How AMD fought its way back to relevance

Norrod says this attitude led to one of the key decisions that defined the first generation of EPYC chips - the decision to offer customers maximum value at the expense of sales margins. Intel, he explains "don't want to have customers buying one piece of silicon when they can sell them two", so AMD chose to take the opposite route. It offered customers an enterprise-ready single-socket processor, leveraging its impressive core density to do with one chip what Intel did with two.

Norrod also believed AMD could differentiate itself from Intel by its approach to security. "Intel had this SGX, which was designed - in my opinion - to enable Intel to get into the enterprise application business. Because if you get SGX security, you have to have the application right to SGX, you've got to rewrite your app to use it, and you have to distribute it on a secure route of trust distribution method, which Intel, out of the goodness of their heart, also happens to offer."

AMD decided to take a different approach with its Secure Encrypted Virtualization technology, Norrod says: "We were, like, 'OK, let's go offer similar or better security that's completely transparent to the application'. We can't make a dollar off of it."

At this point, Norrod says, AMD has a product it can take to Dell, HPE and "the cloud guys" - AWS, Google and Microsoft - which offers more cores per socket and differentiated security. It means cheaper purchase prices for the end-user companies and more profit for the vendors, as it frees up money which can be spent on software and solutions, which make more money for infrastructure vendors than the hardware itself.

"You can launch a single-socket server that generates the same level of performance at substantially better operating cost than the two-socket," Norrod explains, "and you can then upsell storage and other solutions on top of that, which are far higher margin for you."

When in Rome

AMD's next generation of data centre chip, known as 'Rome', looks set to introduce a number of technical benefits, but for the AMD management team this is far more than a chip refresh. "The importance of Rome is difficult to overstate," enthuses Norrod, "not just because it's a hell of a chip, but because it's the second one, and we're delivering it right when we said we would deliver it over two years ago."

Rome is based on TSMC's 7nm manufacturing process, which AMD says has given it a 50% increase in density, meaning twice as many transistors in the same physical space. Intel, meanwhile, is still working on a 10nm process. Its first 10nm chips were initially scheduled to launch in 2016, but have been pushed back repeatedly among numerous development setbacks.

"They're going to launch it, we think, end of next year, and the servers probably mid to second half of 2020," reveals Norrod. "It was a law of nature that Intel, who are the masters of process, had a one generation or greater process lead over everybody else. That was a law of nature, and that law of nature just got broken. And they're a really good team. It got broken because this is super, super, super hard."

"This stuff is incredibly hard," Norrod points out. "We're working on the absolute frontiers of physics here. The complexity of manufacturing these things is just incredible... My impression is they went just a tiny bit too far in a couple of aspects of the process design rules. And they just can't get it to work."

However, now that AMD has delivered two successive generations of chips on time with no hitches or setbacks, he explains, it demonstrates to OEMs and vendors that the company is serious about re-entering the server market, and it means that from now on, when AMD announce the release date for a new chip, its partners will believe it.

It's not just partners whose faith in AMD will be restored by this consistency, though Norrod admits that it has also reaffirmed the faith of the company itself. He tells us that, after the company's slightly ailing fortunes over the last few years, the staff have had to suffer through some hard times.

"The engineering team at AMD is very good, and they're very hardcore," he says. "Anybody that stayed in AMD through the lean years was a true believer, and you know, they were not well served by the management for many years. But they've done a hell of a job."

Tim Danton

Tim Danton is editor-in-chief of PC Pro, the UK's biggest selling IT monthly magazine. He specialises in reviews of laptops, desktop PCs and monitors, and is also author of a book called The Computers That Made Britain.

You can contact Tim directly at editor@pcpro.co.uk.