Technology & Positioning
With the Core Ultra 200 series, Intel is launching a brand-new processor architecture for their desktop lineup. While traditionally, Intel non-mobile CPUs have been monolithic, single-die designs, Arrow Lake changes this completely. The new CPU is made up of four independent silicon "tiles" that come together to form a single processor. This is somewhat similar to AMD's Ryzen approach, but much more complex, because the tiles are so different, each with its own responsibility, while AMD has just the I/O die and the compute tiles. While AMD connects their tiles using Infinity Fabric, Intel leverages their Foveros technology and puts them onto yet another piece of silicon, called "interposer," which allows much higher transfer rates, with more efficiency.

Another major architectural change is that Intel's new microarchitecture does away with Hyper-Threading. In the past, this technology was used to effectively double the processor's thread count, but over time it turned out that it is a major source of security issues. Also, some workloads run better without HT, and there is a die area cost of implementation. Last but certainly not least, the E-Cores have been revamped, too, they now have access to the shared L3 cache and come with a major increase in IPC.

In addition, the naming has been updated to indicate the evolution in Intel's architecture. Instead of Core i9-15900K, the processor in this review is called Core Ultra 9 285K. I'm not sure if that makes any difference for the average Joe on the street who thinks "higher number is better," and enthusiasts will just get used to the new naming, no matter the actual scheme. Today we not only have reviews for the flagship Core Ultra 9 285K (this review), but also for the Core i7-equivalent, the Core Ultra 7 265K, and the Core Ultra 5 245K, which replaces the Core i5 in Intel's lineup.

Application Performance
Intel has been praising their new Lion Cove architecture for months now, and we finally have our own results. Overall application performance of the Core Ultra 9 285K is improved by 1.2% over the Core i9-14900K—not much. AMD's flagship, the Ryzen 9 9950X still wins, by a small 3.4% margin. Once we take a look at individual benchmarks we can see why Arrow Lake isn't winning more battles. In several of our tests, the new processors lag seriously behind the competition, sometimes by significant amounts. I'm not exactly sure yet what causes it, but my first guess would be mis-scheduling of threads, either to E-Cores or bouncing between cores too often. Interestingly, this happens mostly in light and medium workloads. Heavy workloads run very well and show how mighty these new P-Cores are. I'm also impressed by the performance results of our CPU-only AI testing (without NPU), it's amazing how well this new architecture deals with this kind of workload, even without Hyper-Threading. Increasing the power limit really doesn't make any substantial difference for Arrow Lake—these processors are not limited by power.

DDR5 Memory & CUDIMM
With the performance results in mind, I'd like to highlight here that we are running at a relatively slow DDR-6000 CL36, which was picked to ensure fairness for all tested platforms from both AMD and Intel—a common denominator that all systems can run pretty easily. Now that DDR5 memory has matured we're looking into faster memory options for the next full retest of the 40+ CPUs in our test group. For Arrow Lake specifically, you have to consider that this platform has excellent memory support with speeds in excess of 8000 MT/s, without any additional dividers. On AMD, beyond 6400 MT/s you'll have to run at a lower memory controller ratio, which needs a significantly higher memory clock to make up for the increased latency. On Arrow Lake, a higher divider is only needed for 8800+ MT/s, which means memory scales much better. I am working on a full memory scaling article, but for now, I've included some quick testing in the review for you to get an idea what to expect. The numbers confirm that 6000 CL36 isn't monumentally slower than 6000 CL30, but that there are good gains to be had with memory clocked at the limit of what's possible today.

Gaming Performance
If you thought application performance was a mixed bag, then take a look at the gaming numbers. Depending on the game, Arrow Lake can be near the top of our charts (Spider-Man), or near the bottom (Elden Ring). This is quite surprising, because we all expected these new P-Cores to breeze through games, especially, since there's no Hyper-Threading, which often lowers gaming performance a bit. I have no idea what causes these swings in gaming performance, but hopefully it's something that Intel can address in the near future. Compared to last generation's Raptor Lake 14900K, the Ultra 9 285K is around 5% slower at 1080p, 1.5% at 4K. AMD's Zen 5 offers roughly the same performance, but AMD has the X3D ace up their sleeves. The 7800X3D remains the undisputed best gaming processor in the world, and the 9800X3D is launching very soon, too—unless AMD f'ed that up big time I am sure that it will take the gaming crown. Do consider these FPS differences though—5%—not much. At 100 FPS that's 5 FPS more—subjectively, you'll never notice any difference between these processors. They are all very good for gaming, and you should probably look at options with pricing in mind, too. Ryzen 9600X ($250), 7700X ($280), 14600K ($260), 13700K ($280) are all competitively priced with lots of choice in the segment—around $200 cheaper than the 7800X3D ($470), and I'm sure that AMD will make you pay for the 9800X3D.

Integrated Graphics
With the Core Ultra 200 Series, Intel is introducing a revamped iGPU, which is built on a more modern architecture. I'm happy to report that the performance has roughly doubled (!) compared to Raptor Lake—good job! While you still can't play AAA gaming with the integrated GPU, light gaming won't be a problem at all. You also get accelerated video playback with wide codec support and Quick Sync for an exceptional media encoding experience. Any sort of productivity, Internet browsing, Office, Photoshop will run perfectly fine of course, too.

Power Consumption
One of the main targets of Intel's new architecture is to bring down power consumption, and I can confirm that Intel has achieved that goal. On average, in our mix of 49 applications, the 285K consumed 132 W while the 14900K used 180 W—very nice, a 30% reduction. Gaming power is down a bit, too, by roughly 50 W, too, down to 94 W, the predecessor consumed 149 W. Compared to AMD's Zen 5 offerings, the power profile is comparable, "close enough" I would say. For gaming specifically, the Ryzen 7 7800X3D is the best choice for energy efficiency, though, using roughly half the power of the 285K while offering more FPS at the same time. You do have to consider that Intel is on a 3 nanometer process whereas AMD uses slightly less efficient 4 nanometer—both from TSMC. With our 2024 Test Suite we've added full system idle power consumption measurements, here Intel can impress with Arrow Lake, too, just like with their older architectures. Compared to AMD, the difference is around 25 W, which can add up when your PC is running many hours each day.

Cooling Requirements
In the past, Intel's high-end processors were difficult to cool—this is a total non-issue now. Thanks to improvements in energy efficiency the heat output is lower, and that heat isn't as concentrated anymore, because the P-Cores are spread out over a larger area. You can keep the 285K cool at stock easily, even with a standard high-end air cooler. Intel's decision to increase the thermal limit to 105°C allows for more headroom, while still maintaining a full warranty for 24/7 use. AMD uses a 95°C thermal limit, which paired with the thicker IHS makes their processors more difficult to cool, despite the slightly lower heat output. If you plan on serious overclocking, especially with voltage increases you'll definitely want good cooling, a decent AIO should do the trick here.

Overclocking
Just like in previous generations, Intel uses the "K" suffix to denote overclockable processors—is that really still needed? AMD gives you unlocked multipliers across the board. While overclocking is easy, and there are some new dials to play with, at the end of the day it really doesn't make much of a difference. I'm sure in the coming weeks we will learn more about how to boost performance. Interesting topics are D2D (die-to-die) and NGU (Next Generation Uncore), Ring, and of course everything related to thread scheduling and memory performance. The "classic" CPU multiplier increases won't make much of a difference, as the CPU is running pretty close to its maximum limits already, but per core overclocking can provide better than stock performance in all scenarios. I am working on more OC testing right now and will update this review ASAP.

Platform
Arrow Lake brings with it a new platform with a new chipset, new socket, and new motherboards. I really like that you can finally use a PCIe Gen 5 SSD without compromising on the GPU's PCIe lanes—with previous platforms the GPU dropped to x8 when a CPU-attached M.2 SSD was installed. Of course, you're also getting all the latest tech features like modern USB, Wi-Fi, Thunderbolt, etc. While in the past, such platform transitions have been very smooth with Intel, this time things felt much less refined. Throughout the last two weeks of testing, we have seen five separate BIOS versions, and I'm certain more will follow in the upcoming weeks. Personally, I encountered strange memory instability, where slots 1 and 3 would barely run at DDR5-5600. As soon as the modules were moved to the recommended slots 2 and 4, speeds of 8000+ MT/s were possible with the exact same settings and BIOS. Running four memory modules at high speeds is very difficult, you'll usually end up sub-6000 with four DIMMs, perhaps CUDIMMs will shake this up. Many new BIOS options were introduced, but the motherboards currently lack useful descriptions for these items in the UEFI settings, which will make you feel lost at first. When pairing Windows 24H2 with Arrow Lake, performance will be terrible—we've seen games running at 50% the FPS vs 23H2. One solution is to turn off Thread Director or disable the "Balanced" power profile, which is why we decided to use 23H2 for the time being. Last but not least, there are some driver issues and bluescreens when both a dGPU and iGPU are active at the same time.

