intel Intel's TDP ratings for higher-end CPUs make no sense anymore

[Ash_Kechummm]

 

Summary

Dr. Ian Cutress recently posted a review of the core i7 10700, a "65 watt" TDP chip, whose testing revealed that it consumed as much or more power than the 10700k, a "125 watt" TDP chip.

 

Quotes

Quote

The Core i7-10700K, technically the higher performance processor, came in with a 205 W detected peak power draw in our AI Benchmark. The Core i7-10700 by comparison was observed at 215 W during our LINPACK benchmark....If we look at some of the other processors for comparison, both of the Core i7 use more power than Intel’s 28-core Xeon processors, and match the Xeon W-1200 series parts, although there’s still a way to go to match the Core i9-10900K. AMD’s peak here is 142 W.

Quote

1. Up to 3 cores or less, the Core i7-10700K uses more power
2. Between 3-cores to 7 cores, the results are identical
3. Only at 8 cores does the i7-10700 use more.

Quote

...This means that users who buy the Core i7-10700 in this review, despite the 65 W rating on the box, will have to cater for a system that will not only peak around 215 W, but sustain that 215 W during any extended high-performance load, such as rendering or compute. We really wished Intel put this 215 W value on the box to help end-users determine their cooling, as without sufficient guidance, users could be hitting thermal limits without even knowing why. 

 

My thoughts

I had heard people talk about Intel having a bad time making chips that compete well at the high end (at the lower end, the core i3 10100f is slightly better than a ryzen 3 3100, and is actually available for much less than a r3 3100 in most of the world, and that is including platform costs), but holy sh*t did I not expect the power consumption to be over 3 times what's written on the box (215 vs 65 watts).

 

TL;DR: Intel's high-end chips guzzle a lot of power (even the ones that are supposed to be energy-efficient), output a ton of heat, and are still barely able to compete with Ryzen products.

I feel like the only thing that's saving Intel for now in the high-end is the fact that chips like the i7 10700 (non-k and k) and the i9 10850k are actually in stock in most of the world at close to MSRP, unlike Ryzen 3000 and 5000 chips

 

 

 

 

Sources

https://www.anandtech.com/show/16343/intel-core-i710700-vs-core-i710700k-review-is-65w-comet-lake-an-option/

Edited January 26, 2021 by Ash_Kechummm
spelling lol

[WereCatf]

14 minutes ago, Ash_Kechummm said:

Intel's TDP ratings for higher-end CPUs make no sense anymore

To be quite frank, they have never made much sense. I guess the move from "not much" to "completely zero" is still some sort of an achievement, but....

[Stahlmann]

Both the TDP numbers of Intel and AMD CPUs are not what you get. On intel it's often more because most BIOS have some sort of "MCE" enabled by default, which take them beyond stock restrictions. And afaik AMD's TDP formula doesn't even include power consumption.

[Jet_ski]

We should all be used to Intel pumping up the voltage of the same CPUs every year for a 5-10% improvement with a new lake name, instead of actually innovating.

[leadeater]

16 minutes ago, WereCatf said:

To be quite frank, they have never made much sense. I guess the move from "not much" to "completely zero" is still some sort of an achievement, but....

Well the context that needs to be applied is this is specifically related to motherboards and motherboard settings, Intel can only give a product specification and rating for the product itself under their test conditions and parameters.

 

Now I know Intel allows the PL2 values to be changed, power amount and time, and actively encourages and benefits from it for gaming vendors/motherboards but these power draws are artifacts of these parts combinations and configurations alone. Where the i7 10700 is most used is in corporate OEM computers that use the Intel specification so do indeed drop down to 65W after turbo time is exceeded.

 

Like this has to be repeated numerous times for what ever reason because it just doesn't seem to hold in peoples minds, TDP specifications are for cooling designs and coolers are designed for the target device and configuration. You as the buyer shouldn't be looking at the CPU TDP on Intel Ark, you should be looking at the typical power draw for your actual configuration you are going to use i.e. look at gaming reviews and buy a cooler based on that. The motherboard vendors need to tell you what the PL2 settings are not Intel.

 

These issues are one of the many reasons I like Gamers Nexus reviews, Steve does not use motherboard defaults and configures the systems to adhere to the Intel specifications and lists that as stock on the graphs.

 

Yes this opens the can of worms of what is 'stock' and 'out of the box', just remember motherboard manufacturers started this and they hold the majority of the blame, even though Intel has no specific issues with what is happening.

[porina]

34 minutes ago, Stahlmann said:

Both the TDP numbers of Intel and AMD CPUs are not what you get. On intel it's often more because most BIOS have some sort of "MCE" enabled by default, which take them beyond stock restrictions. And afaik AMD's TDP formula doesn't even include power consumption.

The first and biggest take away point is that TDP is not in any way a measure of power consumption. Most misunderstandings start from there.

