New Zen microarchitecture details

[The Stilt]

Out of interest do you have the VFMADD132PD 256bit latency/ TP numbers? same as CPC's?

cheers

Accurate for 128-bit but no idea what happens with 256-bit (i.e 2 ops), sorry.

 

[KTE]

How could you say it's 75%? Link? Or it's a guess? And 81W it's a theoretical limit. Probabily it's under 75W...

Exactly. Where's YOUR link for 81W max power draw?

The ammeter on what? If it's the wall it's obvious... There are the VRMs and PSU losses...

Ammeter is used to check the DC line current.

The other approximations have never quite been reliable enough to use, since the Phenom days at least. Usually always low balling the actual DC power.

I'm not sure why we have to keep rehashing these things with every new chip.

Sent from HTC 10
(Opinions are own)

 

[bjt2]

Exactly. Where's YOUR link for 81W max power draw?

Have you seen the New horizon video pre-conference of december 8th?
Difference between load and idle is 94W (moreless, i don't remember exactly).
Then the calculations are straightforward.
For IDLE powers, there are tons of reviews in which the power is measured at the 12V that show less than 5W in idle.
Do you want links? Do you want a link for efficiencies of VRM or PSUs? Do you want a drawing?

Google for VRM efficiency and PSU efficiency and you will find upper limits for that. I repeat: 88% for VRMs and 90% for PSUs. If you think that are false, post a link...

http://www.infineon.com/dgdl/pswus03vrmdesign.pdf?fileId=5546d462533600a4015357483c833fc2 Here there are also graph. 88% is the max efficiency and it's at 70 Ampere. Over 100A drops at 86% for a 6-phases and lower for less phases.

http://www.corsair.com/en-us/blog/2...at-does-it-mean-and-what-is-the-benefit-to-me This is a good 1200W PSU 80+ gold. At 144W (12%) the efficiency is 90%. Obviously you should take away the monitor and the system alone will draw under 144W...


Where came from your 75%? From your mind? Post a link or a reasoning, otherwise it's of no value.

Ammeter is used to check the DC line current.

The other approximations have never quite been reliable enough to use, since the Phenom days at least. Usually always low balling the actual DC power.

I'm not sure why we have to keep rehashing these things with every new chip.

Sent from HTC 10
(Opinions are own)

Ok, so the TDP was 125W and the measured power was 140W? I assume on the 12V.

140W*.88=123.2W.

Below 125W. And i was generous because over 70A the efficiency is lower than 88%...

 

[coercitiv]

Do you want a drawing?

Well, since you're offering, can you please post the exact same calculation for the 6900K in that test?

 

[bjt2]

Well, since you're offering, can you please post the exact same calculation for the 6900K in that test?

Moreless the same that Zen. But Zen is faster and anyway was overvolted...

 

[coercitiv]

Moreless the same that Zen.

Can you please be more accurate than that?

In case you feel you're missing some data, let me help you - idle to load delta for 6900K system was 85W.

Reference idle power usage for a 14nm 4c Intel core chip is:

• 9W @ 4Ghz w/o C states
• 3.5W @ 800Mhz w/o C states
• 1.8W with C states enabled

 

[Abwx]

Can you please be more accurate than that?

In case you feel you're missing some data, let me help you - idle to load delta for 6900K system was 85W.

Reference idle power usage for a 14nm 4c Intel core chip is:

• 9W @ 4Ghz w/o C states
• 3.5W @ 800Mhz w/o C states
• 1.8W with C states enabled

Idle power of the 6900K is 18W, i already posted the numbers but seems that whatever doesnt suit some public is simply ignored, hence the forever redundant and useless debate by here..

http://www.hardware.fr/articles/946-4/overclocking-consommation.html

Notice that someone said that Hardware.fr numbers are anectdotical, yet he once stated that such numbers would be meaningfull if measured at the 12V CPU rail, wich is exactly what HFR are doing, it s telling about the extent of the bad faith in this thread.

 

[CentroX]

Is there a reason that amd only has talked about their 8c 16t ryzen? Should we be worried about 4c 8t and 6c 12t?

 

[Agent-47]

I think they might not clock as high as Intel as they are salvaged 8c16t parts.

Is there a reason that amd only has talked about their 8c 16t ryzen? Should we be worried about 4c 8t and 6c 12t?

 

[CentroX]

I think they might not clock as high as Intel as they are salvaged 8c16t parts.

Exactly.

 

[dacostafilipe]

...

Congrats, you made it on pcgameshardware.de

http://www.pcgameshardware.de/Mainb...X370-B350-Vergleich-Intel-Chipsaetze-1218450/

 

[coercitiv]

Idle power of the 6900K is 18W, i already posted the numbers but seems that whatever doesnt suit some public is simply ignored, hence the forever redundant and useless debate by here..