To me, this experience felt a bit like AMD Ryzen 1st generation, but I'm confident that Intel can solve all these issues, like AMD did. While Raptor Lake supported both DDR4 and DDR5, Arrow Lake focuses on DDR5 exclusively, which makes a lot of sense to me now that memory prices have reached reasonable levels. Pairing the platform with faster memory definitely helps gain more performance, which is an advantage for Intel over AMD, because you can run faster memory with Intel (if you're willing to pay the premium for it).

We've seen very long boot times from AMD's AM5 processors in the past, and it seems that this is something that AMD can't really fix conclusively. On Core Ultra 200 POST times feel very similar to previous generations, "fast," no matter if booting from power disconnected or doing a cold or warm reboot.

NPU
Intel's new Arrow Lake processors all come with an NPU, which accelerates AI operations. Right now, this is a rare scenario—most AI experiences are cloud-based or run on the CPU/GPU. But going forward, I'm sure that we will see more and more vendors port their algorithms to run on NPUs, which could give such systems an advantage, mostly in terms of power efficiency. When it comes to raw compute horsepower, the GPU and CPU will have the upper hand for a long time, and the hardware/API ecosystem is quite fragmented right now. At this time, having an NPU in a desktop really makes no tangible difference, but this could change in the future.

Pricing & Alternatives
According to Intel, the Core Ultra 9 285K will sell for $590, which is the same launch price as last generation's Core i9-14900K. I'm definitely not impressed. Right now, there are so many performance gotchas with Arrow Lake that it's hard to recommend, especially when lacking a clear pricing advantage. No doubt, some application scenarios run really well on Core Ultra 200, and if you run these, do consider it, but for gamers there are lots of good alternatives to the Core Ultra 9 285K which not only are more affordable, but also offer (slightly) better performance. If you really want to play with Arrow Lake, do consider the Core Ultra 7 265K, which costs just $395, or $200 less, yet lacks only four E-Cores and runs slightly lower clocks overall. In terms of performance you're losing single digits in both apps and gaming, which isn't so bad considering an almost 50% difference in pricing. AMD's Zen 5 hasn't been selling well at all, because it doesn't offer significant improvements over Zen 4, and AMD just announced a price cut. I suspect that Core Ultra 200 will suffer the same fate. Upgrading really isn't attractive these days, especially not with those crazy motherboard prices. For the vast majority of people I would strongly suggest waiting a few months to ensure platform maturity.

Upcoming Releases
AMD has already declared that their new Ryzen 9800X3D is coming on November 7th, which will probably set new records for gaming performance, and it won't be cheap. In 2025, we expect Intel to announce non-K Arrow Lake processors, with lower TDPs at lower price points. I think at this time the platform will have matured a lot, and Microsoft will have had some time to fix Windows 11 24H2. Beyond that would be Zen 6 and Panther Lake, which will arrive in late 2025 at the earliest, so nothing to get excited about yet. We can only hope that the prices for the new Arrow Lake and Zen 5 platforms will decrease in the coming months, encouraging more customers to adopt them.

I mentioned this before, but Arrow Lake reminds me so much of Zen 1, first generation Ryzen. Looking back, somehow that feels like an opportunity, because it establishes a new baseline design that sets a foundation for the coming generations of Intel processors. For me, there's more coverage for Arrow Lake to work on: DDR5 Memory Scaling, P-Cores Only Performance and E-Cores Only Performance are my personal highlights that I'm looking forward to exploring, so stay tuned.