While both AMD and Intel praised their new chips to heavens' high, reality seems to look a bit different: in a chat with the founder of GigaOM, A Facebook VP blasted AMD and Intel. Recently, AMD released its sexa-core Istanbul processor, promising easy upgradeability and roughly 50% increase in performance due to adding 50% more cores per socket. On the other side, Intel launched Nehalem-EP based Xeon 5500 prior to AMD's Opteron 2400, telling all the customers that the real evolution has come. We received some forwarded e-mails citing different sales pitches, and all in all, the situation look rather good for both companies.
However, when it comes to satisfying its largest customers, it seems that both AMD and Intel failed miserably. In his talk with the founder of GigaOM, Jonathan Heiliger, Vice President of Technical Operations on Facebook discussed the recent launch of Facebook Vanity, e.g. www.facebook.com/yournamehere project.
In a 14 minute interview, Jonathan stated that "the biggest thing… was less-than-anticipated performance gains from new micro-architectures, so new CPUs from guys like Intel and AMD. The performance gains they're touting in the press, we're not seeing in our applications."
Just like that, the bomb was dropped on the rosy statements coming from AMD and Intel. While we do not know how the Facebook code works and what type of applications are they running, it looks like guys from Facebook hit either Amdahl's Law or we have a case where the CPU cannot get or crunch enough data.
It looks like Facebook will have to wait for AMD to have deliver their promising Bulldozer architecture, while Intel's Sandy Bridge can't be that far behind. However, just to be on the safe side, we took a look at Facebook's lower tray, and the associated chat client - it seems that Facebook should work on getting their code in the shape, since we cannot and will not understand that a chat client with 50 people or more and five active chats will take 100% of one CPU core. Sorry guys, AMD and Intel may under-delivered, but your code isn't exactly "worth a million bucks".
As always, the truth lies somewhere in the middle.
