AMD unveiled its 7nm Vega GPU for machine intelligence workloads at Computex. This revised version of Vega 10 — the high-end solution that AMD debuted at the end of summer back in 2017 — increases total onboard HBM2 to 32GB, along with a raft of unspecified changes that AMD obviously wants to keep under its hat for now.
The new Vega 7nm GPU wasn’t expected to show up quite this early. In fact, AMD is moving ahead of schedule with Vega, relative to its own previous plans. The original goal was to sample Vega in the back half of this year and to launch in early 2019. Instead, AMD is sampling now with launch expected later this year. This implies that TSMC’s 7nm roadmap is executing well, since the chip is being built in partnership with that foundry, as opposed to GlobalFoundries, which handles all of AMD’s 14nm GPUs.
AMD hasn’t confirmed any other aspects of the GPU design, but the expectation is that this core is built on four stacks of HBM2 memory, with a 4096-bit bus in total. That could put total RAM bandwidth for the card up near 1TB/s (RX Vega 64 has 483.8GB/s of memory bandwidth in two HBM2 stacks). Other questions, like core counts and capabilities, remain unanswered. But AMD is gunning hard to win space in a market dominated almost entirely by its major rivals, Intel and Nvidia, and it’s fighting back from a minority position. Expect the company to bring hardware to market as aggressively as possible to push back and gain share as a result.
As for when this technology will come to ordinary gaming systems, again, that’s not happening any time soon. AMD hasn’t made any adjustments to its roadmaps, which show consumer hardware still based on the RX Vega and RX 500 series of GPUs through the end of the year. Only in 2019, when 7nm manufacturing becomes more widely available and hopefully less expensive, will AMD plan to move to a new line of products. Nothing is known about next-generation Navi GPUs, including whether they’ll rely on GDDR6 or HBM2.
Financially, 7nm Vega isn’t going to be a major winner for AMD. The company is pushing into a new market with a new GPU ramping a new process node. None of those are financial positives, and AMD is taking on all three at once. But the high stakes around AI and machine learning have also emphasized the importance of getting hardware in-market. With Intel, Nvidia, IBM, and a hundred other firms all collectively chasing artificial intelligence, machine learning, and similar markets, AMD needed a product it could take to market to build a business of its own. Bringing out 7nm Vega months ahead of schedule gives the company an additional boost towards making that happen.
Google’s AutoML Creates Machine Learning Models Without Programming Experience
The gist of Cloud AutoML is that almost anyone can bring a catalog of images, import tags for the images, and create a functional machine learning model based on that.
Valve Removed Steam Machines From Its Home Page
You can still access what remains of the Steam Machine landing site via a direct link — not that you'll see much when you get there.
Valve Isn’t Done With Steam Machines After All
Valve developer Pierre-Loup Griffais claims Steam Machines are not dead, and the company has big plans for gaming on Linux and the associated hardware platform.
Google’s Cloud TPU Matches Volta in Machine Learning at Much Lower Prices
Google and Nvidia both offer competitive machine learning products — but Google is beating Nvidia on costs, at least in certain tests.