Intel's Future Chips: News, Rumours & Reviews

Page 120 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

aldaia

Distinguished
Oct 22, 2010
533
18
18,995


Wait!!! Does that mean that the non B 8700 actually has a real TDP higher than 65W ?
But that cannot be!!! We all know that intel stays true to anounced TDP. Only AMD has a TDP higher than anounced.
;)
 


I know, right?

It's really, really, really, surprising! (not)

Cheers! /sigh
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995
Fake rumor, but curious anyway (be careful several sites have reported as if the news where true).
Acording to the rumor intel is preparing a 40 aniversary edition of the 8700K with much higher frequencies to celebrate 40 years of 8086.
Launch Date: June 8, 2018
SKU: "Intel Core i7-8086K 40th Anniversary Edition"
Specs: 8th Gen (CFL), 6C/12T, 5 GHz Turbo, unlocked
https://www.pcgamesn.com/intel-core-i7-8086-anniversary-edition



 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


AMD chips violates the official TDP rating on stock settings and AMD even had to admit the real TDPs: e.g. 128W for the R7-1800X and 90W for the R7-1700.

Intel chips satisfy the official TDP on stock settings. The official TDP when overclocks are enabled at BIOS level. E.g. the 8700k has an all-core turbo of 4.5GHz on stock settings, but it is increased to 4.7GHz with auto-overclockling enabled in boards that support it. With auto-overclocking enabled on BIOS the TDP of the CPU is no longer 95W, obviously, because the official TDP rating is for stock settings/clocks.

Different mobos implement different auto-overclock features.
 


I'm pretty sure Intel breaks the TDP with Turboboost, at least on the non-k SKUs, this is without MCE or an auto overclock enabled. The 8700 has an all core Turbo of 4.3GHz, but it has been shown if you limit it to the 65 Watt TDP with Intel XTU (this limit set by default on certain low end boards) and run it with an aftermarket cooler to eliminate thermal throttling, then you get 3.3-3.5GHz, or closer to the advertised base clock. Bottom line is the 8700 non k probably should have a 95 Watt TDP at minimum, but Intel is trying to say its only 65 Watts because they only guarantee the 3.2GHz base clock and TurboBoost is basically an auto-overclock that only works if you have a good motherboard and aftermarket cooling.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995


Intel disagrees with you. Intel defines TDP at base frequency, and when Thermal dissipation is constrained to official TDP the processor can only work at base frequency. 3.2 GHZ in the case of 8700 as shown by the evaluation of coffee lake B CPUs.

See Intel definitions of Base frequency and TDP:

Processor Base Frequency
Processor Base Frequency describes the rate at which the processor's transistors open and close. The processor base frequency is the operating point where TDP is defined. Frequency is measured in gigahertz (GHz), or billion cycles per second.

TDP
Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


There is no disagreement.

Of course TDP is defined for base frequency, not for all-core turbo. Turbo is a special working stage, not the default state. If TDP was defined for all-core turbo, then the chip could sustain turbo clocks all the time and they wouldn't be turbo anymore, but a new base frequency.

But it doesn't mean that TDP is violated by turbo, because TDP is average, not instantaneous consumption. As explained a hundred of times TDP is "sustained power consumption"

https://books.google.es/books?id=cM8mDwAAQBAJ&lpg=PA24&ots=SHX5Km2Vvb&dq=TDP%20average%20power%20consumption%20Patterson&pg=PA24#v=onepage&q&f=false

During boost stages instantaneous power consumption is above the average defined by the TDP because there is a deficit of power previous to boost, where the instantaneous power consumption was below the average

sandybridge_061.jpg


If you integrate over time, the area of the curve above the TDP is compensated by the area below the curve, and the chip is within the TDP during all the workload.

So Intel chips satisfy the official TDP as all reviewers know. AMD is different. As reviewers have demonstrated, AMD lies about TDPs. The '95W' chips are 128W chips and the '65W' chips are 90W chips.

The 180W threadripper chips also lie but in a different way. Instead maintaining the advertised base frequency and violating the marketing TDP as their AM4 brothers do, the ThreadRipper chips maintain the official TDP and violate the base frequency. Next graph are clocks for the 1950X

IMG0054462.png


Marketing base frequency is 3.4GHz, but we can check several cores are working under 3.4.

We could see such a throttling under several applications on the 1950X although it is far from systematic, the applications using AVX seem the most affected, logically. The limitation in our case after many checks was not thermal, but related to consumption. AMD has confirmed to us that such a throttling was theoretically possible without completely confirming it. The 1920X did not exhibit such behavior in our tests, logically enough.

Somehow, it's a shame to see such a throttling even if in absolute terms, it is rather happy to see the Threadripper satisfy their TDP (an exceeding as the 1800X multiplied by two would have been complicated to accept). Going under the basic frequency is however something embarrassing, even if it is not the first time that one sees this behavior at AMD.

Conclusion:

Intel satisfies the official TDP. AMD either violates TDP (AM4 socket) or violates base frequency (TR4 socket).
 
No, TDP and Power Consumption are not the same. Thermal conductance affects power consumption through resistance. And that graph is PRETTY clear on the why (although the load windows graph has nothing to do with power, only *load*).

The way Turbo works implicitly violates TDP as long as the cooling can sustain higher thermal load to dissipate. Why is that so damn hard to understand? With the "B" series Intel just made sure the TDP = Max Energy Consumption at all times, making those "turbo pockets" of heat smaller and telling the BIOS to don't use any thermal headroom (probably a hard cut asked by the OEMs to have longer life expectancy of fans and other pieces that get affected by heat). The average power consumption of any CPU when it can manage heat transfer well, will be lower (makes turbo pockets bigger) while keeping initial power consumption as high as it needs for as long as the pocket lasts.

I'll stop here; this discussion comes and goes all the goddamn time.
 

goldstone77

Distinguished
Aug 22, 2012
2,245
14
19,965
Yuka makes a great point about TDP and thermal limitation vs. power consumption. Power consumption will vary with the quality of the cooling system, heatsink/fan, based on the thermal regulation of the processor. Take a heatsink and fan off of a processor, and the processor will shut down quickly. Run it on LN2, and sky is the limit.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


No one said they are the same. And in my above post I just discussed differences between TDP and instantaneous power consumption.

Of course TDP continues being sustained power consumption. As I have explained this a dozen of times and as the above graph from Intel and the textbook on computer architecture from Hennessy and Patterson.



No. Turbo is an instantaneous load can be run when the average dissipation is under the TDP. This is what the above graphs means by accumulation of "energy budget". Instantaneous consumption is not average or sustained consumption.



I ask myself the same. About one year ago reviews demonstrated that AMD lied about the TDPs of Zen processors, The marketing values 65W and 95W aren't the real values. The same reviews prove that TDPs communicated by Intel are real. Since then Intel threads like this one are filled with repetitive claims that Intel processors also violate TDPs, which isn't true.

This resembles what happened when reviews showed that Polaris violated PCIe standards. A small group of people in forum was during months pretending that Nvidia cards also violated the standard. It wasn't true, all Nvidia cards were full compatible with PCie standards. People pretending otherwise mixed averages with peak loads and mixed loads in a single channel with loads on all channels.



Official TDP are defined for ambient temperatures not for extreme subzero tempts.

Also if a CPU is rated at 100W on stock settings, then it requires a cooler what can dissipate 100W. If a 50W cooler is used or id the cooler is removed completely, then the CPU cannot work on default settings and have to throttle or shut down (ancient CPUs without thermal protection mechanisms would burn). This is not a mystery.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995


Yes it does mean exactly that.
Power2.png


Your own graphs explain that very clearly.
sandybridge_061.jpg

Intel speaks about thermal budget. If the cooling system extracts more thermal energy than the official TDP, the turbo will engage more frequently and consumptiion will go above official TDP as many reviews show.
No one satisfies anymore the official TDP if the cooling solution allows to dissipate more thermal energy.

But we already had this discussion many times, sig ....
 

manleysteele

Reputable
Jun 21, 2015
286
0
4,810
There are two good reasons to buy a K processor. The first, and for me, at least, the better one, is that one gets better clock profiles within the rated TDP. The other is that you intend to violate the TDP, right form the jump, or in other words, you intend to run higher clocks and use better cooling in order to maintain the higher clocks and higher temperatures those clocks produce. As an aside, you may also wish to run a workload that requires more cooling than the stock TDP can handle.