Sometimes it really takes tons of patience to lead the horse to the water.

Take a look at the numbers I posted for reference power usage, then take a look at the HFR table you quoted. Ask yourself why is it that SB-E in that table only uses 6W, while HSW-E and BDW-E go to 16W and 21W respectively. What parameters other than semiconductor process characteristics greatly affect idle power usage? Look at my reference data again, multiply those figures by 2x and see how they land against the data from HFR.

Do we have any idea on how both CPUs were configured from a power management perspective during the presentation? Think about it.

 

[Jan Olšan]

Notice that someone said that Hardware.fr numbers are anectdotical, yet he once stated that such numbers would be meaningfull if measured at the 12V CPU rail, wich is exactly what HFR are doing, it s telling about the extent of the bad faith in this thread.

Don't put my words out of context. Sure, 12V measurements are way better than idle-load delta (my point exactly?). But the problem was that 1) even then they aren't necesarily the same for every CPU specimen and every board.
2) Yet you tried to extrapolate Zen power consumption from that - different platform, unknown motherboard consumption, unknown idle consumption...

There is no "ignoring insuitable facts", just pointing out that "we don't know well enough to say". YOu need to accept that sometimes there just isn't enough information and any guess will just stay a guess no matter how far you go in attempts to get some support points for it. It's like math, if you have too much unknowns and too little equations, you can't get a result.

I'm a fan of AMD but everyone needs to distinguish between what has some factual support in the confirmed information and what is just thinly-veiled speculation or mistaken extrapolation.

 

[Tuna-Fish]

Is there a reason that amd only has talked about their 8c 16t ryzen? Should we be worried about 4c 8t and 6c 12t?

The proper 4c8t part will be the APU, which will come later. I find talking mostly about the lead product that you are actually about to ship to be quite sensible.

 

[Atari2600]

I think they might not clock as high as Intel as they are salvaged 8c16t parts.

Why would a salvaged part not clock as high?

Salvage =/= Bin
Is there any correlation between a salvaged part and the typical bin of that part?

 

[CentroX]

The proper 4c8t part will be the APU, which will come later. I find talking mostly about the lead product that you are actually about to ship to be quite sensible.

How many skus have been confirmed? 4?

 

[coercitiv]

Why would a salvaged part not clock as high?

Salvage =/= Bin
Is there any correlation between a salvaged part and the typical bin of that part?

The unlocked Pentiums had no problems clocking high. The FX 6300 was streching it's legs as well.

 

[Abwx]

Congrats, you made it on pcgameshardware.de

http://www.pcgameshardware.de/Mainb...X370-B350-Vergleich-Intel-Chipsaetze-1218450/

They should correct the legend below the pic, though, since it s written that they are the source for the diagram...

AMD Ryzen: AMDs X370 und B350 im Vergleich zu Intels Chipsätzen

Quelle: PC Games Hardware

2) Yet you tried to extrapolate Zen power consumption from that - different platform, unknown motherboard consumption, unknown idle consumption...

I'm a fan of AMD but everyone needs to distinguish between what has some factual support in the confirmed information and what is just thinly-veiled speculation or mistaken extrapolation.

There s no mistaken extrapolation, it work the same with all plateforms, laws of physics are invariable and we have an accurate idea of what are the losses in a PSU and in a CPU power supply, once we use the power deltas influence of the rest of the plateform is neutered, and with idle power measuements of CPUs we can corner the estimations within +-5% at worst.

Look at my reference data again, multiply those figures by 2x and see how they land against the data from HFR.

Do we have any idea on how both CPUs were configured from a power management perspective during the presentation? Think about it.

You re using irrelevant data, mainstream CPUs are not triple channel HEDT, that s not even the same die at all, so it s not Hardware.fr who is in the wrong but your inadequate methodology.

FTR Computerbase measured exactly the same idle power as Hardware.fr for the BDW plateform in a similar configuration..

https://www.computerbase.de/2016-05/intel-core-i7-6950x-6800k-test/6/

You can do the comparison with the mainstream chips that you once deemed as being similar idle power wise..

 

[bjt2]

Can you please be more accurate than that?

In case you feel you're missing some data, let me help you - idle to load delta for 6900K system was 85W.

Reference idle power usage for a 14nm 4c Intel core chip is:

• 9W @ 4Ghz w/o C states
• 3.5W @ 800Mhz w/o C states
• 1.8W with C states enabled

I was talking of this video:
https://youtu.be/EsVNQYwlSAo?t=2m18s

 

[Veradun]

Take a look at the numbers I posted for reference power usage, then take a look at the HFR table you quoted. Ask yourself why is it that SB-E in that table only uses 6W, while HSW-E and BDW-E go to 16W and 21W respectively. What parameters other than semiconductor process characteristics greatly affect idle power usage?

