Question Intel vs. AMD CPUs - out of the box POWER DRAW

[Kocicak]

I was surprised, after I read in the recent review, that if you place Intel CPU in the out of the box motherboard, it will draw over 200W instead of advertised 65W.

I have a fresh experience with three different CPUs from the new AMD 5000 series, out of the box they run at their advertised power draw of 105W. I placed them all in one motherboard so I do not know if this is not specific just to my motherboard.

So is it now a general fact across more motherboard manufacturers, that if you put Intel CPU in OUT OF THE BOX (default BIOS settings not changed by user) motherboard, it will draw much more than advertised power, and if you put AMD processor in out of the box MB, it will run at specified power draw?

 

[blckgrffn]

Actually, it was Ian's 4th (maybe more?) article on the same subject:

Why Intel Processors Draw More Power Than Expected: TDP and Turbo Explained

www.anandtech.com

Talking TDP, Turbo and Overclocking: An Interview with Intel Fellow Guy Therien

www.anandtech.com

Intel’s Tiger Lake 11th Gen Core i7-1185G7 Review and Deep Dive: Baskin’ for the Exotic

www.anandtech.com

So still same stuff (I may have missed but I do try to read all the main articles.) I am more interested in something like - if we take the 10850k and give it a range of PL2 max numbers (starting at 125 and up by 25W to it's full glorious ~250W or more max) what does the performance curve look like?

I feel like Intel just set the PL2 level to "stupid high", again, like the 3090 voltages that add ~5% performance on 20% increased power usage? If I am setting up a 10th gen Intel system (because I can really just go buy those parts now) what's a sweet spot? It's possible that in non-core saturating workloads (gaming, normal desktop use) setting a far lower PL2 is going to change performance only a very small bit but reduce power usage spikes considerably. I'd think that be good for all components.

It's like a 2 minute exercise to set it in the bios and forget it afterwards. I always manually configure it on 9/10th gen setups based on the TDP of the cooler and PL2, but perhaps (very likely) I am a tiny minority.

 

[blckgrffn]

OEMs most likely do configure it to run at the short term boost. And when it drops it could very well run at base clock.

This is what makes using an enthusiast motherboard to include 9th/10th gen parts in something like Bench such a waste of Ians time. The vast majority of these non-k CPUs are going into OEM systems that are likely to show very different characteristics, and how do you display that? Starting adding asterisks all over the place?

I find it boggling that Intel doesn't make board manufacturers certify non-K CPU settings to be Intel guideline based in their firmware. Really, why get a K CPU if it, out of the box, has no different performance? OC'ing these things is just nuts given the investment in components and cooling required to make it worth anyone's time, and is super redundant now that AMD has the crown.

 

[Mopetar]

So what is the point of the TDP if everyone is going to ignore it? As if a sustained power consumption of 200W+ doesn't require far heavier cooling plus a bigger Power Supply to begin with.

I would love to see benchmarks with these Processors capped to the same TDP than Ryzens, albeit the results are obvious cause they would be completely stomped.

The TDP values from Intel are what the chip is guaranteed to provide in terms of base clock at the described power budget. Almost all chips can be fed additional power outside of the rated amount and will handle it just fine up to a point, but just the mere fact that we have multiple cores and a turbo boost now means that we have a lot of different ways to measure a chip whereas in the past you had a single core with a fixed clock speed.

So which of the following is the best way to measure and report the single number that we'll designate as TDP:

A) Maximum power draw possible when using a single core/thread at base operating frequency.
B) Maximum power draw allowed by the CPU (artificial limitation) when using a single core/threat at base operating frequency.
C) B except when operating at maximum boost frequency permitted by the CPU.
D) Maximum power draw possible when using all cores at base operating frequency.
E) Maximum power draw allowed by the CPU (artificial imitation) when using all cores at base operating frequency.
F) Maximum power draw allowed by the CPU when all cores are boosting to whatever maximum frequency they can acheive above the base amount.
G) Maximum power draw allowed by the CPU under any operating circumstances, including extended instructions sets like AVX.

There are probably dozens of other small (but important) derivations that someone might care about, but no one will ever agree on which is best. For Intel they've decided that TDP means that the chip they sell you can run all cores as the advertised base frequency within that power budget. For AMD it means the chip will consume up to that much while boosting above the base frequency on all cores, but of course you can still adjust the settings in the CPU or MB to given it additional power beyond that and clock higher.

The only real gripe is that the term isn't used consistently by everyone, but you'll never get everyone to agree to which definition is best.

 

[TheELF]

If I am setting up a 10th gen Intel system (because I can really just go buy those parts now) what's a sweet spot? It's possible that in non-core saturating workloads (gaming, normal desktop use) setting a far lower PL2 is going to change performance only a very small bit but reduce power usage spikes considerably.

PL are upper limits, they don't force the CPU to always use that amount of power, they force the CPU to not exceed that amount.
If you run non-core saturating workloads the CPU will never even reach near PL2 in the first place, not even during turbo, not ever.

 

[Zucker2k]

I had to post the above link for the surprise in there.

 

[blckgrffn]

PL are upper limits, they don't force the CPU to always use that amount of power, they force the CPU to not exceed that amount.
If you run non-core saturating workloads the CPU will never even reach near PL2 in the first place, not even during turbo, not ever.

In the Anandtech article on 9th gen CPUS that had a greater than 200W PL2 set they could never get them to hit that power figure - only managing a 9900K to ~170W.

