Launches and Mirrors AMD EPYC Genoa and Intel Xeon Sapphire Rapids

What does it mean to “launch” a server CPU? What is General Availability or “GA” anyway? If these sound like philosophical questions more appropriate for Plato than the modern server industry, then it may be time to strap in. AMD EPYC Genoa and Intel Sapphire Rapids have “launch” dates two months apart, but what it means to launch is certainly fluid.

Launches and Mirrors AMD EPYC Genoa and Intel Xeon Sapphire Rapids

A few days ago, we discussed that AMD EPYC Genoa Launches 10 November 2022. We have been under embargo for some time for Genoa and will be bringing you a very in-depth launch day piece full of hands-on. We are the only site to have done both AMD EPYC 7003 and Intel Xeon Sapphire Rapids testing, albeit just being able to show you the accelerators at this point.

While I have a fairly good handle on how I think performance will pan out, given the fact that we have multiple unreleased systems, what I have less of today is an understanding of what is launched and what is considered “GA” silicon.

Here is what I can tell you at this point:

• Cloud providers have both Genoa and Sapphire Rapids. I think folks know this in the industry, but there is a different launch timeframe between when a hyper-scaler gets a new solution and when enterprises do. High-volume, lower-margin customers get serviced first, then other customer groups are added to increase margins.
• The Sapphire Rapids delay, from what I have heard, was something done for a late security challenge, as the company has stated. Intel has not gone into the vulnerability, but from what I have heard, I think that:
  • It was correct for Intel to delay.
  • It highlights Intel’s need to get more into a chiplet approach in the data center, well beyond what Sapphire Rapids will offer.
  • This delay happened very late, I think it happened after my Intel Sapphire Rapids at Intel Vision 2022 photos.

• Intel said that there will be a launch event for Sapphire Rapids on January 10, 2023.
• I will tell you I do not think we will be doing many SPR system reviews in January 2023, and that is not from a lack of desire. This may change, but STH’s review calendar has been full for 2022 since late August/ early September 2022 so we are already working on Q1 2023 content.
• On the Genoa side, readers started to pick up on the nugget in the HPE ProLiant Gen11 announcement yesterday. We cannot go into details, but I will say it is not HPE’s engineering that makes all of those features unavailable until 1H 2023.
• From a platform maturity perspective and from the systems we have seen, it actually feels like SPR platforms are more mature at this point, just waiting on chips.

These observations, in total, beg a few questions:

• When is a server CPU “launched”? Is it…
  • When it is shipped for revenue?
  • When cloud providers deploy it?
  • When a launch event happens?
  • When enterprise server vendors start taking orders?
  • When one can buy them through distributors so they are at “General Availability?”
• When is a server General Availability anyway? Is it…
  • When it is shipped for revenue?
  • When cloud providers deploy it?
  • When cloud instances are out of private beta?
  • When a launch event happens?
  • When enterprise server vendors start taking orders?
  • When promised hardware features are all enabled and able to be offered by server vendors?

Given how close we are to the Genoa launch, I have a pretty decent handle on what AMD will have on the above. For the Intel side, one of the interesting things is that the server platforms from enterprise vendors are relatively mature.

Here is one that will help frame the discussion, the QCT system that we used for Sapphire Rapids testing feels ready to go aside from production silicon. We can show you Sapphire Rapids chips and systems, but we cannot use any of our Genoa photos for this piece.

Final Words

The good news is that we should get a lot more clarity over the next 90 days. To those in the industry, many of the above questions will resonate. To those who have not been watching the delays and platforms mature, the above will likely make sense come January 10, 2023. One thing is perfectly clear at this point: a server CPU launch is very different today than it was pre-2016.