon an overvolted Zen @ 3.4 with a ballpark 95W TDP, with further respins it is likely that the early revisions (as CPC had) used much more power per clock than later.

 

[Ancalagon44]

To be fair, I won't be overly upset if an 8 core/16 thread part uses more than 95W under load. That is a lot of cores!

 

[Zucker2k]

It's still an ES. Let's wait retail.

Do you realize you just check-mated yourself with this statement? That's all your debaters have been trying to tell you for many pages. It may be that SOME, ALL, or NONE of the characteristics being exhibited by ES chips are going to be manifested in retail chips so your extrapolations are more than likely to be off.

Tell me that what i said is false (TDP=140W), please. Actual consumption is lower (for not AVX2 code), but my statement it's true.
If they could guarantee 95W they would have rated it 95W and not 140W.

Are you saying AMD's 95W rated octacore is going to achieve the same AVX2 throughput as Intel's 140W BDW-E cpu? In other words, how much power do you think the 3.6Ghz/4.0Ghz Zen part is going to consume in order to match the best throughput rate of BDW-E in best case AVX2 processing scenarios? Wait, don't answer. Let's wait for retail. LOL. I do wish AMD succeeds in lighting a fire under Intel's *** this time though. Competition is good for the consumer.

 

[AtenRa]

thanks for the link

 

[PhonakV30]

AMD TDP =/= Intel TDP
If AMD says 95w TDP , then It's Ok, Just please translate into Intel TDP.
Why do you compare AMD/Intel TDP when both formulas are completely different? Just convert them then use Logic.Simple.

 

[imported_jjj]

"Why should buyers wait even longer, putting off PC builds they so dearly want to start? At CES 2017 I put this question to Jim Anderson, AMD’s senior vice president and general manager of the computing and graphics business group.
“I think what the consumer will see when we fully launch is that the amount of performance we’re bringing per dollar is phenomenal. It’s user experience per dollar. That’s why I’d say I’d wait.” "
http://www.trustedreviews.com/opinions/amd-2017-worth-the-wait

 

[LTC8K6]

"Why should buyers wait even longer, putting off PC builds they so dearly want to start? At CES 2017 I put this question to Jim Anderson, AMD’s senior vice president and general manager of the computing and graphics business group.
“I think what the consumer will see when we fully launch is that the amount of performance we’re bringing per dollar is phenomenal. It’s user experience per dollar. That’s why I’d say I’d wait.” "
http://www.trustedreviews.com/opinions/amd-2017-worth-the-wait

But he'd say that anyway...and what does it really mean?
And compared to what?

 

[cytg111]

"Why should buyers wait even longer, putting off PC builds they so dearly want to start? At CES 2017 I put this question to Jim Anderson, AMD’s senior vice president and general manager of the computing and graphics business group.
“I think what the consumer will see when we fully launch is that the amount of performance we’re bringing per dollar is phenomenal. It’s user experience per dollar. That’s why I’d say I’d wait.” "
http://www.trustedreviews.com/opinions/amd-2017-worth-the-wait

Yea, I'd wait too unless my only rig had caught fire or something. + we need some rev 1.0 testers

 

[Atari2600]

Yea, I'd wait too unless my only rig had caught fire or something. + we need some rev 1.0 testers

Even then - I think I'd scrounge a laptop off someone for a few weeks...

 

[dahorns]

But he'd say that anyway...and what does it really mean?
And compared to what?

Good point. Even without Ryzen, AMD would have used that line. It has kinda been their thing the last few years.

 

[imported_jjj]

But he'd say that anyway...and what does it really mean?
And compared to what?

They can price it pretty much anywhere but phenomenal is a pretty strong term, hopefully it's not vs the upper part of the HEDT line.

 

[KTE]

Now that is utter FUD (worthless opinions are own).

The Barcelona 3.0 GHz was at the level of a fudzilla rumour.


We've now got independently verified CPUIDs of 3.6/3.9 8C Zen which are supposed to have a 95W TDP. If a 3.6/3.9 8C can fit under 95W, why is it a surprise that a 3.4 GHz chip (even with mild overvolting to ensure stability) also comes in under 95W?

If the 3.6/3.9 wouldn't have a hope of fitting inside the 95W AM4 TDP, then AMD would not be building them! (If you think otherwise then your really into the tinfoil hat faked moon landings level of conspiracy theorist.)

All what you've said is a knee-jerk reaction when you don't read what you're replying to.

We're talking about POWER not clocks.

Phenom does not have any sleep state and any clock gate.
It's made in pure dynamic logic, that can't be shut down.
If you look at page 2, there is the CPUID screen.
As you can see, the multiplier and Vcore is fixed.
So in idle is sitting at 2500MHz as under load.
Probabily the power consumption in idle is very high.
170W for MB, NB, SB, GPU all in IDLE seems quite high, even factoring the efficiency.
With Ahtlon 64 was introduced the IDLE at 800 or 1000MHz and reduced Vcore, and there started to draw much less...
Only recently AMD has put enough attention to energy saving.
They are not comparable.
Since only very old reviews agree vith your vison, have you ever had a doubt that maybe that results were not comparable, because are of a very OLD CPU (9 years ago) that don't have any energy saving feature?

Seriously.

I added 35W DC to idle. How much more do you want!?

Phenom DID have C1e and CnQ as standard, as well as separate power planes.

Idle sat at ~1.0v 1.2GHz not 2.5GHz - just 2 of why all the above is incorrect assumptions, again.

