Why is AMD's Zen more efficient than intel's Core?

P

Peppercorn

Limp Gawd
Joined
Dec 8, 2016
Messages
259
Is it the architecture or 14nm process or both? It's stunning that somehow AMD has managed to outdo intel in efficiency when intel was supposed to have such a superior 14nm process. I think a lot of it is from GloFo's 14nm process and i think it'll be even more impressive in mobile. But, all the power saving techniques that AMD put into Carrizo and Bristol Ridge was invaluable reusable IP that surely played a big part in Zen's excellent efficiency.
 
TheLAWNoob said:
We get it, Zen is good.
Click to expand...

Yes, and your point? Getting tired of hearing how good Zen is? Are you being sarcastic? Or am I reading into something that is not being said?

Well, efficiency is one thing they definitely needed to get down and it looks like they nailed it! :)
 
Peppercorn said:
Is it the architecture or 14nm process or both? It's stunning that somehow AMD has managed to outdo intel in efficiency when intel was supposed to have such a superior 14nm process. I think a lot of it is from GloFo's 14nm process and i think it'll be even more impressive in mobile. But, all the power saving techniques that AMD put into Carrizo and Bristol Ridge was invaluable reusable IP that surely played a big part in Zen's excellent efficiency.
Click to expand...

a lot of it comes down to the fact that it doesn't have an IGP built into the processor.. yes even though the igp may be disabled there's still power going to it. some of it has to do with the lower clocks as well but yeah i do think AMD put a lot of effort into the efficiency side of the processor. i think you're exaggerating the scaling potential of the processor though given on mobile platforms it'll be based on the APU which while efficient adds to the power usage of the chip. vega and the rx500's going to be the key to whether or not it becomes an efficient mobile platform.
 
ManofGod said:
Yes, and your point? Getting tired of hearing how good Zen is? Are you being sarcastic? Or am I reading into something that is not being said?

Well, efficiency is one thing they definitely needed to get down and it looks like they nailed it! :)
Click to expand...

Why do you think everyone is against AMD?

Zen was better than I expected, I bought it and moved on.
 
I haven't seen a super in depth look at it (given that's a far more interesting topic for the server, not desktop, parts) but it's certainly fairing pretty well from the few sites I've seen measure usage. It's generally a bit lower than the 6900k depending on load, which is pretty good given it more or less matches the 6900k in non-gaming workloads. But yeah, a far more in depth look at performance compared to watts in a much larger variety of workloads is needed to say much else.

http://www.techspot.com/review/1345-amd-ryzen-7-1800x-1700x/page6.html
http://www.tweaktown.com/reviews/8081/amd-ryzen-7-1700-1700x-cpu-review/index11.html
 
Arcygenical said:
Also, don't underestimate the efficiency gains from having no IGP on die...
Click to expand...

I doubt that has much to do with it actually.

With regards to the 6900k, keep in mind the 6900k has quad channel memory, 4 RAM sticks to run, and a larger PCI-E controller because more lanes.

In general though, it is most likely due to the manufacturing process. The manufacturing process is low power optimized, and that's why Zen is incapable of clocking above 4 ghz. It's a tradeoff. You can bias the manufacturing process towards high frequencies, and you generally get higher watts/performance as a result. Or you can bias it towards power efficiency, but you cannot clock high.
 
Peppercorn said:
Is it the architecture or 14nm process or both? It's stunning that somehow AMD has managed to outdo intel in efficiency when intel was supposed to have such a superior 14nm process. I think a lot of it is from GloFo's 14nm process and i think it'll be even more impressive in mobile. But, all the power saving techniques that AMD put into Carrizo and Bristol Ridge was invaluable reusable IP that surely played a big part in Zen's excellent efficiency.
Click to expand...

They are not. 65W(That is actually able to stay below 65W without being a blatant lie) quadcore with HT at 3.2Ghz base, 3.4Ghz boost with low IPC and no IGP. But who knows if this will still be a 65W part after the SIMD fix that requires the power consumption to be increased.

Now compare to these much faster 14nm 65W parts:
5775C at 3.3Ghz, 3.7Ghz, 128MB EDRAM and GT4 IGP.
6700 at 3.4Ghz with 4.0Ghz turbo with GT2 IGP.
7700 at 3.6Ghz with 4.2Ghz turbo with GT2 IGP.

Or at 35W.
7700T at 2.9Ghz with 3.8Ghz turbo with GT2 IGP.

Or even 22nm.
4770S at 3.1Ghz with 3.9Ghz turbo and GT2 IGP.
3770S at 3.1Ghz with 3.9Ghz turbo and GT2 IGP.

Zen efficiency wise is more around IB levels at 22nm. But even then I doubt it can keep up.

You should also ask around with Dell, Lenovo etc why they dont pick Naples or Ryzen for workstations and servers.
 
Last edited:
Shintai said:
They are not. 65W(That is actually able to stay below 65W without being a blatant lie) quadcore with HT at 3.2Ghz base, 3.4Ghz boost with low IPC and no IGP. But who knows if this will still be a 65W part after the SIMD fix that requires the power consumption to be increased.

Now compare to these much faster 14nm 65W parts:
5775C at 3.3Ghz, 3.7Ghz, 128MB EDRAM and GT4 IGP.
6700 at 3.4Ghz with 4.0Ghz turbo with GT2 IGP.
7700 at 3.6Ghz with 4.2Ghz turbo with GT2 IGP.

Or at 35W.
7700T at 2.9Ghz with 3.8Ghz turbo with GT2 IGP.

Or even 22nm.
4770S at 3.1Ghz with 3.9Ghz turbo and GT2 IGP.
3770S at 3.1Ghz with 3.9Ghz turbo and GT2 IGP.

Zen efficiency wise is more around IB levels at 22nm. But even then I doubt it can keep up.

You should also ask around with Dell, Lenovo etc why they dont pick Naples or Ryzen for workstations and servers.
Click to expand...
You realize you said nothing here and just quoted part numbers? And as far as work station or servers... Intel ahs paid for it before in anumber of ways even just 2 years ago, in other circumventing ways. You sure it is a question you really want to ask.
 
JustReason said:
You realize you said nothing here and just quoted part numbers? And as far as work station or servers... Intel ahs paid for it before in anumber of ways even just 2 years ago, in other circumventing ways. You sure it is a question you really want to ask.
Click to expand...

So now Intel is paying server and workstation OEMs and buyers. Didn´t take long before that excuse came up.
 
As far as power efficiency, as far as I understand it has alot to do with Samsungs 14nm process, but that also limits overclocking to ~4ghz. I think the Ryzen 5 processors will also have the 4ghz limit as well. Maybe in future releases they might be able to tweak the process but who knows.
 
lolfail9001 said:
Is it?
Click to expand...

NO. The old 6900k is 15% more efficient.

getgraphimg.php
 
Here's another one.

power_gaming.png


As for the IGP in SkabyLake i don't think that has any impact as a properly designed chip would have that shut down when using a dGPU. Ryzen just looks like an overall more efficient design. GloFo's 14nm process' sweetspot is around 3-3.5GHz which is extremely promising for mobile and server.
 
Peppercorn said:
Here's another one.

power_gaming.png


As for the IGP in SkabyLake i don't think that has any impact as a properly designed chip would have that shut down when using a dGPU. Ryzen just looks like an overall more efficient design. GloFo's 14nm process' sweetspot is around 3-3.5GHz which is extremely promising for mobile and server.
Click to expand...

That is power consumption, not efficiency... but we already knew (check #17) that 1800X is more efficient than 7700k: 16.78 vs 12.67.

Too bad that Techpowerup is comparing apples and oranges and then making outrage claims, because the 6900k (which they didn't test) is more efficient than the 1800X.
 
juanrga said:
NO. The old 6900k is 15% more efficient.
Click to expand...

What workload(s) is this tested under?

Edit: Found the article. x264 encoding if anybody else was curious.

Edit Again: Interestingly, the R7 1700 is by far the most efficient processor in that test that they had. It's sort of felt like all along the 1800x (and even 1700x) were "factory overclocked" models, given their almost non-existent headroom. Power usage seems to reflect that.

getgraphimg.php
 
Last edited:
rhexis said:
not getting why people are hating on the amd stuff. competition is a good thing or do you enjoy paying out the nose for minor upgrades every year from intel?
Click to expand...

It is only certain folks around these parts and that will probably never change. Me, I have a 1700 and 1700X and enjoy them immensely.
 
FrozenSun said:
What workload(s) is this tested under?

Edit: Found the article. x264 encoding if anybody else was curious.

Edit Again: Interestingly, the R7 1700 is by far the most efficient processor in that test that they had. It's sort of felt like all along the 1800x (and even 1700x) were "factory overclocked" models, given their almost non-existent headroom. Power usage seems to reflect that.

getgraphimg.php
Click to expand...

There is not anything special about the 1700, just a consequence of lower factory clocks. Performance depends linearly on frequency f, whereas power has a non-linear dependence usually quadratic n=2 or cubic n=3

Performance ~ f

Power ~ f^n

Efficiency = Performance / Power = 1/f or 1/f^2

Reducing clocks increases efficiency. That is why efficiency increases as 1700 > 1700X > 1800X.
 
Last edited:
juanrga said:
There is not anything special about the 1700, just a consequence of lower factory clocks. Performance depends linearly on frequency f, whereas power has a non-linear dependence usually quadratic n=2 or cubic n=3

Performance ~ f

Performance ~ f^n

Efficiency = Performance / Power = 1/f or 1/f^2

Reducing clocks increases efficiency. That is why efficiency increases as 1700 > 1700X > 1800X.
Click to expand...

I'm well aware. The rate efficiency deteriorates also accelerates the closer you get to pushing things to the max. I expected the 1700 to be the most efficient of the R7s. I didn't expect it to be that much more efficient, hence why I feel like they probably pushed the 1700x and 1800x a bit harder than they really wanted to, but felt they needed to to compete. And I do think it's incredibly impressive AMD produced a chip whose total performance is good and its efficiency is excellent. FX were turds in both departments.

I also thought it was worth noting that a Zen chip was more efficient than the 6900k. Of course, the 6900k did slightly outperform the 1700 at that bench, so we'd have to see how a lower clocked broadwell did to make it a fair comparison.

At any rate, this is a server argument more than anything, which I don't care a lot about.
 
Last edited:
rhexis said:
not getting why people are hating on the amd stuff. competition is a good thing or do you enjoy paying out the nose for minor upgrades every year from intel?
Click to expand...
AMD is competing? Looking at Ryzens i do not see anything competing with Intel per se. Just a chip that covers a very specific niche in Intel's existing desktop line-up.
 
Results are very early and the current AMD line up is limited. It does appear as you near 4 ghz the efficiency does decline quickly though. You also need to consider this is their first go around with this architecture, Intel has been tweaking theirs for a decade now? Time will tell.
 
FrozenSun said:
I'm well aware. The rate efficiency deteriorates also accelerates the closer you get to pushing things to the max. I expected the 1700 to be the most efficient of the R7s. I didn't expect it to be that much more efficient, hence why I feel like they probably pushed the 1700x and 1800x a bit harder than they really wanted to, but felt they needed to to compete. And I do think it's incredibly impressive AMD produced a chip whose total performance is good and its efficiency is excellent. FX were turds in both departments.
Click to expand...

Again there is not anything special on why RyZen is much more efficient than Piledriver FX. Piledriver was a speed-demon microarchitecture; Zen is a brainiac microarchitecure; speed-demons will always have worse efficiency per physical laws. Moreover, RyZen uses a 14nm optimized for low power whereas Piledriver uses a 32nm optimized for performance. It would be an interesting academic exercise to port Piledriver to 14LPP and see how much its efficiency would increase. A fast computation suggests efficiency would probably double.
 
The benchmarks I've seen point to Ryzen 1800X being more efficient under idle and light loads than the 7700k or equivalent Intel 6 and 8 core chips... but less efficient than the 7700k under full load, and approximately about the same as the Intel 6 and 8 core Broadwells.

So it loses a little overall complared to the 7700k, and is overall a little bit more efficient than the Intel 8 cores. Buuuuuuut put this in proper context: the 9590 chip was a f*cking 220 watt part. AMD has come a long way.

If you're looking for a low power 8 core, though, the 1700 vanilla chip is probably a good bet.
 
juanrga said:
Again there is not anything special on why RyZen is much more efficient than Piledriver FX. Piledriver was a speed-demon microarchitecture; Zen is a brainiac microarchitecure; speed-demons will always have worse efficiency per physical laws. Moreover, RyZen uses a 14nm optimized for low power whereas Piledriver uses a 32nm optimized for performance. It would be an interesting academic exercise to port Piledriver to 14LPP and see how much its efficiency would increase. A fast computation suggests efficiency would probably double.
Click to expand...

I'm not sure what you're arguing about at this point. I never said it was "special". I'm just fairly impressed that AMD, who seemed to be totally out of the game, released a chip which more or less trades blows with the 6900k in both total performance and efficiency in many workloads. Yes, it has some hiccups, hence the lower price point, but I didn't think AMD would even get that close. I figured their 8 cores were going to be 6850k level at best in threaded apps.
 
From what I've been reading around the web, here is what I've picked up:

The Samsung/GloFo process is a mobile-focused process, which as such is meant for low power-high efficiency. As a result that also explains the clock ceiling, though they did get a fair bit out of it...
Based on TheStilt's test results it seems Ryzen was aimed to be more of a 3GHz part to me, but maybe that's in part due to the process?
Ryzen's Stock voltage isn't necessarily what it needs to run at its speed, due to binning being based on voltage not clocks, and a wafer gets a blanketed programmed voltage rating based on the worst performing die.
It's when you clock Ryzen down, that things start to get really interesting and telling in regards to efficiency and power draw.

The bulk of all that can be discerned by reading this: Ryzen: Strictly technical
Which note I hate myself for having slacked off after around page 10, since there's 27 more pages now! lol
 
FrozenSun said:
I'm not sure what you're arguing about at this point. I never said it was "special". I'm just fairly impressed that AMD, who seemed to be totally out of the game, released a chip which more or less trades blows with the 6900k in both total performance and efficiency in many workloads. Yes, it has some hiccups, hence the lower price point, but I didn't think AMD would even get that close. I figured their 8 cores were going to be 6850k level at best in threaded apps.
Click to expand...

I am saying that there is not anything really impressive in the efficiency of Ryzen compared to Piledriver: a Brainiac design plus 14LPP did most of the work.

About the rest it depends of the expectations of each one. For instance I said "IPC+SMT ~ Haswell" and it is ~15% behind Haswell; I cannot be really impressed.
 
juanrga said:
I am saying that there is not anything really impressive in the efficiency of Ryzen compared to Piledriver: a Brainiac design plus 14LPP did most of the work.

About the rest it depends of the expectations of each one. For instance I said "IPC+SMT ~ Haswell" and it is ~15% behind Haswell; I cannot be really impressed.
Click to expand...
You are dodging the question. He wasn't asking your assessment on Piledriver. He is asking your assessment of Ryzen against Intel being both on 14nm and AMD isn't usually known for efficiency and add to that a new architecture and new node with what should be a subpar foundry compared to Intel. That is the question- ANSWER IT.

