Intel Core i9-13900K Power Scaling Performance Explored

Does the Core i9-13900K really need to consume over 300W?

(This is a supplement to our initial Intel Core i9-13900K review.)

In the past, if you looked up the specifications of an Intel Core i9 processor, you would see a TDP (thermal design power) value near the bottom of the list, and at one point in history this was closely related to the amount of power that the processor would actually draw. There is a reason that Intel no longer uses a TDP value for its unlocked desktop CPUs.

Just read this definition of TDP from Intel:

TDP stands for Thermal Design Power, in watts, and refers to the power consumption under the maximum theoretical load. Power consumption is less than TDP under lower loads. The TDP is the maximum power that one should be designing the system for. This ensures operation to published specs under the maximum theoretical workload.

Obviously, the last couple of generations of Intel Core processors – at least at the high end – have not been limited to a set TDP. Intel last used such a number with the Core i9-11900K, which was given a 125 W TDP (configurable down to 95 W). However, based on the power draw numbers that we saw back when that processor launched, including a rise from 210W to 270W all-core after a performance-enhancing BIOS update, it was clear that most motherboard vendors were ignoring official power limits.

The Unlimited Power Era

Beginning with the Core i9-12900K, Intel moved to “Processor Base Power”, which is where we still find the old 125 W value, and Maximum Turbo Power, which for the Alder Lake Core i9 was 241 W. With the new Raptor Lake Core i9-13900K the Processor Base Power value is still 125 W, but Maximum Turbo Power has crept up to 253 W.

But there is another element that affects performance, and while thermals is the obvious choice here, there used to be a time limit in place to keep your cooler from becoming completely saturated. This is the Tau (recommended time limit) for the Maximum Turbo Power value. With the Core i9-13900K there is no Tau by default – just as with the Core i9-12900K before it (see power limit section of the AnandTech review for a good explanation of this).

By default, our MSI MAG Z690 TOMAHAWK WIFI motherboard gives the processor a 4096 W “limit” for both PL1 and PL2

Motherboards have the Tau option set to “auto” by default, and for the Alder Lake and Raptor Lake Core i9 processors this is infinite, unless you manually input something. The traditional Tau in the TDP era – the era of PL1 (base) and PL2 (boost) numbers – was 56 seconds. In other words, your CPU would drop from the boost power limit back down to the base power limit after 56 seconds.

As difficult a thermal load as 253 W is for most coolers, even with the traditional 56 seconds of a PL2 (boost) Tau, imagine running such a limit indefinitely. That’s the reality of running a Core i9-13900K with default / auto motherboard values.

Alternatives to Infinite Turbo Power Draw

Manually enforcing a more practical limit makes sense if you don’t have incredibly powerful CPU cooling, as, for the vast majority not running a custom loop, thermals are going to be the limiting factor in performance – at least for sustained all-core workloads.

While certainly not a unique idea (der8auer had results with a forced 90W limit on launch day), I thought that comparing benchmark results with previous-gen limits would be interesting. Thus, I present some comparisons with the processor limited to a maximum (including boost) of 125 W, which is a throwback to the old days of the hard TDP, and with the CPU set to a max of 241 W (the Core i9-12900K’s boost limit).

At this point I accept that Tau is no longer really “a thing”, though it would be interesting to test with a simple 56 second boost limit. Oh well, maybe next time.

First up, a check of single and multi-core performance at these power settings using Cinebench R23:

While performance within various power limits is going to be application-specific, I am content to look at these Cinebench averages. The results on the above chart are each the average of three separate runs, and as you can see the performance in the single-core test does seem to scale just a bit upward at the lower power limits, while clearly scaling downward in the multi-core tests – as Blender results demonstrate:

I think it’s fair to assume that single-core performance won’t be limited by a manual setting of even 125 W for PL1/PL2, as a single core just doesn’t use that much power. However, multi-core performance is naturally going to scale with power. I think it’s also fair to assume that, at some point, there will be an intersection between power and thermal limits with a given cooler.

Let’s look at some 3DMark results to see if these power limits affect performance with our GeForce RTX 4090 FE:

Well, considering the slight variance between tests (mitigated somewhat by averaging three runs for each result shown), it is hard to say that power limits affected the GPU score, even in the slightly more CPU-bound Time Spy tests. In the GPU-bound Speed Way test the results are even closer, so I think we can assume that in GPU-bound scenarios a low power limit changes nothing.

What about a “real” game result? Here is Metro Exodus, and these are results averaged from (once again) three runs at each resolution/setting:

Well, other than the result from the 241 W power limit in the 1280×720 / Low test, there is little to see here, though at 1080 / Normal there seemed to be a very slight (but consistent) scaling upward as the power limit went down.

Final Thoughts

One thing that became apparent as I saw some results that slightly favored running with a power limit, vs. without any limits: it is possible that our Core i9-13900K, when running without any limits and often peaking above 300 W, was thermally limited. Shocking, I know. I use a 360 mm AiO with the fans and pump at 100% at all times, but I think custom cooling would be required to eliminate possible thermal performance limitations.

I came to this rather obvious conclusion when considering that, if it is indeed possible to achieve higher performance in anything by lowering the power limit, there must be some thermal throttling at play – or at least less opportunistic boosting given the available thermal headroom in a given moment.

I think it’s safe to say that, other than the slower all-core performance when power limited, there seems to be little reason to run at unlimited power all the time. A power limit of 125 W provided the same gaming performance with a powerful GPU, and taking things up to 241 W restored all but a few percentage points of the unlimited performance level in multi-core workloads.

Intel was very successful with the Core i9-13900K launch, with charts all over YouTube and websites like this showcasing Intel’s return to the top of the charts in many instances. But after the benchmarking is done, and especially if you just want to play games, I think it’s ok to turn down the power. You probably won’t even notice – other than, you know, lower temps and a cheaper power bill next month.