For quite a while, Intel has been facing challenges on multiple fronts. On one side there is AMD and their "Zen" architecture, which has shown impressive gains the company kept improving on relentlessly with "Zen+" and "Zen 2". Today, AMD is offering more cores at better pricing than Intel across all product ranges, and they're not done yet. The Ryzen 9 3950X will be the first 16-core/32-thread mainstream desktop processor, and new Threadrippers HEDT chips are coming, too. On the other side is Intel's own lack of innovation in silicon production tech. For over four years, they've been stuck on a 14 nanometer process which has been improved multiple times but is now starting to fall behind what TSMC offers with their 7 nanometer node. In order to regain competitiveness and strengthen their gaming leadership, Intel has been working hard to raise the clock speeds of the Core i9-9900K, because that's the only option they have at this time. Higher core counts will be introduced with Comet Lake next year, and a 10 nm or even 7 nm process is still several years away. The outcome is the limited edition Core i9-9900KS, which uses the same physical silicon as the i9-9900K and 5.00 GHz maximum boost, too. The secret sauce lies in the boost clock behavior of the CPU, which is now basically "all core 5 GHz at all times". The previous 9900K would boost to 5 GHz only with a single core active, and only clock at 4.4 GHz when all cores are loaded.

Looking at our performance results, we see impressive numbers from the Core i9-9900KS, especially considering it's an 8-core/16-thread CPU. In "general" applications, the Intel CPU beats AMD's offerings with ease because of the much higher clock speeds and a slightly better IPC rate. When looking at highly multi-threaded software that really loads all the cores all the time, AMD's Ryzen 9 3900X comes out ahead. Examples of such applications are rendering, media encoding, and benchmarks. Everything that's in-between—i.e., multi-threaded, but not with varying workloads—tends to be a slight win for the i9-9900KS or a tie. Examples of such workloads in our tests are virtualization, software compilation, and even machine-learning. Overall, we see the i9-9900KS around 10% faster than the Ryzen 7 3700X (which is 8-core/16-thread), too) and 6% slower than the Ryzen 9 3900X, which has 12 cores/24 threads, but is priced a little below the 9900KS. It's important to note here that all 3rd gen Ryzen chips in this review have been re-tested on the latest AGESA Combo PI 1.0.0.4B microcode. Compared to the Core i9-9900K, the KS is an impressive 10% faster, which seems surprising at first considering both processors are marketed with "5 GHz Boost". The difference lies in the boost clock behavior, as mentioned before. Whereas the Core i9-9900K will quickly dial back its boost frequencies when multiple-threads are active, the 9900KS will keep all threads going strong at 5 GHz. As expected, Intel's Core i9-9900KS further extends Intel's performance leadership in gaming, but the differences are small, especially at higher resolutions. The differences to AMD's processors are barely significant, so if you play 1440p or 4K, our benchmarks show no evidence that you should discard AMD's offerings from the start.

Just like the Core i9-9900K, the KS model has its TDP limit set too low to really handle 5 GHz at all times. Intel designed a clever way around that several years ago and continues using it in the i9-9900KS. When the CPU senses that it is running at higher power draw than the rated TDP, it will continue running at maximum speed for a short duration of time, while the heatsink can absorb the heat. After around 30–60 seconds, the TDP will get respected, though, and clocks drop to levels that ensure thermal sustainability. This is an excellent approach to handle short intensive bursts of CPU usage without driving heatsink requirements through the roof. With the default settings and multicore enhancement turned off, our ASRock motherboard disables this behavior, ensuring 5 GHz boost is active at all times at the cost of higher temperatures and power consumption. For our "stock" testing we reverted to Intel's defaults to ensure we present performance numbers that truly represent stock—values that will be achievable on all motherboards and with all cooling solutions. An additional data point records performance with fully relaxed turbo limits, which yields around 2% extra performance in applications; improvements in games are below 1%.

Single-threaded power consumption of the Core i9-9900KS is just as good as the 9900K, successfully defending the energy-efficiency throne. Multi-threaded efficiency is slightly worse than the i9-9900K, though. In this test, AMD Ryzen CPUs have the upper hand because of the higher core counts that are manufactured using a more efficient production process. Once we overclocked the CPU, all gloves were off—we measured 313 W system power consumption in Prime95.

While any decent high-end cooler should have no problems handling the heat output of the Core i9-9900KS at stock, things look very different once we start manual overclocking. Due to the massive heat output at 5.2 GHz, our CPU gets pretty close to thermal throttling at the voltage required for stability. With the Core i9-9900KS, Intel introduced an additional overclocking dial that lets you adjust the temperature at which the CPU should begin to throttle. Instead of 100°C, you can now go up to 115°C, which can provide additional headroom, even though I'm not sure running 115°C all the time is the best idea. Still, it can be useful during testing to figure out OC potential without being limited by heat, for example. If you plan on doing serious overclocking, watercooling is your best option. Overclocking potential on our retail CPU was decent considering we reached 5.2 GHz with ease and not too much voltage. Getting 5.3 GHz stable, on the other hand, wasn't possible, no matter the cooling or voltage.

Intel's Core i9-9900KS currently retails for around $525 at a few retailers, which is a $45 increase over the Core i9-9900K's current price. Considering we saw 10% performance gained in applications, that's not an unreasonable price increase. It's also a way to get higher performance than with the 9900K for people who aren't willing to spend the time overclocking their CPU. On the other hand, game performance isn't improved nearly as much. Here, I'd say 2% is what you should expect. We did see 5% better gaming performance at the academic resolution of 720p, but nobody is going to buy such an expensive CPU for 720p. I'd even say that a majority of the target audience for this processor will be gaming at 4K (not 1080p or 1440p), with the GPU being the bottleneck nearly all the time. In that scenario, I'm not sure if spending the extra money is worth it. Our performance numbers show that you can get comparable FPS rates with AMD's high-end processors and cheaper Intel SKUs, like the Core i7-9700K. Some launch-day reviews of the i9-9900KS used a $513 price for comparison, no idea why. That price is Intel's "1000-unit price to retailers" and not something you can expect to see in retail. Some big retailers like Newegg and Amazon have actually been jacking up prices quite a bit, offering the Core i9-9900KS for $600, which is simply too much and seems to be an early adopter's tax. For $150 more, AMD's 3950X, which goes on sale in two weeks from now, offers double the number of cores without compromising on clock speeds, and it is definitely worth checking out if your workload involves a lot of rendering and video editing.

Of course, we would have wished for more with this release, but I have to admit that yet again, Intel managed to squeeze more life out of their Coffee Lake architecture to keep consumers happy until Comet Lake arrives next year, which reportedly should help make up ground against AMD due to more cores and wider availability of HyperThreading. For gaming workloads, it seems unlikely that AMD's upcoming Ryzen 9 3950X and Threadripper can change things, but we don't doubt that the upcoming processors will work very well in highly threaded application workloads.