Tony Yu, CEO of ASUS China, has shared images of the Intel Core Ultra 9 285K seen at the silicon level, allowing us to appreciate it in more detail. all keys on the design level which Intel used in this new processor, which will be the top model of the Arrow Lake-S series.

The Intel Core Ultra 9 285K uses design based on blocks or tilesi.e. a chiplet design, which means it has abandoned the monolithic core design we saw in the Raptor Lake-S Refresh. At its core, the approach is similar to that used by AMD, but in practice the execution is very different.

The chips in the Ryzen processor are separated and interconnected horizontally through the Infinity Fabric system distributed throughout the substrate. This has its advantages as it makes scaling easier and lowers costs, but it also increases latency and negatively affects performance and bandwidth.

The chip giant didn’t separate or interconnect the chiplets in the same way as AMD, and I’ll explain it clearly with these images of the Intel Core Ultra 9 285K. The blocks or tiles of this processor are integrated side by side, leaving us with a design that could appear monolithic to the untrained eye.

We actually have a total of five functional tiles, a filler block that only provides structural strength, and a base block that all the others sit on. All these blocks are connected in parallel by the system EMIB (multi-dies interconnect bridge). Intel also uses Foveros to achieve 3D packaging.

The design that Intel uses is more advanced, more complex and more expensive, but it is It has its advantages. as it reduces latency and performance loss and also allows for more bandwidth. In the attached image you can see a graphically very clear representation with the layout of the different blocks that the Intel Core Ultra 9 285K has.

Intel Core Ultra 9 285K block division

Starting from top to bottom, we first have gpu blockwhich is produced in TSMC N5P node. This block uses the Intel Xe (Arc Alchemist) architecture, has a considerable size, but is obviously smaller than the SoC blockwhich is next on the list. This is produced in TSMC N6 node, It features 5600 MT/s DDR5 memory controllers, integrates NPU3 with up to 13 peak performance, and also has a PCIe Gen5 x16 interface for a dedicated GPU.

With Intel XMP 3.0 profiles it will be possible to use DDR5 memory at much higher speeds. DDR5 modules with speeds between 6000 and 7000 MT/s should work without problems and modules at speeds of 8,000 MT/s or more They shouldn’t cause any headaches if they come with properly tuned latencies.

He computing block is the next and biggest. Use it TSMC N3B nodeit is configured with 8 high-performance cores based on the Lion Cove architecture and 16 high-performance cores based on the Skymont architecture. These two types of cores are connected by a Ring Bus system and have L2 and L3 cache. These types of cores are distributed together, which improves communication and heat management.

We have it on the right I/O blockwhich is also produced in TSMC N6 nodeand which integrates input and output interface elements such as PCIe Gen5 x4 and Gen4 x4 subsystem for SSDs. Right next to it is a filler block that provides structural strength, and underneath all these blocks is the Intel 1227.1 base block.

That basic block is that enables interconnection every single one of the blocks we have seen and therefore plays a very important role because without it Intel would not be able to use this design and would be forced to connect each block through a substrate.

As I said at the beginning of the article, the way Intel has interconnected and packaged each of these blocks gives the impression that we are looking at a processor with a monolithic core design, and as I explained earlier, even this reduces classic problems derived from using chiplets under the model that AMD uses with Ryzen processors.

It kept a rectangular constructionwhich made it possible to use an IHS in the Intel Core Ultra 9 285K very similar to what we saw in the Intel Raptor Lake-S processors. This is important because it made it possible to maintain almost full compatibility with current cooling systems for the LGA 1700 socket.