There is a silver lining to Intel’s Sapphire delay

Avoidance of problems is a good thing

Xeon Bronze logoYesterday Intel ‘delayed’ their upcoming Sapphire Rapids server CPU, but is it all bad news? SemiAccurate may have rose colored glasses on but we see one big advantage to the delay.

You might recall that in the bygone days of yesterday, Intel realigned their server roadmap with what SemiAccurate has been telling you since 2019, Sapphire Rapids was a mid-2022 product. Who knew? Sarcasm aside, looking at the performance of Intel and AMD stock since that news broke, most of the world sees this delay differently from SemiAccurate. We see it as the potential start of a good trend, they see it more as a significant new competitive disadvantage.

Setting aside the fact that the real world dates for delivery of final Sapphire Rapids CPUs didn’t actually change much if at all, and Intel’s stock tanked, why is there a silver lining? It is a bit complex and starts with what was going to be shipped in 2021. The main impetus for the farcical promises of an early 2021 launch was the Aurora supercomputer. It used Sapphire CPUs coupled to Ponte Vecchio GPUs and had a hard delivery date of Q2/2021. Before you check your calendars, that is today and Intel won’t meet the deadline.

Contractual obligations and attendant kissing up to placate a second major miss aside, there was no way Intel would have met the date with release silicon Sapphire Rapids and they knew it in late 2019. Sure Sapphire did tape out in 2019, barely, but that plus the usual year of debug meant Q2/2021 was achievable, right? In the past the answer would have been a resounding yes but this time was different. Sapphire Rapids was in historically bad shape and the tapeout was not done for technical reasons, it was done for marketing reasons.

This is abjectly stupid for a number of reasons, and part of the festering malaise that was literally killing Intel. Tapeouts on 10nm are on the order of $10 million for mask sets, it takes large amounts of engineering manpower and time, plus a lot of otherwise unnecessary work. That time and compute power should be used to fix existing and known bugs, not to to meet arbitrary deadlines so someone can wave a chip at a photo op. Silicon does speed up some parts of testing and debugging so it isn’t all bad, but if you already have bugs you know about, proving them in silicon doesn’t move things forward in a useful way.

The only thing that could make matters worse is if Intel decided they needed something called Sapphire Rapids to plug in to Aurora in Q2/21 and planned multiple known incomplete and buggy tapeouts with lots of chicken bits to turn off broken features. This would cost tens of millions of dollars, add a lot of unnecessary work, and in general likely slow things down. No sane company would do this for a marketing ‘win’. Guess what Intel did?

If you think SemiAccurate is making this all up, a flight of fancy or a gross misinterpretation of the facts, there is one final data point to add. Intel said this directly in a memo sent to customers in late 2019, see the quote about half way down.

When you ship multiple steppings with fused off, incomplete, or broken features, you add a lot of work for OEMs and partners too. They aren’t happy either, essentially no one wins from this strategy other than those who’s bonuses are based on shipping X in time Y, nothing says X has to be complete or work.

So after spending a few hundred words trashing what Intel planned on doing for Sapphire Rapids, and minimizing the impact of the new ‘delay’, why did we say there was a silver lining? Because all this idiocy above isn’t going to happen now. Actually that is not true, but a lot of it went away, mostly on the external side, and likely a lot of it on the internal side too. Not making and shipping known bad steppings is a win/win for all on every front.

The ~1 year delay to Sapphire Rapids means that time can now be spent fixing issues before shipping parts to customers. Better yet those known broken parts that were deployed to meet arbitrary deadlines usually have contracts that specify replacement when the real part is done, hopefully before sketchy semi-accurate news outlets find out about the deals. The users get to keep the expense, ill-will, and overtime though. *COUGH*GOOGLE*COUGH*UPDATE SPACE*COUGH*

That all said, when Sapphire Rapids does finally ship, it will almost assuredly be final or something really close to final. Given the lessons learned with Skylake, Intel is unlikely to ship early Sapphire parts in significant volumes but small high profile designs like Aurora could get pre-release silcon, just not as emphatically ‘pre-‘ as it used to be. With any luck this isn’t a one-off event but the start of a new way of doing things at the new Intel. For those that have been around the industry for a while, you might recall this is also the old way of doing things, and it worked out very well then.S|A

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate