Intel Skylake / Kaby Lake

[TahoeDust]

He gets all his technical information from youtube videos, so obviously there will be some mistakes here and there. Don't hold it against him, some people just aren't big readers.

The graph was from this article...

https://www.overclock3d.net/reviews/cpu_mainboard/intel_core_i9-7980xe_18_core_hedt_cpu_review/1

 

[Zucker2k]

Just taking the first one, a 20% lead is a fair amount, but decimates is not the word I would use. Now twice the price for 20% performance increase ? If you have the money, sure, but its a poor choice IMO.

And that 20% is only in some things. 1950X actually wins a few, and comes close in quite a few.

So, say your Threadripper 1950x build sums up to $5,000. Your i9 7980XE build will be at $6,000 - all other parts and accessories being equal. That's your 20% right there.

 

[eddman]

Not an insult (unless saying he is 'wrong' is an insult to you). In fact, I told the guy not to hold it against him for posting fake graphs he got from youtube.

If you'd like to complain about somebody "insulting" others, maybe you should check out his last post...

I should have written "disrespectful" and yes, the whole tone of the post was that. You could have checked where the source was from.

As for that post you mentioned, I actually reported it.

 

[Jan Olšan]

It wasn't long ago a Hardware Unboxed review was considered "trash data" because it didn't line up to the silicon world view of some forum members, now Overclock3D - a review site more than 10 years old - is being trashed for publishing power data that doesn't fit right with other folks.

Maybe, just maybe, we should consider sample variance and use data from more reviews to get a more accurate image?! Oh, who am I kidding, Tiny 'Tim' Logan is a YouTube amateur! /s

You know, the person we talk about didn't even post the link to the source. Do you think it is weird we doubt the obviously fishy-looking result in those circumstances? We don't know how credible reviewer it is, what were the settings, components, etc. I will trust the PCPer data much more eagerly. If there is some good explanation for the 700W nonsense, I'm all ears, but in the meantime, I'm going to ignore it as some fringe outflier. I might be partial and fanboyish, but I'm not gonna "+1 like" obvious bullshit just because I wish it was true.

-------------------------------------------------------------
On another topic... (source)

Throttling to meet the TDP? Hardware.fr has one of the better power consumption results, so it would not be surprising. The regression is though - seems that when running x265, the clocks drop so much that the IPC boost from AVX2 assembly is more than negated and performance suffers.

 

[Markfw]

What the hell did the guy do with the Threadripper, 1.5V? Sounds kinda bogus (read = not representative) to me.
Look at what PC Perspective measured (edit: link):

The i9-7980XE is at 4,3 GHz, 1.2 V.



Even though Threadripper also shoots up when OCing on raised voltage, Skylake-X seems to have much bigger "potential" in this. And note that the CB R15 test doesn't even use AVX(2) or even AVX512

I agree. I can do 4.0 at 1.2 vcore.

 

[wildhorse2k]

If you are talking about the 7980xe, OC3D is running it at 4.5GHz all 18 cores easily using an off the shelf AIO...

Thats probably the highest overclock that makes sense on the 7980XE, as above that this monster will require a delid and draw too much power anyway. He managed to get very low voltages, its probably why it seemed so power efficient. Its still interesting that even with the standard Intel TIM it can run so fast on all cores without delid. Power efficiency seems to be surprisingly good without OC.

His TR 1950 probably draws so much power because its an ES.

 

[LTC8K6]

Once the initial purchase has been made, how well a CPU like this pays for itself has almost no bearing on the actual performance offered, except in very specific instances, and those instances aren't representative of the kind of users buying these CPUs off retail and etail.

If you want the best performance in something like Ansys Fluent, you'd ask your company to invite tenders for one of these instead.

I don' t know why you keep agreeing with me while sounding like you don't.

 

[R0H1T]

Why some members like to insult others for no reason. It's from overclock3d.

Doesn't say under what program it was measured but the footnote says CPU+GPU. A game? That can't be it, can it?

The numbers are odd to say the least.

I should have written "disrespectful" and yes, the whole tone of the post was that. You could have checked where the source was from.

As for that post you mentioned, I actually reported it.

That's a serious anomaly, if truly that.

Having 18 cores on tap means that increasing voltage causes power consumption to quickly spiral out of hand. After some trial and error we found that the chip became incredibly thirsty if voltage was increased to 1.2V. Appreciating that we're using air cooling we took a balanced approach to frequency and voltage. All told, we were able to run the chip at an all-core 4.5GHz using 1.17V in the BIOS.
This combination was at the very edge of what the Noctua cooler could handle without causing the chip to throttle under demanding workloads. To give you an idea of the at-wall system wattage, the default configuration consumed 233W when running Cinebench. The overclocked chip pulled an average 488W on the same test, suggesting that premium power delivery will make a genuine difference in your overclocking ability.
We've been in touch with leading system builders who have seen the same phenomenon. The Asus X299-A Prime is decent enough, of course, yet they say that switching over to the Deluxe offers more stability at the bleeding edge.

http://hexus.net/tech/reviews/cpu/110072-intel-core-i9-7980xe-14nm-skylake-x/?page=13

 

[TahoeDust]

You know, the person we talk about didn't even post the link to the source. Do you think it is weird we doubt the obviously fishy-looking result in those circumstances? We don't know how credible reviewer it is, what were the settings, components, etc. I will trust the PCPer data much more eagerly. If there is some good explanation for the 700W nonsense, I'm all ears, but in the meantime, I'm going to ignore it as some fringe outflier. I might be partial and fanboyish, but I'm not gonna "+1 like" obvious bullshit just because I wish it was true.

-------------------------------------------------------------
On another topic...

Throttling to meet the TDP? Hardware.fr has one of the better power consumption results, so it would not be surprising. The regression is though - seems that when running x265, the clocks drop so much that the IPC boost from AVX2 assembly is more than negated and performance suffers.

Here is the source for the graph I posted. OC3D is not new to this game...

https://www.overclock3d.net/reviews/cpu_mainboard/intel_core_i9-7980xe_18_core_hedt_cpu_review/14

I can't believe that we are really down to complaining about the power consumption of an overclocked 18 core cpu.

 

[LTC8K6]

Perhaps there is just more variability in all of these modern 14nm chips?
Perhaps if you test 5 of them, you will get 5 different power consumption results?

 

[Zucker2k]

I'll

Perhaps there is just more variability in all of these modern 14nm chips?
Perhaps if you test 5 of them, you will get 5 different power consumption results?

look at board and bios versions first.

 

[R0H1T]

Perhaps there is just more variability in all of these modern 14nm chips?
Perhaps if you test 5 of them, you will get 5 different power consumption results?

That is to be expected, since we're talking total system power consumption in most of the reviews.
However when there's a delta of 180W in just one review, there has to be something odd, perhaps he ought to redo the TR OCed test & recheck the total (system) power consumption numbers.
Also 1.5v is insane for TR, or Skylake-X for that matter.

 

[TahoeDust]

Real question. Did TTL ever say he was using 1.5v on that 4.0GHz TR overclock?...or did that just start by someone saying it in this thread?

 

[Markfw]

The bottom line on power consumption, IMO. Anandtechs review and methodology appear to be the best, and the majority of OTHER reviews seem to support that. Since I OWN a TR overclocked to 4.0, I KNOW what it takes, and I think Anands review is the most representative.

 

[R0H1T]

Real question. Did TTL ever say he was using 1.5v on that 4.0GHz TR overclock?...or did that just start by someone saying it in this thread?

Don't know but he's using previous ES results for TR.

Goto ~ https://youtu.be/shLd_JPnPLs?t=250

 

[TahoeDust]

The bottom line on power consumption, IMO. Anandtechs review and methodology appear to be the best, and the majority of OTHER reviews seem to support that. Since I OWN a TR overclocked to 4.0, I KNOW what it takes, and I think Anands review is the most representative.

OK. Using Anandtech's number the 7980xe takes ~15w more than the 1950x, stock vs stock, is that correct? I think the 15w are being put to good use.

I mean 8% more power for 12.5% more cores. Kind of logical.

 

[TahoeDust]

Don't know but he's using previous ES results for TR.

Goto ~ https://youtu.be/shLd_JPnPLs?t=250

Yep. Because he can only get 3.8GHz out of his retail TR 1950x. He said he wanted to put the best results he had for the 1950x.

 

[Bouowmx]

From OC3D TV: Threadripper 1950X & ROG Zenith Extreme X399 Motherboard Review

I did manage to get a 4 GHz overclock across all of the cores at 1.42 V

To settle doubts on Overclock 3D's result on AMD Ryzen TR.

 

[Zucker2k]

From OC3D TV: Threadripper 1950X & ROG Zenith Extreme X399 Motherboard Review

To settle doubts on Overclock 3D's result on AMD Ryzen TR.

Some people are pumping 1.4v into their oc tr rigs in the tr builders thread. Also, a leaky ES chip could easily consume 1.5v @ 4ghz

 

[TahoeDust]

From OC3D TV: Threadripper 1950X & ROG Zenith Extreme X399 Motherboard Review

To settle doubts on Overclock 3D's result on AMD Ryzen TR.

So, the 1.5v was just pulled out of thin air?

 

[Bouowmx]

OK. Using Anandtech's number the 7980xe takes ~15w more than the 1950x, stock vs stock, is that correct? I think the 15w are being put to good use.

At stock, AMD Ryzen TR and Intel Skylake-X power figures are not directly comparable. AMD Ryzen TR on ASUS ZENITH EXTREME is power-limited to 180 W, whereas Intel Skylake-X on practically any motherboard has no such limit. I explained earlier: 39089929

 

[Markfw]

OK. Using Anandtech's number the 7980xe takes ~15w more than the 1950x, stock vs stock, is that correct? I think the 15w are being put to good use.

I mean 8% more power for 12.5% more cores. Kind of logical.

I even said that in a post above.

 

[R0H1T]

Yep. Because he can only get 3.8GHz out of his retail TR 1950x. He said he wanted to put the best results he had for the 1950x.

Strangely there's no power consumption numbers in the retail sample review, why would you put ES numbers then for comparison? There's any number of things that could go wrong with that theory.
https://www.overclock3d.net/reviews/cpu_mainboard/amd_ryzen_1950x_threadripper_retail_retest/1

 

[eddman]

The bottom line on power consumption, IMO. Anandtechs review and methodology appear to be the best, and the majority of OTHER reviews seem to support that. Since I OWN a TR overclocked to 4.0, I KNOW what it takes, and I think Anands review is the most representative.

I can't seem to find out which program they used for power consumption. The efficiency part has graphs for corona and handbrake but no info on the top, consumption charts.

Techspot uses corona and lists both 7960 and 7980 using more power. Is anandtech using handbrake then?

Also, why does corona cause skylake-X to consume more power compared to other tests? Is it AVX?

 

[TahoeDust]

I even said that in a post above.

Sorry, I missed it. The thread is moving fast.