Now I am not certain where it is you get your information but AMDs SMT shows ~45% and Intels HT ~35%. You don't compare SMT/HT designs like you did with Haswell unless HT % changes with architecture but even then you did it wrong. And add to that the only place AMD results may even give that result is in gaming, whereas everywhere else ie: productivity they surpassed most all Intel designs up to current quite handily.
 
juanrga said:
About the rest it depends of the expectations of each one. For instance I said "IPC+SMT ~ Haswell" and it is ~15% behind Haswell; I cannot be really impressed.
Click to expand...
15% behind Haswell? Where? Let alone with SMT. From my perspective it has almost the Skylake IPC and ~10% better SMT, if i compare it to my i5 and some random 6900k from GB4 data base.
 
JustReason said:
You are dodging the question. He wasn't asking your assessment on Piledriver. He is asking your assessment of Ryzen against Intel being both on 14nm and AMD isn't usually known for efficiency and add to that a new architecture and new node with what should be a subpar foundry compared to Intel. That is the question- ANSWER IT.
Click to expand...

I can guess just by how the conversation is going that you're talking to Juan??? or whatever his name is.

He will dance and prance and try to convince you that this really isn't the question you should ask, that you're wrong for asking it and that he can adjust your world view to accept his question and answer. (which is always correct). So , good luck.

AMD is making do with less transistors and Ryzen doesn't have an IGP. When we see the APU's we'll have better balance to compare like to like. That's what I'm thinking. Still, I don't think they're out in left field here, I think AMD made huge inroads. It's like they hid the incremental performance they should have been bringing for years and dumped it in our laps all at once.

You ever look at your female child in my little pony clothes and then all of the sudden some fuck boy is picking her up for a date with his shitbox Acura thumping away out front of your house? This is Ryzen.
 
Simplyfun said:
You ever look at your female child in my little pony clothes and then all of the sudden some fuck boy is picking her up for a date with his shitbox Acura thumping away out front of your house? This is Ryzen.
Click to expand...

LMFAO!
 
Here we go again. Fucking another thread that is about to turn in to shit fest. Someone asks a simple questions and we are talking about fuckin Piledriver. WTF?

Ryzen is a decent chip. It is pretty effecient, it may or may not beat intel. But regardless! Are we really that ignorant that we can't admit this is a decent chip knowing how much more man power and R&D budget intel has over AMD?

But there are some people that go out of their way to disprove that Ryzen is somehow same pile of shit Piledriver was.

Get real. Move on and no point making threads anymore or asking a question because you are feeding the trolls it seems. You know who you are.
 
JustReason said:
You are dodging the question. He wasn't asking your assessment on Piledriver. He is asking your assessment of Ryzen against Intel being both on 14nm and AMD isn't usually known for efficiency and add to that a new architecture and new node with what should be a subpar foundry compared to Intel. That is the question- ANSWER IT.
Click to expand...

Huh? He wrote "I do think it's incredibly impressive AMD produced a chip whose total performance is good and its efficiency is excellent. FX were turds in both departments." And I just explained in a reply to him why getting RyZen providing efficiency very superior to Piledriver doesn't have any special merit, because Piledriver was a speed-demon microarchitecture on a non-efficient 32PDSOI node. I even suggested that only porting Piledriver to 14LPP would probably double the efficiency of the chip.

About RyZen getting efficiency close to Broadwell. Again there is no magic or mystery here. RyZen uses a process node more optimized for power consumption than for performance, The RyZen microarchitecture is more distributed [1] and optimized for thoughput workloads (that is why RyZen shines in Handbrake or Cinebench); again throughput-optimized microarchitectures are inherently more efficient than latency-optimized ones [2]. Finally, Broadwell has power hungry 256bit SIMD units and datapaths, which add a bit to the overall inefficiency in workloads cannot use 256bit vectors.

[1] Separate INT and FP clusters, distributed schedulers, dual CCX...

[2] 6-wide vs 8-wide, high-latency IF,...
 
Dejavu? I could swear I've replied to this thread before, but it's new and I haven't. Whatever, probably for the best. :/
 
You must log in or register to reply here.
Back
Top