Intel CEO, Brian Krzanich, has made the bold prediction to analysts that it will have discrete graphics chips available in 2020. That’s only ten years after the last time Intel promised it would have a discrete graphics card shipped… and failed to deliver.
I’ve got to say, a 2020 target seems massively ambitious, especially with Raja Koduri, chief architect of its Core and Visual Computing Group, only starting out in his role around six months ago. Having a product ready to roll in two years might sound like a long time, but going from inception to manufacture to release in under three years is going to have to take some effort.
At the moment there are only two GPU players, and these are the best graphics cards around.
And that’s without considering the amount of time it might take for Intel to actually tune its GPU to actually be in any way competitive with whatever AMD and Nvidia will have on the shelves by that time. Intel may still just about hold sway in the CPU market, but they’re currently a long way behind in the graphics game.
By 2020 AMD will have a 7nm Navi GPU already on the shelves for a year, and potentially a more powerful refresh of the initial mainstream design ready to roll too. We might even have a GTX 1180 from Nvidia by then too.
The announcement from Krzanich came during an analyst event in early June, though whether the specific 2020 chip he was referencing is going to be the first gaming, AI, or data centre GPU we’re not sure. Intel hasn’t really spoken about a focus yet, just speaking in broad strokes, but it would make sense for it to focus on the big money AI space to start with and hope that will produce graphics silicon that’s competitive in the gaming market too.
Intel also followed up with a tweet featuring Raja’s beautiful, beaming face and a reassertion that its first discrete GPU was coming in 2020.
Now it’s possible that Intel could just take its existing GPU technology, the silicon currently baked into its Coffee Lake CPUs, and just throw more of its execution units into a discrete GPU package. That would give it a quick way to produce a graphics card that would at least have nominally competitive performance.
But the hardware is only one side of the argument - software, the drivers, are equally important in making sure that your GPU and your games play nicely together. And that’s going to require a lot of work on Intel’s side and it spending a lot of time working with game developers.
That almost means Intel can’t really take a particularly radical approach to its new GPU because they’ll need to be as familiar as possible to encourage developers to code for them. You can’t ask a triple-A developer, who’s already spending hundreds of millions of dollars, to devote another few million to coding specifically for a new set of graphics cards that are likely to have a very small install base.
So, how is Intel going to manage such a gargantuan software feat? Well, given its track record of hiring ex-AMD staffers we wouldn’t be surprised to see the next Intel GPU hire being Mr. Catalyst himself, Terry Makedon. He’s a passionate driver guy, and it’s going to have to be a passion project to get Intel’s graphics drivers up to spec in two years.