I understand what you are saying, but if you can get some significant percentage of the sustained all core performance with much lower levels, wouldn't that at least be interesting? That's what I found missing from the review - if Intel gives soft guidance and motherboard makes do whatever the crap they want, how should we as system builders configure these easy to set values to get the best performance for power consumption that we feel is reasonable. That's the kind of data I come to reviews for. Based on other sites, I've seen that 65W is pretty restrictive as a PL1, but what about 80W? 100W? For x86 CPUS at this time it seems pretty reasonable.

Heck, according to the Anandech article, loading a 9900K at 4C/4T and 3 GHZ resulted at 23W max power usage. While this 14++++ or whatever, Apple has shown that core count and rough speed to be a real bogey for performance and power consumption. I hope that we'll see some additional testing like that going forward. Anyway.

The Intel 9th Gen Review: Core i9-9900K, Core i7-9700K and Core i5-9600K Tested

www.anandtech.com

I also don't know exactly what people who use the computers I build for and since I don't like their PL2 to be outside of what their coolers could handle it get configured.

There are going to be times (like DX12 games building/caching shaders) where you are going to get some pretty solid burst power consumption. For any board/PSU that's going to be a stressful event and I'd prefer to keep them minimized.

I could do this investigation myself but I am not that motivated, only just motivated to come here and write a too long post about it

 

[Abwx]

Corrected that for you. Ian's review gave me quite a laugh. First thing he did was pull off the 65W cooler that would keep the average power to 65W in the long term. Then he complained that with better cooling, the chip would turbo longer, run faster, and use more than 65 W.

So if i understand your "logic" one who wants to run at 65W max has to use a crappy cooler so that the CPU will reach rapidly the necessary 100°C that will make it throttle at rated TDP..

If Ian used a better cooler it s to have a lower operating temp at full theorical TDP, that s exactly what everyone in this forum would do, or perhaps that you prefer noisy PCs with fans spinning at full speed.?.

 

[blckgrffn]

So if i understand your "logic" one who wants to run at 65W max has to use a crappy cooler so that the CPU will reach rapidly the necessary 100°C that will make it throttle at rated TDP..

If Ian used a better cooler it s to have a lower operating temp at full theorical TDP, that s exactly what everyone in this forum would do, or perhaps that you prefer noisy PCs with fans spinning at full speed.?.

Yeah, that’s a fairly ludicrous way to do it.

More realistically just he could have given us two sets of numbers. One set as it was (shenanigans nearly all aftermarket boards pull) and with the Intel spec PL1, PL2 turbo duration specs to show “OEM” performance. When I was researching the 9700F (it was $180 at MC for quite some time) I found reviews that did this and it was interesting, at least. Depending on how threaded a game it varies quite a bit.

 

[TheELF]

if Intel gives soft guidance and motherboard makes do whatever the crap they want, how should we as system builders configure these easy to set values to get the best performance for power consumption that we feel is reasonable. That's the kind of data I come to reviews for. Based on other sites, I've seen that 65W is pretty restrictive as a PL1, but what about 80W? 100W? For x86 CPUS at this time it seems pretty reasonable.

65W is only for the locked CPUs and is probably what the intel cooler is rated for.
PL1 should be set to whatever the cooler can handle for "ever" at the temp you consider good. (equal platform thermal capacity)

Intel® Core™ Processors, FPGAs, GPUs, Networking, Software

Browse Intel product information for Intel® Core™ processors, Intel® Xeon® processors, Intel® Arc™ graphics and more.

www.intel.com

You can always ask your customers if they prefer cool and quiet or max performance and set PL2 accordingly either to the recommended 1.25 of PL1
or higher.

 

[TheELF]

So if i understand your "logic" one who wants to run at 65W max has to use a crappy cooler so that the CPU will reach rapidly the necessary 100°C that will make it throttle at rated TDP..

If Ian used a better cooler it s to have a lower operating temp at full theorical TDP, that s exactly what everyone in this forum would do, or perhaps that you prefer noisy PCs with fans spinning at full speed.?.

Yes if you are too stupid to change some settings and need everything automated that's what you are supposed to do.
Anybody with half a brain can figure out how to lock down the TDP so that a better cooler only provides better cooling and not also a higher power draw,
or how to use XTU to only boost the apps you want to the degree you want.

 

[Abwx]

Yeah, that’s a fairly ludicrous way to do it.

More realistically just he could have given us two sets of numbers. One set as it was (shenanigans nearly all aftermarket boards pull) and with the Intel spec PL1, PL2 turbo duration specs to show “OEM” performance. When I was researching the 9700F (it was $180 at MC for quite some time) I found reviews that did this and it was interesting, at least. Depending on how threaded a game it varies quite a bit.

Problem with Intel s CPUs is that you cant get rid of PL1 even if the bios is set to say 65W, the CPU will still exceed its rated TDP for a few dozen seconds, this is the case even with 35W T models, they will start at 65W (even 112W for the 8700T) and then settle at a loose 35W whatever the temperature, what Intel call tau 1 (PL1 duration boost) will prevail whatever the bios settings.

The test with the Core i7-8700T shows how differently a T-model can do with it. With activated turbo, the processor, including AVX2 load, can allow itself up to 112 watt package power. It then operates at 3.8 GHz for all cores, which is also 1,400 MHz more than the base clock that is used for the TDP calculation. If the turbo is deactivated, it is still 45 watts in the AVX test by Prime95, whereas in applications such as blenders with a maximum of 37 watts it is pretty much the 35 watts mentioned by Intel.

https://translate.google.com/transl...ore-i7-8700t-i5-8500t-cpu-test-coffee-lake/2/

[/URL]

They should simply allow for brick walled TDPs within the bios, AMD is somewhat better in this respect since they have more or less proper user power limitation.