News Core i9-13900K Cooling Guide: Testing Intel's Flagship With Budget Air and Big AIOs

[Albert.Thomas]

Without going on "whatifsms", I will just trust Albert did not test this inside a closed case with no airflow. That would be kind of stupid.

Some more details would be welcome though.

Regards.

These tests were completed using a Cooler Master HAF 700 Berserker. The system fans were manually reduced to 35% speeds for acoustic reasons, but that still provides plenty of airflow.

 

[baboma]

Like this?

I didn't include the numbers for 125w and 200w for the different coolers because there won't be much variation at the same power limits. You'll see slightly higher numbers with the better coolers, but barely so.

I appreciate the additional info. Some observations and comments:

. A perf/power graph shows a CPU's sweet spot. As said, 90-95% perf for substantial power reduction seems a reasonable consensus. For 13900K, 90% is reached @ 200W, and 95% @ 250W.

. Given that 253W is 13900K's official recommended PL2, and therefore is likely to be adopted by many who enabled power limits, I think the CPU's official PL2 should be included in all perf/power/temp measurements.

. As a flagship CPU, 13900K gets most of the attention, but most people will buy a cheaper chip, especially those considering cheaper cooling solutions. I think people would appreciate this same piece repeated for the 13600K and 13700K.

 

[TerryLaze]

I appreciate the additional info. Some observations and comments:

. A perf/power graph shows a CPU's sweet spot. As said, 90-95% perf for substantial power reduction seems a reasonable consensus. For 13900K, 90% is reached @ 200W, and 95% @ 250W.

That's not what the sweet spot is, it is the point where power starts to increase much more than what the performance increase is.
In this case that would be 125W since an 80% increase in power to 200W only yields 15% in performance increase.
There are steps between 125 and 200 that are missing to give a better answer, with only this data 125W is the sweet spot though.

Lowering power by a giant amount and getting only 5-10% loss is a fact but it has nothing to do with the sweet spot.
(Other than getting closer to it)

 

[bit_user]

That's not what the sweet spot is, it is the point where power starts to increase much more than what the performance increase is.
In this case that would be 125W since an 80% increase in power to 200W only yields 15% in performance increase.
There are steps between 125 and 200 that are missing to give a better answer, with only this data 125W is the sweet spot though.

Lowering power by a giant amount and getting only 5-10% loss is a fact but it has nothing to do with the sweet spot.
(Other than getting closer to it)

You're defining the "sweet spot" as the peak in perf/W? That logically follows from power increasing faster than performance. The problem with that approach is that I'm sure 125 W is nowhere near the peak efficiency point, for these CPUs.

What I tend to look for is the "knee" in the curve, where efficiency starts to decrease at a much higher rate.

 

[-Fran-]

The "sweet spot" is hard to define/identify when you have a big portion of the "curve" that is basically linear scaling. If you want to be pedantic, most "sweet spots" are around 65W for the consumer CPUs, because both AMD and Intel are trying to target low power envelopes (as much as possible) for the Server and Mobile markets.

Regards.

 

[TerryLaze]

You're defining the "sweet spot" as the peak in perf/W? That logically follows from power increasing faster than performance. The problem with that approach is that I'm sure 125 W is nowhere near the peak efficiency point, for these CPUs.

What I tend to look for is the "knee" in the curve, where efficiency starts to decrease at a much higher rate.

No, I was trying to say the same thing as you, not the peak perf/w but the point where you have to use much more power for much lees perf.
Because as fran said 45W is the peak perf/w and for each "step" up you do need more power than the perf increases but after 125w the amount of more power far exceeds the increase in perf.

 

[-Fran-]

I think it's just better to call it "point before diminishing returns"?

Regards.

 

[TerryLaze]

I think it's just better to call it "point before diminishing returns"?

Regards.

That point is as soon as you power on the CPU...
at 45w you get 399 points/per watt and at 65 it's already down to 348 points.

 

[-Fran-]

That point is as soon as you power on the CPU...
at 45w you get 399 points/per watt and at 65 it's already down to 348 points.

If you define that points difference as "diminishing returns" comparing to another two power points, it fits what I said. You define those based on deltas between data points and how they project their differences (starts growing, remains the same or slows). This in optimization is the 1st derivative of the curve you're analyzing and 2nd derivative for "speed" of that change.

Regards.

 

[dalauder]

Depends on the power envelope and your cooling. With Cinebench the 13900k starts to beat the 7950x when it's allowed to go above a PL2 of 205 watts. Assuming you can cool it above 205 watts then the 13900k wins. If you can't then it loses.

Image source: https://www.pcworld.com/article/135...ive-into-13900k-power-use-and-efficiency.html

That's my point. All the articles say the i9-13900K dominates the R9 7950X. In this article, they even claim the i9-13900K doesn't lose much with worse cooling. My point is that if you've got average OEM cooling (like HP or something), your i9-13900K would actually lose in benchmarks to the 7950X. To me, that sounds like thee systems are a lot closer than the reviewers make it sound like.

 

[dalauder]

These tests were completed using a Cooler Master HAF 700 Berserker. The system fans were manually reduced to 35% speeds for acoustic reasons, but that still provides plenty of airflow.

So what if AMD and Intel both were tested in a case with mediocre airflow, like a common $75 case?

 

[Albert.Thomas]

That's my point. All the articles say the i9-13900K dominates the R9 7950X. In this article, they even claim the i9-13900K doesn't lose much with worse cooling. My point is that if you've got average OEM cooling (like HP or something), your i9-13900K would actually lose in benchmarks to the 7950X. To me, that sounds like thee systems are a lot closer than the reviewers make it sound like.

Has anyone tested the 7950x with a SFF cooler?

 

[dalauder]

You're defining the "sweet spot" as the peak in perf/W? That logically follows from power increasing faster than performance. The problem with that approach is that I'm sure 125 W is nowhere near the peak efficiency point, for these CPUs.

What I tend to look for is the "knee" in the curve, where efficiency starts to decrease at a much higher rate.

Ooh! Are we talking about second derivatives?!!

 

[dalauder]

Has anyone tested the 7950x with a SFF cooler?

In general, all tests are done with FANTASTIC cooling. That's what bugs me. Most people don't have that. Especially when the performance differences seem to not even show up until you go to 200W or even 300W CPU power. I want to see comparison benchmarks of OEM systems.

 

[dalauder]

...Overall, Intel came out better than what I would have expected to be honest, so I'd love to see those additions to this piece and then do the same testing on Ryzen and other Intel models (although I think it's fair to extrapolate from highest SKU).

Regards.

Well "fair" might be a stretch. If you actually take 12% performance off, Intel loses in every performance category. We REALLY need to see the follow-up on Ryzen performance.

But I want to see both in a cheaper case. Most people don't have a $300 case.

 

[Albert.Thomas]

In general, all tests are done with FANTASTIC cooling. That's what bugs me. Most people don't have that. Especially when the performance differences seem to not even show up until you go to 200W or even 300W CPU power. I want to see comparison benchmarks of OEM systems.

This article was delayed a little bit because I decided to include the SFF results, I had initially submitted the results without the AXP120-X67 .

After considering how well the Assassin X 120 R SE handled the i9-13900K, curiosity got the best of me and I setup the SFF cooler. I finished testing just in time to add it to the article 😉

I still find it hard to believe how well the weaker coolers handle the CPU when you stop looking at watts cooled and look at actual performance instead. It's certainly changed how I'll be approaching future reviews.

 

[Albert.Thomas]

But I want to see both in a cheaper case. Most people don't have a $300 case.

That's fair, but really - especially with the way I've configured the system fans - thermal results are comparable to what you'd see in smaller, more reasonably priced cases.

I've tested DeepCool's CK560 case with the same components otherwise, a case which is ~$90 USD, and it's results were only a few degrees worse. I should probably add that information to future reviews.

 

[TerryLaze]

That's my point. All the articles say the i9-13900K dominates the R9 7950X. In this article, they even claim the i9-13900K doesn't lose much with worse cooling. My point is that if you've got average OEM cooling (like HP or something), your i9-13900K would actually lose in benchmarks to the 7950X. To me, that sounds like thee systems are a lot closer than the reviewers make it sound like.

If you compare OEM systems then how good are the chances that the 7950x in an OEM system will have more than one stick of mem or good ram at all?!

You can push the intel farther than the ryzen, for you this might be a bug for a lot of people that is a feature.

Also even if intel loses in performance it still wins overall due to the price.

So what if AMD and Intel both were tested in a case with mediocre airflow, like a common $75 case?

Why? What do you expect to be different?!
You have several measurements at different TDPs, whatever you can provide in cooling in whatever case you have will determine the performance,noise,temps.
The only difference is that intel has an 10 degree higher limit before it starts to throttle compared to ryzen.

But I want to see both in a cheaper case. Most people don't have a $300 case.

Most people also don't have a $700 + CPU ...

 

[dalauder]

If you compare OEM systems then how good are the chances that the 7950x in an OEM system will have more than one stick of mem or good ram at all?!

You can push the intel farther than the ryzen, for you this might be a bug for a lot of people that is a feature.

Also even if intel loses in performance it still wins overall due to the price.

Why? What do you expect to be different?!
You have several measurements at different TDPs, whatever you can provide in cooling in whatever case you have will determine the performance,noise,temps.
The only difference is that intel has an 10 degree higher limit before it starts to throttle compared to ryzen.

Most people also don't have a $700 + CPU ...

I'm trying to respond to each of your questions/comments, but I'm biased so I might overlook some things.

From what I've seen, Intel chips the last three years or so have had more issues with heat than Ryzen and they're only compared when given optimal cooling conditions. This is one of the few investigations into Intel performance with less than optimal cooling. Three degrees could be the difference between throttling at 5.5GHz and 4.7GHz. That could be a 5% performance difference, which is less than the difference Raptor Lake is crowned "winner" over. This investigation already showed that Raptor Lake is significantly worse than Ryzen 7000 if it's not cooled properly.

If it doesn't make a difference, fine, but someone should at least compare common builds, not just with great cooling.

Plenty of people by boutique systems with poor cooling cases and buy a $700 CPU to go with their $300 graphics card in their "gaming" system. I'm sure you've seen tons of i9's pared with GTX 1060's or similar over the years. Plenty of people have expensive CPUs. If they didn't, we'd all still have an i5-6500 because it does everything 98% of users need.

That said, I'd prefer to see more midrange comparisons than we get.

 

[dalauder]

This article was delayed a little bit because I decided to include the SFF results, I had initially submitted the results without the AXP120-X67 .

After considering how well the Assassin X 120 R SE handled the i9-13900K, curiosity got the best of me and I setup the SFF cooler. I finished testing just in time to add it to the article 😉

I still find it hard to believe how well the weaker coolers handle the CPU when you stop looking at watts cooled and look at actual performance instead. It's certainly changed how I'll be approaching future reviews.

Thanks for replying to me directly. I hadn't realized you were the author.

I agree with the methodology of ignoring "Watts cooled" and just looking at performance for a given cooler. Then the thermals and throttling manage themselves realistically...so long as you allow a few heat cycles, which all of Tom's reviews have done the last decade.

I do want to see how Ryzen stacks up with the same coolers though.

As you stated, using a $90 case costs you a couple degrees. If Raptor Lake is already throttling, that could matter quite a bit. As awesome as the Haf 700 is, it's not a reasonable representation of a typical i9 build.

For example, HP is one of the biggest OEMs and their gaming line, the OMEN 40L achieves 10% lower CPU scores than benchmarks claim:

https://www.newsdirectory3.com/omen-40l-desktop-review-high-performance-gaming-pc-with-high-cost-performance-frequent-blog/

Here the OMEN 45L loses 17% performance (see multi-core benchmarks):
https://www.digitaltrends.com/computing/hp-omen-45l-review/#dt-heading-productivity-performance

I can't find any reviews of an AMD version, but at least on Reddit, it has heat headroom available for overclocking:
[URL="http://www.reddit.com/r/HPOmen/comments/vj3ohj/hp_omen_40l_amd_build_featuring_5800x_and_6600_xt/[/URL]"]www.reddit.com/r/HPOmen/comments/vj3ohj/hp_omen_40l_amd_build_featuring_5800x_and_6600_xt/[/URL]
But this Dutch site (yes, that's the best OEM Ryzen benchmark I could find) has Ryzen at about 11% slower than the Cinebench R23 Notebookcheck scores:
https://tweakers.net/productreview/287702/hp-omen-40l-desktop-gt21-0810nd.html

I think Intel throttles more than AMD and with any regular thermal limit, it doesn't beat AMD, they're about equal. For most people, that means choose either. For people who use all their PCI-e lanes like myself, it means pick AMD. I will concede that viewpoint when I see someone compare them.

 

[Albert.Thomas]

I do want to see how Ryzen stacks up with the same coolers though.

I'd love to see how coolers compare on Intel vs Ryzen, unfortunately I don't have a Ryzen system to test with at the moment.

As you stated, using a $90 case costs you a couple degrees. If Raptor Lake is already throttling, that could matter quite a bit. As awesome as the Haf 700 is, it's not a reasonable representation of a typical i9 build.

That's one of the reasons I've configured it with lower system fans, it's thermal performance in this limited scenario is comparable to more mainstream cases. Unless you're using a case with extremely poor airflow, results shouldn't vary too much.

As far as throttling goes, one of the reasons I tested a SFF cooler for this article was to test how much is lost - and really, there's not a lot of total performance lost with lesser coolers.

I think Intel throttles more than AMD and with any regular thermal limit, it doesn't beat AMD, they're about equal.

I feel modern Ryzen 7xxx CPUs will be more impacted than past generations by limited cooling, but given the CPU's lower overall wattage potential there shouldn't be as much potential performance lost from throttling.

For people who use all their PCI-e lanes like myself, it means pick AMD. I will concede that viewpoint when I see someone compare them.

I don't understand what you mean by PCI-e lanes here. How is this relevant to cooling?

 

[-Fran-]

Has anyone tested the 7950x with a SFF cooler?

I forgot to answer this one, but I see dalauder already replied. In any case, I've been trying to find more tests using different coolers, but most reviewers to fiddle around with the power targets while not changing the cooling, so no.

The closest one was Paul's Hardware when testing the Spire low end coolers with a 7600X.

Regards.

 

[TerryLaze]

but most reviewers to fiddle around with the power targets while not changing the cooling, so no.

It's the same thing as long as you know how much cooling a cooling solution can cool.
Looking up cooler reviews might be more helpful than searching for CPU reviews with certain coolers.

 

[-Fran-]

It's the same thing as long as you know how much cooling a cooling solution can cool.
Looking up cooler reviews might be more helpful than searching for CPU reviews with certain coolers.

No, it is not. Heat does affect conductivity and it is notorious in the resulting speeds vs power targets. Having superb cooling all the time will hide these inefficiencies when using lesser cooling.

Regards.

 

[TerryLaze]

No, it is not. Heat does affect conductivity and it is notorious in the resulting speeds vs power targets. Having superb cooling all the time will hide these inefficiencies when using lesser cooling.

Regards.

You only get an heat increase of the CPU if the cooler isn't capable of transferring it away from the CPU.
If you look at a 200W power limited review and use a 200W cooler at 100% you will never get any hotter than you already are, it will be able to constantly draw away the full 200W of heat that the CPU creates.
So yes, it is.