Conclusion

During the 14 nm era, Intel processors outshone AMD products by offering remarkable power efficiency in addition to superior performance, but this didn't last for long. AMD's recent success thanks to the wonderful Ryzen architecture, means that to maintain a meaningful bleeding edge over AMD Intel has had to resort to extremely high power limits, especially for the overclocked K models. In the case of the new Core i9-14900K "Raptor Lake Refresh" the power limit is as high as that of its predecessor, at 253 W maximum turbo power and PL1=PL2. Intel motherboards from decent vendors have had BIOS options for many years to control the power limits of these processors, and so we decided to see just how much performance is at stake if you tried to lower the power limits. If you're having a sense of Déjà vu reading this article, it's because we've already done such a battery of tests with the Core i9-12900K "Alder Lake," in which we found some interesting answers to the questions we're raising here.

Why would you do so? You might not be gaming all the time, or engaged in heavy productivity work all the time, and would want a daily driver setting where you lose an insignificant amount of performance for a tangible amount of power savings. Intel processors have excellent power management by themselves, and for light workloads, they automatically turn down and even clock-gate the P-cores. Thread Director ensures that the E-cores are made to deal with a workload first, and only when found too taxing for the E-cores, the P-cores are involved, which is the basic premise for Hybrid architecture. Having said all this, there is still scope for power savings by lowering the maximum turbo power value down from its default 253 W, just as there is a tiny amount of performance to be gained by unlocking this power limit—something most Z690 or Z790 motherboards allow you to do in their BIOS settings.

We chose six distinct power limit settings to put the i9-14900K through, 253 W (stock), 125 W, 95 W, 65 W, and 35 W. We also did one test run at 95 W with an additional undervolt, voltage set as low as we need to still run 100% stable. The first of course is the stock setting of 253 W, which will serve as our baseline for this article. This is when the processor is made to run as Intel intended. You can just refer to our main review of the processor for a more comprehensive look at how the i9-14900K performs out of the box, but put simply, it's Intel's fastest processor, and trades blows with the limited edition i9-13900KS that sells for $200 more than the i9-14900K.

The second power limit setting is 125 W. In this setting, the processor is told that both its PL1 and PL2 are no higher than 125 W, which also happens to be the on-paper processor base power value Intel advertises for its 14th Gen Core K-series SKUs. When averaged across all non-gaming applications, this setting sees the processor perform within 92% of its default 253 W setting. Most lightly threaded application workloads don't need to roast all 24 cores on the processors, and even if they need a few of the 8 available P-cores running, they end up close to the stock i9-14900K. An extreme illustration of this would be SuperPi or MP3 Encoding, which only load one P-core (the best one determined by Intel's preferred-core logic), and here there is no performance loss whatsoever, across all six of the power limit settings. This shows that it takes less than 35 W to run just one P-core, while leaving the others at their lowest C-state. Y-Cruncher stresses all CPU cores with a math-heavy workload, where the 125 W setting ends up around 12% slower than stock. This remains the case with Cinebench R23, where the multi-threaded test sees performance drop by 19%. Real world applications lose an assortment of performance percentages depending on just how parallelized their workloads are, but their averages work out to 92% the performance of a stock i9-14900K, which is pretty damn close.

95 W is the third power limit, and coincides with what used to be the power limit of Intel K-series processor SKUs once upon a time (read: a decade ago). Here, the performance loss is more pronounced, you get just 85% of the performance of a stock i9-14900K, which ends up slower than even the Core i7-13700K, and definitely slower than the i7-14700K. Multi-threaded data-heavy rendering workloads can lose as much as 28% performance, and there are further performance losses seen across the application workloads.

65 W was our next stop. It's important to pause for a moment and note that this 65 W setting is not reflective of the non-K processor that will come out soon, if the convention set by 13th- and 12th Gen is anything to go by—65 W is just the base power value, the i9-13900 non-K still enjoys a 219 W maximum turbo power, and ends up within 6% of the i9-13900K, but without the overclocking feature-set. You can always unlock the power limits of non-K chips in your motherboard BIOS to improve performance by a fair bit. So the 65 W hard-limit for the i9-14900K was particularly harsh, with the processor ending up at just 75% the performance level of the stock chip. Much of this is shored up by workloads that don't use more than a couple of P-cores, but those that utilize all cores, such as Y-cruncher or Cinebench, see their performance drop like a rock by over 50%. There simply isn't enough power for the processor to run all its cores at high enough clock speeds. 35 W was a setting we ran purely out of academic interest. It was to see just how much performance is lost for single-threaded workloads. This is barely enough power to run 1 or 2 P-cores at high clock speeds. The SuperPi result was pretty telling, the processor did not lose significant performance, nor did Cinebench single-threaded. The story repeats with similar real-world workloads such as web rendering, single-threaded media encoders such as MP3 LAME, etc. However, should an application need more than one core, the performance begins to tank.

As final data point we ran our "95 W" test, but spiced it up with some undervolting. Undervolting runs the processor at lower voltage, which reduces the power draw and heat output. This lower power draw means that the 95 W limit isn't hit as quickly during the same workloads, so the processor can boost higher, yielding additional performance. As mentioned before, we gradually lowered the voltage until the system got unstable. In the end we settled for a voltage offset of -0.125 V, which gave us perfect stability. For applications this made a noticeable difference, especially in heavily threaded tasks that hit the power limit fairly quickly.