Personally, I refuse to buy a cooler when the manufacturer doesn't list a target TDP. For some manufacturers, the TDP rating of a cooler is some kind of state secret, or, at least it appears so from an outsiders perspective.

As far as the listed TDP is concerned, it is not violated when you perform any of the above acts except the first. If you are running a mainstream program and the CPU won't maintain it's specified clocks with a good cooler, correctly installed, then the TDP is violated. Otherwise, you're not getting what you paid extra to get. If you are doing any of the other things in the list, the TDP violation is your doing and not the manufacturers problem.

 
Because OEMs do care about TDP. This is not a "consumer" vs "companies" discussion, but how Intel and AMD go "off-spec" when they-themselves define a metric for OEMs. Here Juan is saying Intel is within those limits and we're saying they're not.

Or, let me add a bit more. If you're a consumer looking at an OEM PC and Intel says to the OEM: "yeah, 65W TDP", then the OEM puts a 65W capable dissipation unit and it burns in a couple days. You will have an angry client complaining to the OEM and the OEM complaining to Intel. This doesn't apply to Intel, because they stopped bundling a cooler capable of handling the off-spec TDP long ago and gave that burden to the consumers willing to get a K CPU. If an OEM gives a K CPU in one of their offerings, they have to deal with that as well.

Cheers!
 

manleysteele

Reputable
Jun 21, 2015
286
0
4,810
Fair enough. As yet another aside, I quit using Intel's coolers because they are dust magnets. In my experience that worked fine, when they were clean. The problem was, they were never, ever clean. Handy if you didn't want to search for the stray dust in your case, but otherwise, problematic.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


It is not only me. It is reviews as well. The same reviews that proved that AMD lied about TDPs, whereas Intel doesn't. That is the reason why those reviews write stuff like

Ryzen Threadripper 1950X : 16C/32T, 180W, 999$
Intel Core i9-7900X : 10C/20T, 140W, 999$
Ryzen Threadripper 1920X : 12C/24T, 180W, 799$
Intel Core i7-7820X : 8C/16T, 140W, 599$
Ryzen 1800X : 8C/16T, "95W", 499$

Only the 1800X has its TDP between marks, because they know the real TDP is 128W.

And talking about OEMs. Those OEMs can check things by themselves. They know that 15W Intel chip is a 15W chip, but a 15W AMD chip is in reality a 20--25W chip, so they had to replace the display and memory by ones consuming less power to maintain battery life similar between AMD-powered and Intel-powered chips.
 

manleysteele

Reputable
Jun 21, 2015
286
0
4,810
I saw that, too. My reaction? Meh. An ingenious solution to a non-existent problem? Maybe. I'm sure they'll sell some. Just not to me. Maybe next time. Or the one after the one after that.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


A company as Intel has a plan A and a couple of B plans in the pipeline in case plans A fails. It is not how if engineers have designed, implemented and checked an 8-core CFL in three months.
 

Gon Freecss

Reputable
Apr 28, 2015
448
0
4,810
So, if Intel made an 8 core Coffee Lake die, do you think it's feasible to think they're making a 10+ core Icelake die?

Also, I'm really interested in knowing why they didn't just backport Cannonlake onto 14nm++. They already had an 8 core design ready back in 2016.
 

aldaia

Distinguished
Oct 22, 2010
533
18
18,995


The canonlake backport to 14nm++ already exists and is named Coffee Lake :)
Coffee Lake is the third 14 nm process refinement following Broadwell, Skylake, and Kaby Lake intended for desktop.
Cannon Lake is Intel's 10-nanometer die shrink following Broadwell, Skylake, and Kaby Lake intended for mobile.
Upcoming Cascade Lake is the third (or maybe forth) 14 nm process refinement following Broadwell, Skylake, and Kaby Lake intended for server and enthusiasts.
Skylake, Kaby Lake, Coffe Lake, Cannon Lake, and Cascade Lake are all the same microarchitecture.
Broadwell to Skylake was last microarchitecture change.
 
Status
Not open for further replies.