PCI-E lanes, chipset and motherboard features?

 

[KTE]

Have you seen the New horizon video pre-conference of december 8th?
Difference between load and idle is 94W (moreless, i don't remember exactly).
Then the calculations are straightforward.
For IDLE powers, there are tons of reviews in which the power is measured at the 12V that show less than 5W in idle.
Do you want links? Do you want a link for efficiencies of VRM or PSUs? Do you want a drawing?

Google for VRM efficiency and PSU efficiency and you will find upper limits for that. I repeat: 88% for VRMs and 90% for PSUs. If you think that are false, post a link...

http://www.infineon.com/dgdl/pswus03vrmdesign.pdf?fileId=5546d462533600a4015357483c833fc2 Here there are also graph. 88% is the max efficiency and it's at 70 Ampere. Over 100A drops at 86% for a 6-phases and lower for less phases.

http://www.corsair.com/en-us/blog/2...at-does-it-mean-and-what-is-the-benefit-to-me This is a good 1200W PSU 80+ gold. At 144W (12%) the efficiency is 90%. Obviously you should take away the monitor and the system alone will draw under 144W...


Where came from your 75%? From your mind? Post a link or a reasoning, otherwise it's of no value.



Ok, so the TDP was 125W and the measured power was 140W? I assume on the 12V.

140W*.88=123.2W.

Below 125W. And i was generous because over 70A the efficiency is lower than 88%...

I tested SMPS ATE units... I know all this. I'm not talking about any of it.

I'm telling you to check previous CPU data and compare your methodology of actual power draw compared to how accurate it is for them.

Hint: very inaccurate.

Q9650 used to show as 60W full load in P95 with your 'deductive' methodology a la ~50W with Blender.

That Blender scene is not full load or max power, why does this have to be spelt out to you? It is lower than average load.

Sent from HTC 10
(Opinions are own)

 

[Abwx]

I'm telling you to check previous CPU data and compare your methodology of actual power draw compared to how accurate it is for them.

You re telling to use irrelevant datas of CPUs that have no AVX, FMA and such instructions that increase power drain under Prime 95, not counting SMT that also increase the loading capacity..

Q9650 used to show as 60W full load in P95 with your 'deductive' methodology a la ~50W with Blender.

That Blender scene is not full load or max power, why does this have to be spelt out to you? It is lower than average load.

Flawed methodology, one more time, you cant compare CPUs that have no SMT with SMT enabled ones, those latter can max out the otherwise low throughput inherent to Blender and hence have much higher loading than the former..

Anyway look like you made a specialty of using exemples that are in no way relevant, let alone half accurate.

 

[Rifter]

It is evident that KL i7s will be the CPU of choice for those who want absolute best ST performance because IPC+ better OC could result in up to ~20% better one thread performance. Ryzen will definitely be a right choice for almost anyone else due to superior price/perf. ratio.

Without knowing how Zen OC's or even its exact IPC its to early to make this assumption. Although I also suspect that the intel chips will OC better.

 

[Agent-47]

Interestingly Canard PC measured SR's VFMADD132PD throughput to be 2x BR's throughput, or the same as KL (2/cycle).
SQRTPD gets executed on just 1 unit, DIVPD, too, while VDIVPD has a bit better throughput.

Edit:
I just found out about two new articles by Hiroshige Goto (auto translated to Googlish):
The integer unit of AMD's next generation CPU "ZEN" is completely different from the Bulldozer series
Floating point / SIMD unit of AMD next generation CPU "ZEN" of relatively mature design

This very interesting for sure. But is there a way to deduce the ST IPC from the numbers if we assume that the front end if not a bottleneck? I can see that the latencies are the same more or less, but I am not sure if the TP is multithreaded or ST.

 

[coercitiv]

You re using irrelevant data, mainstream CPUs are not triple channel HEDT, that s not even the same die at all, so it s not Hardware.fr who is in the wrong but your inadequate methodology.

I never questioned the validity of HFR data, you seem hell bent on the idea that I want to discard the 18W power figure for BDW-E.

The same HFR table has Sandy Bridge-E and Ivy Bridge-E at 6W and 3.6W idle power. I ask you how can that be explained and your only answer is I'm using "irrelevant data" and "inadequate methodology"? Is this a joke? I'm using your data!

I was talking of this video:
https://youtu.be/EsVNQYwlSAo?t=2m18s

And I was talking about the same video.

The systems in that video were configured for optimal performance. Optimal performance likely entails disabling sleep states, maybe even keeping CPUs at base speed even when idle. This strongly affects idle power usage and skews power usage estimates based on power delta.