Gaming is a whole different beast than application workloads. Nearly all gaming tests in our bench confine themselves to just the P-cores (as they should), gaming remains a workload that isn't too parallelized, and hardly a few games need 8 CPU cores (and their 16 threads). 720p is an academically relevant resolution that highlights CPU-level performance bottlenecks. Our high-end GPU can cut through 720p like a red-hot knife through butter, whatever performance differences you see are because the GPU is waiting for the CPU to perform its share of the game rendering workload. Here, the i9-14900K at 125 W loses a negligible 2.5% performance when averaged across all our game tests. Even 95 W + undervolt does reasonably well, losing just 4.5% compared to the stock i9-14900K. 95 W but with stock core voltage sees a 6% performance drop that puts it on par with Core i7 chips such as the i7-13700K. It's only with 65 W and 35 W where the major performance losses are, with 65 W posting a 15.5% performance loss, and 35 W hemorrhaging 45%.

1080p is our lowest real-world resolution that's relevant to the competitive e-sports crowd, where they'd pair 1080p with displays that have a crazy 480 Hz refresh rate. The performance losses are less pronounced for 125 W, 95 W + undervolt, and 95 W, which lose 2.5%, 4.5%, and 6.9%, respectively, however, when compared to 720p, the performance loss for the 65 W setting is a tiny bit more pronounced. 65 W ends up 15.7% slower than stock, and 35 W ends up 43% slower. 125 W ends up a decent setting for 1080p when paired with a high-end GPU like ours, you retain your high refresh rate for an insignificant performance loss, but a nice little efficiency gain.

1440p is where serious AAA gaming is at, and where our GPU begins to feel the heat, as the performance bottleneck shifts toward it. Here, the gap between stock at 125 W is narrowed to just 1.4% when averaged across all tests, and the processor remains faster than a stock i7-14700K that enjoys up to 253 W on tap. 125 W at 1440p is how you get to retain bragging rights for having the fastest Intel processor at lower power. The Ryzen 7800X3D remains the fastest processor at 1440p, ending up 3% faster. 4K Ultra HD is where the performance bottleneck is squarely in the hands of the GPU. Here, the 125 W, 95 W + undervolt, and 95 W settings are within 1% of the stock performance, so if you're gaming at 4K, any of these should do for you. The performance loss with 65 W is just 2.5%, however, this is where you should draw the line, as 35 W loses a whopping 22%. In real world terms, this is like dropping from 165 FPS to 135 FPS—noticeable but still not the end of the world.

This analysis is moot without a discussion on the power savings. 35 W is emerging to be a consensus for what it takes to run the i9-14900K with one P-core roasting with workload, this is confirmed with the stock, 125 W, 95 W + undervolt, 95 W, and 65 W settings. Interestingly, the power limits that we set in the processor's configuration barely correspond to the actual measured power values obtained using dedicated test equipment. For example, the 35 W package power limit setting actually tops out at merely 21 W actual power draw, which explains why it's getting hit with such performance penalties. 95 W reaches 91 W or 79 W, depending on whether the processor is undervolted or not. This shouldn't happen! By definition a power limit is independent of the actual voltage and shouldn't vary in such ways. Unlike NVIDIA GeForce cards, Intel processors don't come with actual power measurement capabilities and while NVIDIA strictly enforces the rules to avoid board partners cheating, there's several ways that motherboard vendors can "optimize" the power consumption values reported to the processor. Just to clarify, for our testing we turned off all performance enhancers like MCE and set the power limits manually. Interestingly, the 125 W power limit setting is spot on and matches our measurements exactly, whereas a limit of 253 W results in 282 W actual power draw.

Gaming power draw is vastly different from application power draw. At stock settings, the i9-14900K draws just 144 W, for which it has plenty of headroom within its stock 253 W limit, and can provide its P-cores with the best possible clock speeds. The 125 W setting (which provides within 96-97% the performance of stock), sees the power draw drop to 106 W, an excellent power saving, which makes running your SSDs and motherboard chipset "free." The 95 W setting sees power draw drop to just 80 W, and interestingly the 95 W + undervolt setting, which improves performance over the regular 95 W setting, ends up with just 69 W power draw. It's only with the 65 W and 35 W settings that power drops below the 50 W mark, but at great performance losses. The Ryzen 7 7800X3D draws just 49 W at stock settings for gaming, but hammers out performance on par, or better than a stock i9-14900K that pulls three times the power when gaming at stock.

All in all, we've had great fun playing with the power limits of the Core i9-14900K. The 125 W setting looks like an interesting daily driver. You remain with a processor that's faster than a stock i7-14700K at gaming, and the application performance loss isn't enormous. You can switch back to the 253 W power limits at will in the BIOS settings or through XTU/ThrottleStop, like for an intense video editing session where you need your processor to be at its best. The 95 W + undervolt config is an interesting setting to retain much of the gaming performance of the 125 W setting, even if at a more pronounced application performance loss. It's about 3 percentage points faster than 95 W, so undervolting is worth your time. After looking at the data I would probably pick 125 W + undervolt as my power-optimized settings for the 14900K. The 65 W and 35 W settings are a hard sell unless you're running your machine off a power backup solution and want to be absolutely sure there are no power draw spikes than can cut power. It could also an option for hot summer days, when you want minimal heat dumped into your room.

On the other hand, aren't we buying the 14900K for its amazing performance? Why sacrifice some of it for energy savings? This also explains why Intel had to give the 14900K such a high power limit setting. If they made it efficient, by lowering clocks/voltage/power limits it would perform similar to the 14700K, and lose the application performance duels to competing AMD CPUs like the 7950X and 7950X3D. It would also be too slow to justify its $590 price point vs the $410 14700K. It will be interesting to see how the non-K models do once they are released, they'll run lower clocks and lower power limit, which should help improve energy efficiency.