It looks like Intel is not just delaying a single server project, their entire roadmap has just slid significantly. SemiAccurate is now worried about what we are hearing from the company, things do not appear to be getting better.
We told you about the latest near-term delay to an Intel server CPU yesterday, the word games about it’s predecessor, the previous delay to the platform first mentioned earlier, and many more. Some of these delays are pretty small and won’t have a big effect on customers, other are just the opposite.
Upping the TDP of Whitley last June is nothing short of catastrophic for customers, the change means you can’t reasonably cool the CPU with air. Water cooling isn’t the end of the world but it does entail platform changes, has strong resistance from some customers, and requires a new set of skills from those deploying it, all manageable things. Unfortunately Intel sprung this on customers 2Q before the (then) release date. That isn’t enough time for most to validate the platform much less design new chassis and cooling parts to support the higher TDP.
Why did Intel do this? AMD has 64 cores on their Rome CPUs, Intel was on track for ‘only’ 40-48 so TDP went up. This is the same thinking that resurrected Cascade Lake-AP to no positive acclaim. The only thing these dual-die parts are good for is grabbing headlines and lessening benchmark losses among potential customers that don’t read the basics of the specs. A 400W part with crippling restrictions, no 3rd party systems, and a price that Intel refuses to disclose is not a recipe for a sales winner. Raising Cooper Lake to 300W does the same thing to the mass market line.
It would be one thing if that bump pushed Cooper over the top and had it win some benchmarks but, well, it doesn’t. The 64-core AMD Rome based Epyc 7xx2 CPUs win almost everything handily even against the mythical dual die Cascade-AP line, if you take TDP into account things just get far worse. And as AMD has shown with the 32-core Threadripper 3, and the Epyc 7H12, AMD can up the TDP to 400W+ with ease and crush anything Intel can offer by an even wider margin. The fact that AMD has not responded to the TDP wars should tell you everything you need to know about the TAM of this space and market acceptance of the dual-die line.
That brings us back to today’s story, the changes to Intel’s roadmap other than the bits we told you about yesterday. Intel put some statements into the customer letter that were intended to sound soothing but as long as you don’t move your lips much when you read, were pretty transparent. The net effect was they horrified customers and OEMs which caused SemiAccurate to dig in to why those statements were made. The customers were right to be horrified, the dire picture we painted a year ago now looks rosy in comparison.
Note: The following is analysis for professional level subscribers only.
Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.