The summary of Intel’s process-related roadmap, which was well documented by Dr. Ian Cutress over at Anandtech, is a slippery slope that continues to obfuscate the problems which continue to plague the company. Looking at the results of the Rocket Lake desktop platform, where Intel back ported a 10nm design to the long-in-the-tooth 14nm process node, many concessions had to be made. Reduction of the performance of the iGPU in Rocket Lake when transitioning to the older 14nm manufacturing process have also been benchmarked and proven by Dr. Cutress. When one combines the data, the outcome further amplifies how severely Intel has lost many of its prior advantages over the years.
With failed and devious marketing being present in most of Intel’s marketing materials, Pat Gelsinger had the option to come clean and alter the messaging. Something that encapsulates the key themes provided below would have gone a long way toward building trust and enhancing patience with customers that have been victims of the basket of lies during the Krzanich and Swan eras at Intel.
- We may not be the fastest right now, but our extremely optimized architecture is still competitive with market-leading solutions.
- We’ve proven that we can get more performance out of an existing design methodology through iteration and optimization. We may not have the truly smallest or densest process at a given node, yet improvements will continue to manifest as we further explore what’s possible per process.
- We’ll continue to compete even when fabrication doesn’t involve a like-for-like metric.
- We will not compromise on security for the sake of performance and marketing.
Instead of laying the cards on the table and allowing Intel to correct course over a multi-year period, the new roadmap obfuscates and misrepresents the manufacturing nodes which will be used for future platforms. The guy who called Apple a “lifestyle company” is now “innovating” by doubling down on bad branding. While the transparency is good, the fact that premium margins will be applied to process nodes that trail the capabilities of TSMC and Samsung is abhorrent. We doubt that equivalent transparency will be applied for potential acquisition targets SiFive and GlobalFoundries. There are many intelligent engineers at Intel that have developed viable solutions for multi-chip designs and vertically stacking elements. The shortened production runs for Kaby Lake-G (Intel CPU + AMD GPU via Co-EMIB) and Lakefield (first hybrid architecture released in 2019) highlight how this company is not optimizing the return on investment for groundbreaking designs. While Alder Lake is an evolution of the design and approach from Lakefield, there is still hard ceiling of eight maximum performance cores in the architecture. There was no effort placed on iterating on Lakefield to make it more competitive or compelling to use in a wider portfolio of products.