AMD's Threadripper line of high-end, workstation CPUs based on the Zen architecture wasn't originally part of the company's chip strategy, but a 'skunk works' project implemented by AMD engineers in their spare time.
That's according to Forbes, based on interviews with AMD staff working on the Ryzen project.
"It's not really a story of roadmaps and long-term planning or huge R&D budgets - it's a lot more personal than that and stemmed from a skunkworks project and a small group of AMD employees who had a vision of a processor they'd really want in terms of a high-performance PC," AMD's Sarah Youngbauer told Forbes.
She continued: "They worked on it in their spare time and it was really a passion project for about a year before they sought the green light from management, which is quite unusual - it was something they really cared about. The result, several years later was Ryzen Threadripper."
James Prior, AMD senior product manager continued: "Myself and a few others were in a very cross-functional team that get together for various different projects and as we got the first hints of what the Zen core performance and efficiency were like and started looking at the internal roadmap, which is a constantly changing thing and noticed a gap between Ryzen and Epyc."
However, a processor with more memory bandwidth, cores and PCI-E lanes, with the Zen architecture, shouldn't require too much modification, AMD engineers figured.
"So we put together this skunkworks team where we had platform architects, people that deal with core design, business unit, marketing team, to work out how to use what's already here and to go to the boss - Jim Anderson and say we'd like to do this," added Prior.
That was in 2015, when the Zen project was still in its early stages. As Zen developments progressed, the group working on Threadripper - the name they applied to the project, which was adopted by marketing - went with the grain, absorbing and working with the Zen technology, rather than rejecting elements and trying to create their own alternatives.
When AMD senior vice president and general manager Jim Anderson joined from Intel, and found out about the idea, he threw his weight behind it.
"I thought it was fantastic! I was really blown away when we worked out we could actually build it - I came into the industry as a CPU architect and it was the kind of product I'd love to build," Anderson told Forbes.
He continued: "I asked for two things - the specifications and also how quickly we could build it. I fell in love with it seeing the specifications, but the timeline was disappointing as it originally pointed at 2018 for launch. I immediately asked them to pull it in by about a year - so summer 2017."
Threadripper, he added, never had a business plan: "We were building it because we knew it was awesome, because we could and to make it best product we could, even the name had to be big."
In a bid to make the device in time, and at a reasonable cost, the engineers rejected the idea of making a larger die.
"The efficiency in manufacturing came from the fact that we didn't need entire wafer runs just to produce Threadrippers, we could use Ryzens… It meant we didn't need millions of dollars for this one design; we found a way to use our existing great product and make it even more powerful," said Prior.
"You can consider the resulting size of the CPU to be a negative aspect - the size of the socket, the heatspreader and number of pins, but it turns out it was a blessing in disguise. We could use the same socket as Epyc and just re-wire things. We'd already defined all the supply and ironed out the issues for Epyc, and this made it really easy to persuade our motherboard partners [to support Threadripper]."
14nm Cavium ThunderX2 CPUs deployed in HPE Apollo 70 supercomputer for US National Nuclear Security Administration
MWR's Countercept platform and phishd technologies key to F-Secure acquisition
Brexit labour shortages will lead to higher adoption of robotics
Newbies will be thrown in with the big boys on Sanhok as Kar98 fodder