Core i9-9900K Power & Thermals, Did Linus (and OC3D TV) Get it Wrong?

[schwellmo92]

 

Quote

Recently we released a Core i9 9900K review and sighted thermals as being a pretty major issue for the 8 core CPU, in fact thermals were so bad I was worried I done something wrong but prior to release fellow aussie youtubers confirmed the horrendous thermal results. There were however a few outliers, OC3D TV being one, they reported shockingly low operating temperatures, this didn't make sense till I dug in to the system specs and discovered the issue.

 

Hardware Unboxed dig in to the outlier results for the Intel 9th generation reviews, being OC3D TV and Linus Tech Tips. His belief is that both OC3D TV and Linus Tech Tips used motherboards with under-powered VRM's, causing VRM throttling allowing the processor to run cooler, draw less power and perform slightly lower. The results Steve comes up with are interesting and definitely go a long way to describing the discrepancies, I am interested in hearing what OC3D TV and Linus Tech Tips have to say about this.

 

EDIT: It could also be caused by TDP throttling.

[Shally]

I was confused when Linus said thermals were great, even though others said they were only slightly better. Hopefully Linus can respond by using a couple other motherboards, we know they have them

[Spotty]

Quote

 Linus Tech Tips
Pinned by Linus Tech Tips
Linus Tech Tips
15 hours ago (edited)
Since our thermal and power consumption numbers are very different from other publications, here's some clarification on our board settings:

-Multi-core enhancement: Disabled
-SVID behavior: "Intel Fail-Safe" / Intel's default
-XMP: Enabled
-All power duration and limit settings default
-Core multiplier at defaults
-Core voltages at defaults

Cooler: Corsair H115i w/IC Graphite thermal pads

The performance numbers we got were double-checked against Intel's engineers, and they were in line; This suggests that our thermal tests are indicative of stock performance (with motherboard vendor optimizations disabled). We did not ask Intel about their thermal or power consumption results.

Response issued by LTT is pinned to the top of the comments for their video [on youtube].

Edited October 22, 2018 by Spotty

[Rauten]

Doesn't LTT normally use the highest end ASUS board available? Those usually have overkill-level VRMs, don't they?

[Deli]

1 minute ago, Rauten said:

Doesn't LTT normally use the highest end ASUS board available? Those usually have overkill-level VRMs, don't they?

The Asus Hero isn't normally a cheap board. Only this time Asus gives it a cheap VRM solution. Well done, Asus.

[Sakkura]

23 minutes ago, Spotty said:

Response issued by LTT is pinned to the top of the comments for their video.

It's still worth pointing out (outside of a youtube comment nobody reads) they were getting VRM throttling. Their results are not really representative of what the 9900K is going to do with a proper motherboard.

 

On another note, LTT were also wrong about the percentage price difference between the 9900K and the 2700X and 8700K.

[schwellmo92]

4 minutes ago, Rauten said:

Doesn't LTT normally use the highest end ASUS board available? Those usually have overkill-level VRMs, don't they?

 

13 minutes ago, Spotty said:

Response issued by LTT is pinned to the top of the comments for their video.

Watch the video, Steve explains the board that LTT used was a 4 phase VRM which he had also tested a 4 phase VRM board himself and experienced thermal throttling with, so much so that the CPU stabilised at about 4.2GHz all-core boost (which would explain the lower power draw and thermals).

[Deli]

I want to see other tech channels do further testing to validate Hardware Unbox's result. So we can have a solid conclusion.

 

For all the tech channels. I trust der Bauer the most in term of CPU testing. His result is in line with Hardware Unbox's.

[schwellmo92]

Just now, Sakkura said:

It's still worth pointing out (outside of a youtube comment nobody reads) they were getting VRM throttling. Their results are not really representative of what the 9900K is going to do with a proper motherboard.

 

On another note, LTT were also wrong the percentage price difference between the 9900K and the 2700X and 8700K.

Well yeah the point was they didn't say (or maybe know) they were getting VRM throttling (assuming they were).

[porina]

So much data to analyse... if the LTT system was VRM throttling, would that hit the performance results? 

 

Also Anandtech had to update the power part of their review because their first mobo used over-volted.

