Hacker News Reader

> Customers which the article goes on to name as Google and Facebook. Although, I cannot find any corroborating evidence to support this claim. Supercomuting 16 is now over, and I see no other articles detailing how "Intel is playing favorites with Skylake Xeons in a way that will shatter the industry status quo." SemiAccurate is literally the only site I can find professing this scenario.

SemiAccurate has a pretty good track record with these kinds of rumors and analysis. They're not right 100% of the time, but they're right enough that it's clear they aren't just making things up.

One tidbit is Google themselves saying they'll roll skylake out in cloud early next year.

https://cloudplatform.googleblog.com/2016/11/power-up-your-G...

Looking at the article date (Nov 17) and the dates for Supercomputing '16 (ended Nov 18), if Intel had actually announced preferential release to specific customers, it would have been confirmed already. Combing through Google News, there doesn't appear to be any article covering Intel's announcements at SC16 that mentions such a preferential release. So, it's probably just a false rumor.

Would the screwed companies that had been promised these chips immediately go to the public?

And with Thanksgiving, I'm not sure there's been enough time for it to further leak out.

On the other hand, it's a claim that Intel will do something that's both extraordinary and extraordinarily stupid for the long term, so we indeed need some more proof.

Apparently not: https://newsroom.intel.com/press-kits/intel-at-sc16/

Although I would be really surprised if this was something they actually announced, it doesn't really help Intel to announce something like that and as the article points out it can anger other customers.

That said I have worked in Intel's System Validation group which was responsible for determining when a new chip was ready for general availability. I have directly witnessed Intel's preferential customer treatment where Intel allocated all of their processor production on a new processor to their largest customer(s). In my case they were selling Nehalem chips to Dell long (6 - 9 months) before others like NetApp could get them to start building filers with them. I have also worked in the platforms group at Google so I have an idea of the scale of their hardware effort and how many chips such a deal with Google might entail.

As a result, I think the actual story is this -

Intel got back first silicon that was acceptably functional on the new parts. This kicked off a discussion between their top customers and the potential early availability of parts. Those customers did a calculation on what advantage they might get by getting hardware early. Then Intel went down their customer list from most strategic to least strategic and offered them access. Either Google or Facebook was on the top of that list (likely Google) and they bought some, but not all of the production capacity, and then Intel offered the remainder to the second customer on the list who agreed to take the rest. Everyone else has to wait until the part is released to general production in all of its target fabs.

Further, it won't cause Intel any grief because all their customers know that the ones with the most volume get the most "love" from Intel. And the situation might be different the next time around.

I don't understand what the big deal is. Almost every integrated circuit manufacturer does this because supporting a few big customers is far cheaper than many smaller ones especially when you're operating on the cutting edge, a chronic condition for Intel. Google and Facebook will only call support when there is a discernable pattern of failures of processors or in previously verified code and that's the most optimal way for Intel to deal with early problems instead of handling thousands of support calls for intermittent failures in tiny sample sizes. The difference here is that microprocessors are a key piece of IT infrastructure and there are now a few providers that can soak up production of new products for months. For better or worse, economies of scale have always been a natural competitive advantage and Intel's dominant position in the server CPU market doesn't change that calculus, nor obligate them to make their products widely available when they first roll off the fab.

The big deal is the claim that companies that were promised access RSN to these chips will now have to wait ... 6 months, was it?

It's the breaking the promise more than anything else that problematic, if the sources, presumably from these screwed companies, are accurate.

Ah, I wasn't aware that Intel promised them early access. Did Intel break a contractual obligation or an implied agreement with previous partners? I'm unclear as to what promise means in this case.

I also wonder if this has anything to do with how Google and Facebook both make their own-ish servers ( https://code.facebook.com/posts/1711485769063510/facebook-s-... // https://www.google.com/amp/s/www.wired.com/2016/02/googles-h... ) which would potentially allow Intel more wiggle room with early release chips vs supplying all the chipsets and reference designs themselves. It could also just be the volume of FaceGoog. It's a pretty normal thing for a business to do. I don't really see the cause for outrage here. I do wish that Intel had more competition, and hopefully I'll be building a new gaming PC based around the AMD Zen SR7 when that comes out.