 

On Intel side, Intel have some "serving suggestions" to power limit, but these are NOT enforced, and it is up to the system builder to dial in values for their goals. They may choose to include better cooling for higher performance, or cheap OEM boxes might put in the bare minimum. Enthusiast "Z" chipset mobos tend to go for the performance end, but the locked chipsets may be more likely programmed with something closer to the suggested values. Changing the power limit by itself is NOT an overclock. This is unrelated to MCE which IS an overclock. I have not personally seen mobos which default to MCE on, although some push you towards it when you manually enable XMP for example, so you have to decline it.

 

On AMD side, stock CPUs run to a PPT power limit. For 65W TDP CPUs, this was 88W, at least in the Zen 2 era. I'm not sure about other generations. Similarly the higher TDP CPUs have an even higher PPT limit. Unlike Intel, AMD do consider changing the power limit to be an overclock.

 

Anyway, while the article is a new take, it is an old "problem" in that this deviation has been around for many years. So, there isn't really any radically new information here other than a specific example of two similar CPUs.

[Taf the Ghost]

This isn't too surprising, this is what Intel needs to do to compete. We see it in GPUs every generation. Whoever is a little behind on the efficiency curve will choose to crank up the clocks until they're well ahead. Nvidia has done with this Ampere, to the point AMD was able to market a 300w card as being far more efficient than their competitor. (They aren't wrong, they've got a big advantage in that space at the top end.)

[porina]

41 minutes ago, leadeater said:

Yes this opens the can of worms of what is 'stock' and 'out of the box', just remember motherboard manufacturers started this and they hold the majority of the blame, even though Intel has no specific issues with what is happening.

At least in the enthusiast area, I think there is a kinda "understanding". Chances are, if you get a Z chipset overclocking board, drop in a k CPU, it'll go unlimited power limit. It's more a grey area if either chipset or CPU is a locked model. You can probably still go higher power limit, but it may also revert to the default setting. So I agree mobo manufacturers have the responsibility to communicate that.

 

OTOH I really wish Intel would publish more values, but can understand why they don't. Of personal interest, I'd like to see max power consumption and all core boost clock. The big problem here is that if max power is specified, it has to be a value that will never be exceeded by any sample ever by any software load, so it will be worse than what most will end up getting. If they mention a typical value, you'll end up with people complaining they got one below average. Similarly with the all core boost clocks, it will vary depending on the workload. It will vary so much, you might as well go back to stating base clock. Oh, that's what we actually have.

[Stahlmann]

1 hour ago, porina said:

I have not personally seen mobos which default to MCE on, although some push you towards it when you manually enable XMP for example, so you have to decline it.

Asus boards often do this with Intel CPUs. Even a lower-end Z board like my Prime Z370 did automatically use MCE on my 8600K. Even clearing CMOS did restore to MCE enabled. Idk exactly how it is with other brand boards, but even Hardware Unboxed did run their 9900K test bench with MCE enabled because it was a setting that is enabled by default and "will be what most people will run" as most people don't do any BIOS tweaking whatsoever.

[porina]

5 minutes ago, Stahlmann said:

Asus boards often do this with Intel CPUs. Even a lower-end Z board like my Prime Z370 did automatically use MCE on my 8600K.

To be clear then, without you changing any setting in bios, your 8600k does 4.3 GHz all core, not 4.1 GHz as would be expected without MCE?

 

The newest Asus boards I have are X299 and Z270. In those and older boards, MCE defaulted to off. From memory, if you turn on XMP, it will immediately ask you if you want Asus core enhancement or words to that effect. If you say yes, then it turns on MCE. Since I don't own a newer board from Asus, I can't say if they changed policy since then.

 

On newer Intels, I have Asrock Z370/Z390 boards, and Gigabyte Z490 board. None of them default MCE on, hence my statement I have never seen a mobo that defaults to MCE on.

[Stahlmann]

2 minutes ago, porina said:

To be clear then, without you changing any setting in bios, your 8600k does 4.3 GHz all core, not 4.1 GHz as would be expected without MCE?

Yes, all cores were boosting to 4.3 and stayed there. Even for longer loads.

 

I don't quite remember if there was a prompt when enabling XMP. I don't have the hardware anymore.

[AndreiArgeanu]

I'm sorry what? 215 watts? Is Intel genuinely on their way to overtake the AMD fx series in power consumption? I mean they're already near the FX 8150, but they might get to the 350W power consumption the FX 9590 was able to pull stock.

 

And I'm genuinely worried about low end B and H series boards that have poor VRM's, I'd imagine they'll end up struggling holding the full boost clock.

[RejZoR]

Intel's TDP makes absolutely no sense because they are measuring it at CPU's base clock. Show me one desktop Intel CPU that actually runs at its base clock when under normal operation. Maybe those crap chips that don't even have boosting or laptop ones that are either scorching hot and drop to base clock through that or they are just limited to actual power that they need to obey otherwise they'd just melt.

[zheega]

AMD's TDP doesn't make any sense either, people just aren't as focused on it, since the difference between the TDP number and actual power consumption isn't as big as with Intel.

Both companies should give a per-CPU number for "heat dissipated under sustained load at 80C".

[SavageNeo]

9 minutes ago, zheega said:

AMD's TDP doesn't make any sense either

yes it does at some cpus. my old 3400g used just over 65w under laod, and that is 65w tdp chip. 

[MageTank]

2 hours ago, leadeater said:

Well the context that needs to be applied is this is specifically related to motherboards and motherboard settings, Intel can only give a product specification and rating for the product itself under their test conditions and parameters.

 

Now I know Intel allows the PL2 values to be changed, power amount and time, and actively encourages and benefits from it for gaming vendors/motherboards but these power draws are artifacts of these parts combinations and configurations alone. Where the i7 10700 is most used is in corporate OEM computers that use the Intel specification so do indeed drop down to 65W after turbo time is exceeded.

 

Like this has to be repeated numerous times for what ever reason because it just doesn't seem to hold in peoples minds, TDP specifications are for cooling designs and coolers are designed for the target device and configuration. You as the buyer shouldn't be looking at the CPU TDP on Intel Ark, you should be looking at the typical power draw for your actual configuration you are going to use i.e. look at gaming reviews and buy a cooler based on that. The motherboard vendors need to tell you what the PL2 settings are not Intel.

 

These issues are one of the many reasons I like Gamers Nexus reviews, Steve does not use motherboard defaults and configures the systems to adhere to the Intel specifications and lists that as stock on the graphs.

 

Yes this opens the can of worms of what is 'stock' and 'out of the box', just remember motherboard manufacturers started this and they hold the majority of the blame, even though Intel has no specific issues with what is happening.

This is absolutely correct. We recently got an MSI Z490 board in the lab that had stock power and current limits of 4096W which blew my mind. Normally, you'd see limits imposed to prevent damage from overclockers that didn't know what they were doing, or an errant spike in power delivery that went awry, but to ship boards with no power/current limit in-place as the default configuration is absolutely silly. The funny part is, when you install a CPU for the first time, you are greeted with a message that states "CPU changed, Press F1 to load Overclock, F2 to load defaults" or something along those lines, and both have unlimited power/current limits with infinite boost durations.

 

Board partners are doing anything and everything they can to squeeze out points on benchmarks and its starting to make me sick. We've seen it in the past few years with those sketchy "performance modes" in the BIOS that favored specific benchmarks and now we are seeing them go further down this rabbit hole with tweaks like this that could actually damage a customers hardware if left unchecked.

[porina]

26 minutes ago, AndreiArgeanu said:

And I'm genuinely worried about low end B and H series boards that have poor VRM's, I'd imagine they'll end up struggling holding the full boost clock.

The CPU doesn't just take the power by itself. The mobo sets the limit. If a mobo manufacturer makes a cheap board that can't sustain higher powers, they'll (hopefully) set the bios accordingly for long term power. Remember the power limit is a system set level, not CPU chosen.

 

13 minutes ago, RejZoR said:

Intel's TDP makes absolutely no sense because they are measuring it at CPU's base clock. Show me one desktop Intel CPU that actually runs at its base clock when under normal operation. Maybe those crap chips that don't even have boosting or laptop ones that are either scorching hot and drop to base clock through that or they are just limited to actual power that they need to obey otherwise they'd just melt.

It doesn't work how you want it to, and you don't get to dictate that. The Intel specification is that if you provide a cooler rated that the TDP, you will get at least the base clock. That's what it means. In practice, even if you have a horrible Dell or HP built down to the minimum, you can still boost above base for short times.

 

3 minutes ago, zheega said:

Both companies should give a per-CPU number for "heat dissipated under sustained load at 80C".

AMD have a fixed power limit at stock so you'll just get that, unless you remove it, which most people probably wont.

Intel give system builders the choice to set whatever limit they want, so it may vary on the system.

It doesn't take into consideration how much work is done e.g. Zen was often praised for its low power consumption under Prime95 for example, but that's in part because it absolutely sucked in that kind of workload. Likewise AVX-512 in Skylake-X was notorious for being even hotter than ever, but the throughput was massive, up to double IPC compared to AVX2. 

It will vary a lot depending on the workload.

 

Basically, it is hard to distill this down to a single number.

[porina]

4 minutes ago, MageTank said:

This is absolutely correct. We recently got an MSI Z490 board in the lab that had stock power and current limits of 4096W which blew my mind. Normally, you'd see limits imposed to prevent damage from overclockers that didn't know what they were doing, or an errant spike in power delivery that went awry, but to ship boards with no power/current limit in-place as the default configuration is absolutely silly.