[Morgan MLGman]

27 minutes ago, schwellmo92 said:

MODS: This is a video - do I still need to provide a quote? Do I just free-type it?

Just take a few key quotes from the video and put them in the OP. That'd suffice to meet the Tech News post format You should also write your short opinion at the end of the post.

Just now, porina said:

if the LTT system was VRM throttling, would that hit the performance results?

Seems to me that if the VRMs throttle, they lower the CPU frequency meaning lower performance, so potentially yeah.

[leadeater]

27 minutes ago, schwellmo92 said:

MODS: This is a video - do I still need to provide a quote? Do I just free-type it?

To quote a video you must record a video of yourself quoting the source video .

Yes that's a joke

[leadeater]

14 minutes ago, porina said:

So much data to analyse... if the LTT system was VRM throttling, would that hit the performance results? 

 

Also Anandtech had to update the power part of their review because their first mobo used over-volted.

The VRM wasn't throttling, the Asus Maximum XI Hero defaults to adhering to the Intel TDP spec so only allows short bursts above TDP then drops down to that. Gamers Nexus used the same motherboard and shows the same behavior in that video. All other motherboard sent out for review, the high end ones, default to removing the TDP limit which has actually been the standard for a very long time.

 

Basically all CPUs in the past have been reviewed with the Intel TDP limit removed, how the reviewers want to deal with that going forward is up to them. It's similar to the MCE situation, realistically everyone is going to have it on but how should a review handle that.

 

GN graph of package TDP over time:

[MMKing]

So my somewhat lengthy rants regarding boost performance only being available, provided the end user provides enough power, cooling and a suitable motherboard has not been without it's merits?

 

I've said it before and i will say it again. Intel can shout that their CPU is the worlds first 5GHZ all they want, without an actual guarantee however, it's completely meaningless.

 

@leadeater It should not be the standard. The motherboards are essentially running the CPU outside it's specifications. When testing the CPU, the stock settings should be strictly adhered to. If you're testing a motherboard however, you can run the CPU higher, as you are testing the motherboards ability to do so.

 

But 5GHZ boost is within specifications... but then you're outside TDP specifications. Basically, the 5GHZ boost and the 95W TDP are at odds with one another. However, 5GHZ is NOT guaranteed while 95W is guaranteed at stock 3.6GHZ all core clock.

[Sauron]

This has a huge FX 9590 vibe... it's a pretty big red flag when a chip VRM throttles at stock speeds on a supported motherboard (if that's actually the case).

 

If @leadeater is right, however, I think it's only fair to test it with the advertised TDP - that's what Intel's own marketing uses.

[leadeater]

14 minutes ago, schwellmo92 said:

Well yeah the point was they didn't say (or maybe know) they were getting VRM throttling (assuming they were).

It's not VRM throttling it's power target throttling, similar to Pascal/Turing GPU overclocking. At some point no matter how good the VRMs there is an imposed power limit you cannot exceed. For Intel CPUs you can actually configure these limits in multiple ways, max boost TDP and for how long the boost will last for and you can also remove the time limit and allow permanent boost within the set power target and thermal limits.

 

Example below my Asus Rampage IV Black Edition has set these CPU package power targets from the setting configured within the BIOS, the boost time limit is also disabled. (Picture is reused from a different FP post about this, ignore the red circles)

Spoiler

 

 

[mrthuvi]

How much performance are we sacrificing for how much cooler?

[leadeater]

17 minutes ago, MMKing said:

@leadeater It should not be the standard. The motherboards are essentially running the CPU outside it's specifications. When testing the CPU, the stock settings should be strictly adhered to. If you're testing a motherboard however, you can run the CPU higher, as you are testing the motherboards ability to do so.

 

Not sure if this is going to work but to quote myself from the FP post I made about this, I posted that video over there (let me know if you can't see it non FP members)

1 hour ago, leadeater said:

Wonder what the implications of this long term are for reviews, it's not often a widespread default out of spec configuration is highlighted with the last time being MCE. Should older CPUs be revisited, such as the 8700k and 7700k, with the TDP limits in place, should the 9900k be reviewed with the TDP limit removed or should both configurations be shown which would be rather messy. Maybe have the TDP limit in place and list results with limit removed as 'Overclocked' but with standard clocks along with OC results at higher clocks.