You keep saying not comparable, why? Because your deduction doesn't work for them?

Science does not change for chips. Power measuring remains the same.

Sent from HTC 10
(Opinions are own)

 

[raghu78]

Do you realize you just check-mated yourself with this statement? That's all your debaters have been trying to tell you for many pages. It may be that SOME, ALL, or NONE of the characteristics being exhibited by ES chips are going to be manifested in retail chips so your extrapolations are more than likely to be off.

One thing is for sure. The retail chip will be better in every aspect than an ES. So whatever we are seeing with ES the retail chip should do better.

Are you saying AMD's 95W rated octacore is going to achieve the same AVX2 throughput as Intel's 140W BDW-E cpu? In other words, how much power do you think the 3.6Ghz/4.0Ghz Zen part is going to consume in order to match the best throughput rate of BDW-E in best case AVX2 processing scenarios? Wait, don't answer. Let's wait for retail. LOL. I do wish AMD succeeds in lighting a fire under Intel's *** this time though. Competition is good for the consumer.

Very simple. For AVX2 workloads Intel will still be significantly ahead in perf and draw more power. Whether Intel has better perf/watt on AVX2 loads we have to wait and see how much the extra perf and power usage are. But for the majority of FP and integer workloads AMD Ryzen should provide competitive perf and perf/watt against Broadwell. Now that would be a refreshing change after almost a decade of Intel dominance. Intel would still rule gaming and lightly threaded workloads (<= 4 threads) with the 7700k and its 4.5 Ghz turbo.

 

[bjt2]

Do you realize you just check-mated yourself with this statement? That's all your debaters have been trying to tell you for many pages. It may be that SOME, ALL, or NONE of the characteristics being exhibited by ES chips are going to be manifested in retail chips so your extrapolations are more than likely to be off.

Retail can't be worse that the best ES. Actually i would expect that they are better, because we don't know how old are that ES (they are somewhat bugged), the MB and the BIOSes are not final and last steps could be better.
This is what i mean with that phrase.
+400MHz with a slightly lower power for an ES vs retail is in any case a good result. Even if retail can't do better, and for sure they will not perform worse, it's a good chip. What is the statement that you found strange? That a chip with +400 default clock and less power consumption will overclock more? Yeah... You are right... It will overclock 500MHz less... It will actually explode if you try to overclock... Ok? Are you happy?

Are you saying AMD's 95W rated octacore is going to achieve the same AVX2 throughput as Intel's 140W BDW-E cpu? In other words, how much power do you think the 3.6Ghz/4.0Ghz Zen part is going to consume in order to match the best throughput rate of BDW-E in best case AVX2 processing scenarios? Wait, don't answer. Let's wait for retail. LOL. I do wish AMD succeeds in lighting a fire under Intel's *** this time though. Competition is good for the consumer.

Never said that. I would expect that 3.6/4.0 Zen will draw 95W, if it will be rated 95W.
Blender is not the right bench for this comparison, because it seems that The Stilt's build increase throughput also on BD, and of a big percentage (80%), so i don't expect AVX2 clock to kick in in 6900K...

 

[bjt2]

All what you've said is a knee-jerk reaction when you don't read what you're replying to.

We're talking about POWER not clocks.


Seriously.

I added 35W DC to idle. How much more do you want!?

Phenom DID have C1e and CnQ as standard, as well as separate power planes.

Idle sat at ~1.0v 1.2GHz not 2.5GHz - just 2 of why all the above is incorrect assumptions, again.

You keep saying not comparable, why? Because your deduction doesn't work for them?

Science does not change for chips. Power measuring remains the same.

Sent from HTC 10
(Opinions are own)

The rated TDP is 125W but at that times It was INTEL that was famous to be generous for the TDP and AMD was conservative, if you remember. Anyway that CPUs didn't have turbo, so the TDP was calculated with a power virus, with Tj=125C and Vcore +10% nominal.
If you read what you posted, in a game the CPU peaked at 269W, over 20W more than you posted. And a Game is not a 100% load, since we know that all modern CPUs will turbo in this case. And 2008 games were not famous to be multithreading... So we have 80+35=115. In line with 125W. But ok, link me a review of an 8088 to prove your point...

 

[Atari2600]

We're talking about POWER not clocks.

I don't think you know what you are talking about. POWER and CLOCKS are directly linked. They always have been and unless something dramatic changes, as far as silicon is concerned, always will be.

Relax your power requirements and you will obtain better clocks for 99+% of silicon. (Unless you live in a land where this doesn't apply, neither does gravity and HTC10s do indeed have their own opinion on internet forums?)


CPC have indicated they see AMD getting their target clocks. They have said, based on extrapolation from their early (A0?) ES, they didn't see this happening at <95W.

But, AMD have subsequently demo'd on later respin supposedly overvolted parts @3.4 GHz at 95W. Even later steppings are said to have further improved on the clock/power ratios.


edit: For clarity, I don't give 2D for the discussion over relative power consumption of Zen vs. BWE, its too finely balanced to know now for sure how it will go, and it will probably change depending on the instructions requested by each software pack. But I do note, and don't care for, the strawman argument which attempts to say that methods that could be valid now are invalidated due to the results of older generations of CPU - which have very different Cstate(ULT) values.

 

[Abwx]

This is the tweet.

Translation:

Translation is 100% wrong, and i already pointed that such ltteral translations have nothing to do with what is stated.

Nowhere it is said that 95W are exceeded easily, the term "bien" mean "for good" or "effectively", but is in no way related to a quantification of the excess.