Intel’s flagship $589 Core Ultra 9 285K headlines its new ‘Arrow Lake’ Core Ultra 200S series, leading the charge with 24 cores melded into a completely new chiplet architecture that comes with plenty of new leading-edge tech, like 3D Foveros packaging, support for new DDR5 CUDIMM memory tech, and the first dedicated AI engine fused inside a desktop PC chip.
Intel says Arrow Lake provides an up to 150W reduction in system power during gaming and other improvements, like a claimed 20% gain in threaded horsepower and a 5% gain in single-thread performance over the prior-gen, which helps offset the lack of gen-on-gen gaming gains.
The Intel launch comes on the heels of AMD’s tepid Zen 5 Ryzen 9000 launch, which saw AMD’s newest chips providing limited generational gaming improvements, so they couldn’t quite catch up to Intel. Naturally, given the performance we’ve seen with Intel’s new chips, AMD’s Zen 5 processors, which recently had pricing adjustments and firmware/OS enhancements, look much more promising than before — at least compared to Intel’s new chips. However, AMD also has its Ryzen 9000X3D processors slated for release early this month, and they will almost certainly be the new gaming performance champions.
That’s not to say that the Core Ultra 200S series doesn’t have its own charms. Intel employs a range of TSMC nodes for the different chiplets (called “tiles” in Intel parlance) in Arrow Lake. In fact, this marks Intel’s first mainstream desktop PC chip entirely fabricated using another company’s process node technology. Intel combines the more efficient process nodes with a radical new CPU core design that intersperses E-core clusters among the P-cores and discards Hyper-Threading entirely, thus claiming to deliver drastic power reductions that will result in a cooler and quieter PC.
Arrow Lake supports up to 192GB of DDR5 memory, but now in two flavors with two different base speeds. The chips support DDR5-6400 with DDR5 CUDIMMs, a new type of DIMM with an integrated clock driver (ckd) that boosts easily attainable stable clock frequencies by amplifying the signal, thus stabilizing the data eye. Unlike the clock redrivers present on fully-buffered registered DIMMs, the CUDIMM redrivers are said not to impose an additional clock cycle of latency (they use a less complicated and cheaper design).
Intel also points to much higher overclocking headroom with CUDIMMs and says DDR5-8000 appears to be the sweet spot (Gear 2). CUDIMMs should enable the use of poorer-quality DRAM ICs in higher-speed kits while simplifying the pricier DIMM PCB designs often required for higher-end memory. But motherboards with CUDIMM support may cost extra, and the CUDIMMs themselves are likely to carry a price premium, so you’ll need to pay close attention to the final cost before deciding whether CUDIMMs make sense.
Intel also supports standard DDR5, but at lower base speeds than it supports with CUDIMMs (the same DDR5-5600 as with its 14th Gen CPUs). Naturally, both types of memory are overclockable. Arrow Lake does support ECC memory, but it won’t be supported on consumer platforms — instead, that feature is reserved for enterprise-focused W-series motherboards.
[mai mult...]