 

 

The problem is all the reviews in the past, as a reviewer you want to test things at stock but if you don't know the stock configuration of the motherboard is to take the CPU out of stock bounds then you just don't know. I'm not going to ride anyone hard on that because not knowing something is not something to punish people over.

 

I would however argue that for such a motherboard and the wide spread type of users likely to buy it having both MCE on and TDP limits removed or increased as default is correct and will be helpful to those users unaware or afraid to set them.

 

It's up to the reviewers to agree upon or disclose the settings used and explain their reasoning as to why, something displayed as stock really should be stock though.

[schwellmo92]

18 minutes ago, mrthuvi said:

How much performance are we sacrificing for how much cooler?

Honestly, not sacrificing much performance (5%~) considering how much lower the thermals and power consumption are.

[porina]

33 minutes ago, Morgan MLGman said:

Seems to me that if the VRMs throttle, they lower the CPU frequency meaning lower performance, so potentially yeah.

That was my thinking too, so not only would the power draw be difference but also the performance. Although VRMs may not be relevant any more for this case, given leadeater's posts since.

 

27 minutes ago, leadeater said:

It's similar to the MCE situation, realistically everyone is going to have it on but how should a review handle that.

Things like this make me happy I'm not a professional reviewer  Every extra variable has the potential to at least double the testing scenarios... 

[leadeater]

2 minutes ago, porina said:

Things like this make me happy I'm not a professional reviewer  Every extra variable has the potential to at least double the testing scenarios... 

Yep, and just when you think you've got it all sorted and have your testing procedures perfected you get slapped in the face by someone else's decision. For someone like Hardware Unboxed this is like a worst case because they do those horrifically long 35 game benchmarks and if you have to add in just one extra data point that could be day(s) extra work.

[asus killer]

how could thermals be so off if 9th gen is backwards compatible with 8th gen boards? 

 

still haven't seen any videos or know what boards were used, so...

[Dabombinable]

TBH, people should be glad that the 9900K didn't have its power consumption with the entire CPU (cores+IMC+iGPU) running at 100%. This is what a 4790K with my Gigabyte H97 HD3 with the core clock at 4.5GHz (1.2V) and iGPU at 1500MHz:

Spoiler

A stock 9900K in the same scenario would probably pull well over 300W.
 

[RejZoR]

It also depends on chip binning. My old 5820K has an incredibly low operating voltage. Running at overclocked 4.5GHz using AUTO voltage and it only runs at 1.180V. If I'm using stock clocks (3.6GHz) and AUTO, it'll run even lower. Given how these new ones turbo as far as they can go based on other parameters, it is quite important how CPU scales and at what voltage. Two 9900K probably hit lets say 4.8 or 5 GHz at different voltages in AUTO voltage mode.

[leadeater]

More interesting information. Remember to translate the page.

 

Quote

For our tests, we have the Core i9-9900K first operated with 95 watts, where he takes under load with Blender even after half an hour exactly this performance and remains cool - at reduced frequencies mind you. The used Noctua NH-U14S at full speed keeps the chip at low 67 degrees. But if we set the cTDP to 300 watts, the processor clocks all cores to 4.7 GHz, reaches a package power of 207 watts and heats up to 100 degrees and thus throttle limit.

 

It makes more sense, therefore, to configure the cTDP with 200 watts and thus achieve 194 watts at up to 93 degrees. So the chip does not overheat, because then it reduces the clock and slows down - alternatively there is the option to increase the temperature limit to 110 degrees. We consider this to be too high for continuous operation, or there are too few reserves. If you do not run the Core i9-9900K with a very strong air cooling system like the Noctua, you should set the cTDP to 180 watts or less to get a good mix of speed, power, and temperature

 

 

https://www.golem.de/news/core-i9-9900k-im-test-acht-verloetete-5-ghz-kerne-sind-extrem-1810-136974-4.html

 

Default setting is 95W as noted by this reviewer, they also go on to say anything below 207W will reduce performance/drop sustained all core clocks.

 

Edit: