Some Parting Remarks

This isn't a conclusion since this article isn't a review, but a preview of the bits and pieces of information Intel allowed us to reveal. We wish we could post our entire review today, but that will have to wait until next week.

From what we can tell, based on what we can say today, Intel seems mighty confident that it has a product that can restore its performance and efficiency leadership in the market. The key to all this is a massive 28% IPC gain for the P-cores over Skylake and Intel not getting carried away with increasing core counts beyond eight to catch up to AMD, instead deploying its more efficient E cores. These little cores are no slouch and able to comprehensively increase multi-threaded performance, letting Intel catch up with AMD's processors made entirely of P-cores.

Much of this was possible because Intel finally got its foundry act together and is able to allocate enough 10 nm foundry capacity to desktop processors. The jump from 14 nm++ to 10 nm Enhanced SuperFin (aka "Intel 7") allows Intel to execute many of the key changes to its core design, letting it meet its IPC targets. It's also able to cushion these cores with plenty of middle- (L2) and last-level (L3) caches. With maximized turbo power values of 241 W and a base power value of 125 W, Intel seems to be maximizing the power headroom of the new node in a bid to catch up to AMD. So far, we haven't seen power numbers from Intel that wowed us; however, expectations from this architecture in idle and light desktop use cases are high since both rely heavily on the nimble E cores. If Intel does end up beating 16-core AMD Ryzens, a bulk of the credit must go to the team that designed Gracemont.

Platform I/O is another encouraging aspect about these chips. We now have next-gen connectivity, including PCIe Gen 5 and DDR5. Until last year, the company was still on PCIe Gen 3 + DDR4. It will be a little while before we see PCIe Gen 5 graphics cards that actually benefit from all that added bandwidth, but certain motherboards able to redirect some of those Gen 5 lanes to NVMe slots could offer users with a huge storage I/O edge. We've heard from SSD controller makers that they're expecting first PCIe 5.0 consumer SSDs in the second half of 2022. It's also encouraging to see plenty of PCIe Gen 4 connectivity, both from the processor and chipset, which will drive stuff such as NVMe RAID setups, Thunderbolt 4, USB 3.2x2, 10 GbE LAN, etc.

DDR5 will bring monumental changes to the memory sub-system, and a gen-on-gen doubling in data-rates is only the beginning. DDR5 introduces on-module power-management features, independent memory channels per module, intrinsic ECC, lower module voltages, and a doubling in generational DRAM density. 16 GB is the new 8 GB, and 32 GB is the new mainstream memory amount. 64 GB is the new premium memory amount, and you should finally be able to max out the 128 GB memory size limit of these processors using commonly-available UDIMMs. We're extremely happy that the transition to DDR5 can be painless, as customers may choose boards with DDR4 slots and retain their older memory, or choose from any of the inexpensive DDR4 memory options. Early adopters willing to play with the new tech will be able to find DDR5 from all the major vendors, though it seems the new memory modules won't be cheap.

Our performance reviews will be worth waiting for. They'll answer some fundamental questions, such as whether Intel managed to score a major technological leap over AMD and the wacky Hybrid core design has overwhelmed AMD's 16-core designs, and, if so, at what cost to efficiency. We'll also study just how this architecture affects performance of various application types while highlighting those use-cases where the approach is most optimal or sub-optimal.

See you on November 4.