New Zen microarchitecture details

[Atari2600]

Right folks - we are never going to settle the argument about what the upper limit of OC will be right now!

Its like a crowd of skinheads arguing over whats better, a comb or hairbrush.


What is more interesting to me is - does AMD's approach with PurePower/PrecisionBoost/XFR pretty much allow the user to set:
1. A noise limit on their cooling system
2. An electrical power limit on the CPU

Then the CPU automatically gates some cores and clocks others to get best performance within temperature and power limits. If adjusting a couple of settings (in software), allows users to easily get up to 4.4 GHz with the wraith cooler --- does this mean the vast majority of users* would be better served with a Zen than the alternative?


*The vast majority don't mess around with overclocking in bios. About the most they might do is adjust power profiles in windows. This might appear as sacrilege to some hardcore geeks, but it just happens to be the way things are.

 

[cytg111]

Someone is not saying the truth or the ES that are going around are very different. IF IT'S TRUE that someone got a 5GHz air overclock, even on one core, there is no way that the air OC is limited to 4.4GHz. Every overclocker knows that the power will exponentially rise with clock, so 5GHz on air, even if it's a suicide run, means at least 4.6-4.7GHz on all cores. This IF Canard PC source said the truth.
And that was a buggy early ES. According to B&C newer steps clocks even better... So... I don't think that on retail chips they will be limited to 4.4GHz OC. Maybe early ES. And even this is questionable if Canard PC source said the truth...

That is some heavy assumptions to begin calling someone a liar.. IMO.

 

[KTE]

That is some heavy assumptions to begin calling someone a liar.. IMO.

Furthermore, that is what happened with the first Agena. 1Core could validate 2.9-3.1GHz while whole chip could only do 2.4-2.5GHz stable.

Because the 65nm process was dog.

Sent from HTC 10
(Opinions are own)

 

[LTC8K6]

Did we see any indication of the single core max in any of the leaks?

If the leaked ES chip was 3.2/4.0 for example, then we should see roughly the same ratio on the retail chips.

For 5.0 max, that would be a 4.2 base clock in my example.

But, I don't recall any turbo info leaking out at all.

 

[JoeRambo]

Furthermore, that is what happened with the first Agena. 1Core could validate 2.9-3.1GHz while whole chip could only do 2.4-2.5GHz stable.

And that was just 3 extra cores, we are talking about 7 more now. The gap between 1 core and 8 core will not be 300hz, that's pretty much certain?

 

[bjt2]

So Your claim is that AMD is incompetent in only releasing 3.4+Ghz chip, when they could release an uber winner @4ghz or so at higher TDPwattage? Instead of releasing a winner, they are sticking with 3.4Ghz base and leaving >1Ghz on the table???

It is way more likely 8C Zen will top out @some 4.2-4.3Ghz on top air and power usage will be rather high.

Note that those are still very respectable clocks for performance they are showing us, but zealotry is doing no good for Zen.

3.4GHz+ was the baseline (can be higher) and we don't even know turbo. FX8370E is an 8 core, 95W, has 3.3GHz base, 4.3 turbo and we all know how much it can overclock... With a fast search on google I found at least 4.5GHz stable OC, and 5GHz+ windows boot on air at Tom's hardware... And this is 32nm and even not power optimized (from Carrizo on 28nm they took seriously the power saving route). Not mentioning the Bristol Ridge at 4.9GHz@1.43V... 14nm has more transconductance, less Vth, less capacitance... Why could not overclock more or at least same?

 

[KTE]

And that was just 3 extra cores, we are talking about 7 more now. The gap between 1 core and 8 core will not be 300hz, that's pretty much certain?

Mate, I have no idea what trip he is on.. He sees things I don't so I'll let him explain



Sent from HTC 10
(Opinions are own)

 

[bjt2]

That is some heavy assumptions to begin calling someone a liar.. IMO.

If someone says 4.2-4.4GHz max and another says 5GHz, either one of the two is lying or the ES were highly inhomogeneous... I think that i didn't say nothing surprising... What other explaination is there?

 

[CatMerc]

I asked Buildzoid on reddit about how recent his knowledge is. He said less than a month old, so we're talking one of the newer samples.

 

[Nothingness]

Every overclocker knows that the power will exponentially rise with clock

That's wrong. Power will grow linearly with frequency unless you start playing with voltage

 

[antihelten]

If someone says 4.2-4.4GHz max and another says 5GHz, either one of the two is lying or the ES were highly inhomogeneous... I think that i didn't say nothing surprising... What other explaination is there?

