AMD's Radeon Vega GPU is headed everywhere, even to machine learning

You'll see it in even more laptops and desktops this year. 

393 Shares
Share
Tweet
Share
Save

    For 2018, AMD is building on its major innovations from last year: its new Ryzen CPU architecture and Radeon Vega graphics. Indeed, Vega is so versatile even its main competitor, Intel, has tapped it for its upcoming high-performance laptop chips. At CES, AMD revealed the Radeon Vega Mobile GPU, a dedicated graphics offering that'll go toe-to-toe with NVIDIA's GTX 10-series. Additionally, it'll be bringing Vega graphics into upcoming Ryzen desktop chips, as well as a new machine learning product.

    While we don't know much about the Radeon Vega Mobile GPU yet, it's not exactly a surprising announcement. Gamers have been waiting eagerly to see when AMD's new graphics hardware would make it into high-powered laptops. In October, the company revealed that Vega was coming to its new Ryzen mobile processors. It was only a matter of time until it had a more powerful dedicated offering. AMD is also positioning it as something you'd find in ultrathin notebooks, and not just chunky gaming machines.

    On the desktop side, you'll also find Radeon Vega in AMD's new Ryzen 5 2400G ($169) and Ryzen 3 2200G ($99) APUs. What's really intriguing about those is the amount of performance they pack into relatively inexpensive chips. AMD claims that the 2400G's 3D Mark Timespy benchmark score matched an Intel Core i5 8400, which retails for $199, paired together with an NVIDIA's $80 GT 1030. It's impressive that a $169 chip could easily compete with $280 worth of hardware from competitors -- at least, according to AMD.

    Looking ahead, you can also expect to see Radeon Vega in the Ryzen 3 mobile chip and a slate of Ryzen Pro APUs for enterprise and commercial users. If you're looking for a bigger architecture upgrade, AMD says it'll unveil the second generation of Ryzen desktop chips in April, and they'll also be the company's first 12nm products. Beyond that, the company revealed that its Zen 2 design is completed and expected to debut in 2019. AMD also has a followup to Radeon Vega in the works, Navi, which will be built on a 7nm process, and hinted that there's a next-gen architecture coming on 7nm+.

    Before AMD gets to those futuristic designs, though, it's starting out with the Radeon Instinct Vega, its first 7nm chip. Like the company's other Instinct products, it's a GPU tuned specifically for machine learning. It'll come with support for new software stacks and libraries that'll make it easier for developers to build machine learning applications. You'll also be able to port your apps developed for NVIDIA's CUDA hardware and, with only a bit of work, optimize it for Vega. One big difference between AMD's approach to machine learning and NVIDIA's? It's relying on open-source solutions, whereas NVIDIA tools are proprietary.

    Addressing the Meltdown and Spectre CPU vulnerabilities, AMD CTO Mark Papermaster reiterated, once again, that the company takes security seriously. While he was quick to point out Meltdown only affected Intel chips, he also said that the differences with AMD's architecture also "means there is near zero vulnerability" with the "Branch target injection" security hole. The other issue related to Spectre, "bounds check bypass," is mostly resolved with OS and software updates. AMD claims the performance loss from those patches is "negligible" -- but, of course, that could easily change with specialized use cases. While AMD is relatively safe for now, a future vulnerability is still possible.

    Click here to catch up on the latest news from CES 2018.

    From around the web

    ear iconeye icontext filevr