I'm not at all surprised at the practically unlimited power limit. I think this is pretty much the norm on most OC boards with K CPUs at least back to Skylake era. This is an easy way to get more performance without overclocking. When actually overclocking beyond that, I often find you hit the current limit before you can do anything really stupid. That is not default unlimited!

 

4 minutes ago, MageTank said:

The funny part is, when you install a CPU for the first time, you are greeted with a message that states "CPU changed, Press F1 to load Overclock, F2 to load defaults" or something along those lines, and both have unlimited power/current limits with infinite boost durations.

I haven't seen that, but combined with @Stahlmann earlier comment about MCE it does seem some mobos may be pushing towards OC by default.

 

BTW to check, the current limit was also unlimited?

[AndreiArgeanu]

18 minutes ago, porina said:

The CPU doesn't just take the power by itself. The mobo sets the limit. If a mobo manufacturer makes a cheap board that can't sustain higher powers, they'll (hopefully) set the bios accordingly for long term power. Remember the power limit is a system set level, not CPU chosen.

Regardless that'll mean that the CPU will throttle due to power limits, therefore not able to hold full boost clocks which is what I was worried most about.

[porina]

2 minutes ago, AndreiArgeanu said:

Regardless that'll mean that the CPU will throttle due to power limits, therefore not able to hold full boost clocks which is what I was worried most about.

That is working as intended, both by AMD and Intel. Boost is opportunistic, there is no guarantee you will always hit maximum clocks.

[MageTank]

18 minutes ago, porina said:

I'm not at all surprised at the practically unlimited power limit. I think this is pretty much the norm on most OC boards with K CPUs at least back to Skylake era. This is an easy way to get more performance without overclocking. When actually overclocking beyond that, I often find you hit the current limit before you can do anything really stupid. That is not default unlimited!

 

I haven't seen that, but combined with @Stahlmann earlier comment about MCE it does seem some mobos may be pushing towards OC by default.

 

BTW to check, the current limit was also unlimited?

Yes, both power and current. The power limit was one thing, but it was the current limit that actually scared me. Power limit by itself is harmless, but an unlimited current will outright kill hardware even if the user isn't overclocking. A poor quality VRM having a random issue can cause current to flake out and go nuts for the briefest of moments and that would end poorly.

 

Also, the specific board in question was a Z490-A Pro, which I'd consider on the lower-end of "overclocking" boards. Definitely wouldn't trust that VRM to run with an unlimited current by any means. We were testing the board for system integration too, meaning a mass production environment, lol.

[porina]

2 minutes ago, MageTank said:

Yes, both power and current. The power limit was one thing, but it was the current limit that actually scared me. Power limit by itself is harmless, but an unlimited current will outright kill hardware even if the user isn't overclocking. A poor quality VRM having a random issue can cause current to flake out and go nuts for the briefest of moments and that would end poorly.

Ok, unlimited current by default is new to me and I agree that is more scary than unlimited power.

 

I feel I've picked poorly in the past before power delivery was as much a thing. I have Asrock AMD and Intel mobos I don't consider 24/7 compute stable with the CPU at stock on either side. Because of the low frequency of occurrence, it took a while to come to that conclusion. The B450 board the only thing I found to help was manually putting in a small positive voltage offset. On Z370, my solution was to disable turbo on the CPU (so I go from 4.3 to 4.0 all core, no big deal). In either case, it appeared to work fine under casual user loads. Certainly for future purchases I'm paying a lot more attention to power delivery.

[AndreiArgeanu]

14 minutes ago, porina said:

That is working as intended, both by AMD and Intel. Boost is opportunistic, there is no guarantee you will always hit maximum clocks.

But benchmarks aren't done at base clock, most if not all are done where the CPU is boosting to the best of it's ability and where it is not held back by the motherboard. And these power consumption metrics are rarely mentioned in benchmarks and almost never considered when it comes to how the power consumption will affect the performance when pairing with a low end motherboard with poor VRM's, which is guaranteed to happen, especially in prebuilts.

[porina]

1 minute ago, AndreiArgeanu said:

But benchmarks aren't done at base clock, most if not all are done where the CPU is boosting to the best of it's ability and where it is not held back by the motherboard. And these power consumption metrics are rarely mentioned in benchmarks and almost never considered when it comes to how the power consumption will affect the performance when pairing with a low end motherboard with poor VRM's, which is guaranteed to happen, especially in prebuilts.

Do you want to benchmark a CPU or benchmark a system? The two are not the same. From an enthusiast point of view, most will self build, and the tech reviewers catering to that audience will review accordingly, including power usage indicators.

[CarlBar]

OP:

 

 

This has been a thing with intel for ages. It's like saying the sky is blue at this point.