It is actually possible that both were telling the truth, since the two claims are possibly for very different scenarios.

It's important to note that the source saying 4.2-4.4 GHz said that this was the max with a realistic number of cores and a realistic voltage* (i.e. non-suicide run), whereas the 5 GHz was a single core run and potentially with extreme voltage (i.e. suicide run).

So a non-suicide run might be 4.2-4.4 for all cores and 4.5-4.7 for a single core, whilst a suicide run would be around 4.7 for all cores and 5 GHz for a single core.

*The exact quote was "it's not going to happen at any voltage or core count that's actually practical"

 

[jur]

Maybe some ESs came from GF, wheareas others were from Samsung.
https://twitter.com/BitsAndChipsEng/status/814894218544037892

 

[bjt2]

That's wrong. Power will grow linearly with frequency unless you start playing with voltage

At same temperature. Power goes exponentially with temperature. At same heat sink, increasing frequency, increase power. And for high clock, you must raise also Vcore and then the temperature will rise further... We are talking of extreme OC in which you need higher Vcore even for 100MHz...

 

[KTE]

Thats dynamic power you guys are talking about. Typically f^3.

Static is unaccounted for here, and rises greatly above a process dependent V.

And yes, temps influence leakage current and Pd.


But nothing beats my new Duron Morgan

Sent from HTC 10
(Opinions are own)

 

[dark zero]

If someone says 4.2-4.4GHz max and another says 5GHz, either one of the two is lying or the ES were highly inhomogeneous... I think that i didn't say nothing surprising... What other explaination is there?

Or maybe is 4.2 on air and 5 on water cooling.

 

[antihelten]

Or maybe is 4.2 on air and 5 on water cooling.

CPC (the ones saying 5GHz) specifically said that it was on air

 

[Abwx]

That's wrong. Power will grow linearly with frequency unless you start playing with voltage

That s a fallacy here, you cant raise frequency if you dont raise the voltage unless you are eating in the voltage margin and hence in stability margin, so it s you who is wrong technically speaking.

Thats dynamic power you guys are talking about. Typically f^3.

Only at the extremity of the frequency/voltage curve, for average values power scaling as a square ensure good enough stability margin, check the voltage/frequency curves of current CPUs if you have any doubt about it...

 

[inf64]

I think that CPC's point was to tell us that Zen core is capable of running at those clocks without extreme cold which means that process node is very good and core can scale frequency wise. It doesn't matter that it will not clock that high across all 8 cores for 24/7 usage. 6900K cannot do it and I seriously doubt SKL-X will be any better than Broadwell-E. AMD is basically set to compete with intel's best for the whole 2017 and 2018 with one core.

 

[cytg111]

Besides the added number of cores, why wouldnt SKL-X clock alot like Kabylake?

 

[Abwx]

I think that CPC's point was to tell us that Zen core is capable of running at those clocks without extreme cold which means that process node is very good and core can scale frequency wise. It doesn't matter that it will not clock that high across all 8 cores for 24/7 usage. 6900K cannot do it and I seriously doubt SKL-X will be any better than Broadwell-E. AMD is basically set to compete with intel's best for the whole 2017 and 2018 with one core.

Did someone mention that CPC stated that the MB VRMs where unfit to feed all 8 cores at 5GHz, the MB being an early and minimal plateform..?..

Anyway if it can work at 5GHz with one core probability is about 100% that it can work at 4.5GHz with all cores loaded.

 

[Nothingness]

That s a fallacy here, you cant raise frequency if you dont raise the voltage unless you are eating in the voltage margin and hence in stability margin, so it s you who is wrong technically speaking.

I'm not discussing that. When you are aiming at technical correctness, you need to be accurate. Saying that power increases exponentially with frequency is simply not fully correct. Someone with your technical knowledge should be well aware that you have to be careful with simplistic statements with hidden assumptions

Can't wait to get more info about Zen. Is it known when AMD will talk about it at CES?

 

[Vesku]

4.2-4.4 before power draw gets "crazy" is a fairly reasonable leak from someone very likely to have access to Zen.

Sounds like it may be the design limit of Zen on the node they're using. As the person highlighting the 4.2-4.4GHz statement pointed out, no 4.8/5GHz 4 core on air would mean the Kabylake Ks will keep their niche among extreme OCers.

