After a few years of remarkably similar Intel desktop CPUs, it's all change with the "Arrow Lake" Intel Core Ultra Desktop Processors (Series 2). We were sent an Intel Core Ultra 9 285K and an Intel Core Ultra 5 245K to test.
What happened to Series 1, you ask? Well, they were the Meteor Lake mobile processors launched in December 2023. These new desktop Arrow Lake processors, like their mobile predecessors, feature an onboard NPU or neural processing unit for AI operations.
The new processors use a new socket, LGA 1851, so the old motherboard that served you well for the 12th, 13th and 14th-generation Intel Core CPUs will not work with the Intel Core Ultra desktop processors. Say hello to the Intel 800-series chipset motherboards, the ones we were expecting for the 14th-generation chips.
The Arrow Lake Intel Core Ultra processors are a big change for desktop machines. In addition to the need for a new motherboard, DDR4 memory is no longer supported. Arrow Lake CPUs only support DDR5 memory modules, including the new DDR5 CUDIMM modules.
Plenty is going on with the actual chips' manufacturing, as well. Arrow Lake processors have the same P-core and E-core arrangement Intel has used since its 12th generation CPUs. P-cores, or performance cores, are the traditional high-powered cores that provide the computing power of a CPU. E-cores, or efficiency cores carry out the more mundane housekeeping tasks required of a CPU, freeing up the P-cores to do the heavy lifting. For these chips, instead of separating the cores, the E-cores are placed in between the P-cores to reduce latency.
Interestingly, the Core Ultra Series 2 processors do not use hyperthreading. This is a bid to reduce power consumption. So, the 285K with its 24 cores (8 P-cores and 16 E-cores) has 24 processor threads and the 245K, with 20 cores (8 P and 12 E) has 20 processor threads.
The 285K has a maximum P-core turbo frequency of 5.7GHz which drops to 5.4Ghz with 3+ active cores. By comparison, the 245K has a P-core frequency of up to 5.2GHz. Of course, these chips having the K suffix means that they are unlocked and thus able to be easily overclocked, netting a potential performance boost, but more on that later.
As well as the neural processing unit, both the 245K and 285K processors also have an Intel Xe graphics processing unit on board. It's of little use to gamers but could be handy for machines that need the computing power of the CPU, but not the graphics performance of a discrete GPU. The chips also have integrated Thunderbolt 4 and USB4 capabilities that can be utilised by the motherboard vendors.
I initially tested these processors back in late November. I was made aware that there were some concerns regarding the performance of the processors at launch. These issues would, apparently, be sorted out by driver and motherboard BIOS updates over the coming months. It's been a few months now, so I thought I'd see what the Core Ultra 9 285K and Core Ultra 5 245K were capable of.
I tested both processors in an MSI MEG Z890 Unify-X motherboard, cooled with a Corsair iCue Link Titan 240 RX RGB AIO cooler, paired with an AMD Radeon RX7800XT GPU.
The motherboard BIOS and Intel chipset drivers were all up to date as of 13th February. For both processors, I ran the following benchmarks: Bapco Crossmark, PCMark 10, Cinebench R23 and 3Dmark. Whilst a newer version of Cinebench is available, I have comparable data for Intel's 14th-generation CPUs across these tests.
The Crossmark results, the 2532 overall score, 2358 productivity, 2823 creativity and 2262 responsiveness all had the Core Ultra 285K scoring less than the Core i9-14900K. The results were not far behind, but still lower. It was almost the same with the PCMark 10 results, except for the digital content creation score which was substantially better for the 285K over the i9-14900K.
The Cinebench R23 single-core result had the 285K pretty much on parity with the i9-14900K, but the multi-core result was some 13% faster. It would seem that the 285K's extra horsepower from multiple cores being active is enough to top the results from the 14900K, which is good to see.
The 3Dmark Timespy test resulted in higher overall 3DMark and CPU scores for the 285K over the i9-14900K of 9757 and 14288, respectively. This is a 3% increase on the 3DMark score and 4% for the CPU score, and the best indicative test for gamers.
The results for the Core Ultra 5 245K were eerily similar compared to the Core i5-14600K. Again, the older processor performed better in the Crossmark tests and the PCMark 10 tests. The PCMark 10 digital content creation test, however, had the 245K beating both the i5-14600K and the i9-14900K. The same with Cinebench, with the multi-core test outperforming the i5-14600K, but a lower score for the single-core test.
The results show that there are some gains to be had with the newer processors when it comes to higher-end applications such as video/image editing that use multiple cores. The 3D Mark results suggest that gamers should actually see a marginal improvement with the new processors as well.
Overall, though, neither the Intel Core Ultra 9 285K nor the Core Ultra 5 245K perform favourably when compared to their predecessors, which is disappointing. It's still a bit too early to assess the viability of the on-chip NPU artificial intelligence technology, with a decent GPU being better placed to fulfil AI duties. This could change in the future, as processors hand over more background housekeeping tasks to AI.
Whilst it is difficult to get overly excited about the Intel Core Ultra 9 285K and Intel Core Ultra 5 245K, themselves, the new architecture has potential. But for this generation of Intel processors, it seems to be one step forward and two steps back.
The Core Ultra chips' energy efficiency improvements are obviously important for laptops, but likely of no immediate concern for the Intel Core Ultra Series 2 processors' intended audience. Someone running a gaming rig with an RTX 4090 humming inside is after higher frame rates, not lower power bills. But Intel needed to do something.
The past few generations of CPUs have been getting increasingly hotter. The revised architecture of the Arrow Lake processors represents a reset that should bode well for the future, if not right now.
I got a glimpse of this, as the processors running a little cooler made them easier to casually overclock using Intel's Extreme Tuning Utility. The overclock closed the gap a little between the 285K and 245K's synthetic benchmark results and those of their 14th-generation counterparts. I'm fairly confident that with decent cooling and a bit of patience, enthusiasts could squeeze quite a bit more performance out of the new processors.
The expense of a new motherboard and potentially having to upgrade to DDR5 for little to no performance improvement makes the Intel Core Ultra Series 2 desktop processors unappetising as an upgrade. That being said, the Core Ultra 9 285K result scored in the top 99% of PCs tested with PC Mark 10, so a new Arrow Lake-based gaming rig will still make you proud. For most of us, though, we'll likely be holding on to see what Intel come up with next, in the hope of some real performance boosts.