 

[.vodka]

4.2 to 4.4 from GloFo first slice of the wafer. Ill take it. Bodes well going forward, I am seeing ~5 in my crystal ball here.

http://www.anandtech.com/show/2754

Die size hasn’t changed, clock speeds barely went up, and performance per clock also remained static. But what’s this?

Ah yes, AMD is improving its 45nm manufacturing process and today we have the latest incarnation of AMD’s 45nm silicon.

The first versions of AMD’s 45nm Phenom II couldn’t really go much higher than the final 65nm Phenom without increasing voltage. By comparison, Intel’s Core i7 920 could go from 2.66GHz all the way up to 3.80GHz without so much as a single extra millivolt in our tests.

This new Phenom II however can also hit 3.80GHz without increasing the core voltage. At least that’s what one of our samples did in our testing. Whether it’s 3.8GHz or 3.6GHz, the fact of the matter is that AMD’s 45nm process is improving and that’s what’s behind todays introduction of the Phenom II X4 955. Architecturally the Phenom II hasn't changed; if you're curious about what makes these things tick, please look at our original article on the CPU.

AMD/Glofo is known for improving their processes as time passes... Deneb shipped at up to 3.8GHz on the PII X4 980 near Bulldozer's launch. Same thing happened with 32nm Vishera vs the latest batches of Vishera that debuted on the FX 8370e. This is bound to happen as 14nmFF matures more and more, and the design gets tweaked here and there.

I agree, for the first batches out of the fab, considering everything, those are VERY respectable clocks for 24/7 usage. Of course the 95w TDP window will be left behind at those clocks and voltages, still I'd like to see how much the CPU overclocks itself when properly cooled.

Did someone mention that CPC stated that the MB VRMs where unfit to feed all 8 cores at 5GHz, the MB being an early and minimal plateform..?..

Anyway if it can work at 5GHz with one core probability is about 100% that it can work at 4.5GHz with all cores loaded.

Yes they did. CPC stated that their source's sample could do 5GHz on a single core, because the motherboard's VRM that it was tested on became unstable with more cores at those speeds. This is excellent news, the Zen core and its current implementation CAN hit 5GHz. It's not clock speed capped by the microarchitecture or the process, and that going forward seeing the evidence of past CPUs getting tweaked and refined... means better overclocking on newer batches. This is no Agena (thankfully). Then there will be variability between Samsung and Glofo parts even though they both use the same processes..

I suppose better, final motherboards, properly overengineered ones (hope we see some in CES) can and will probably have a better chance at trying a bit higher, 4.5GHz for example, on all cores on current samples.. or whatever we get at launch.

4.2-4.4 before power draw gets "crazy" is a fairly reasonable leak from someone very likely to have access to Zen.

Sounds like it may be the design limit of Zen on the node they're using. As the person highlighting the 4.2-4.4GHz statement pointed out, no 4.8/5GHz 4 core on air would mean the Kabylake Ks will keep their niche among extreme OCers.

This clock range has been mentioned here in this forums a lot of posts ago, now Flank3r says so, and now this. I'm okay with a first gen 8c16t CPU doing these clocks with lots of tweaking to be done going forward. Broadwell-E does these clocks.

Kabylake will take the OC crown, that's expected. Still, that's a 4c8t CPU, 4c8t Zen can definitely be competitive with Kabylake if priced right. It needed clock speed to compete... 4.2-4.5GHz is more than fine for these purposes.

Hope we get some information on prices at CES, too!

 

[zinfamous]

If someone says 4.2-4.4GHz max and another says 5GHz, either one of the two is lying or the ES were highly inhomogeneous... I think that i didn't say nothing surprising... What other explaination is there?

you know....there's an actual word for that.

 

[The Stilt]

Ask the_stilt. Hé Will explain you how zen on 14lpp is limited to 2.8/3.2ghz turbo.
Best process analyst ever.

Shall we wait for the information about the clocks Ryzen is going to ship at, and the information about how much the power consumption increases when you push it beyond the stock speeds?
If above 4.0GHz is commonly achievable as 24/7 on all cores, at let's say over twice the original power consumpion I find the absolute overclock rather irrelevant regardless. Or do you disagree?

Also I haven't found any clock estimations you have made, 9kk prior to the launch of the product? Let alone one which would have potential being more accurate than mine? Hindsight etc