AMD Roadmap Hints At Three GPU Launches By 2018

News
By Kevin Carbotte
published

(Image credit: AMD)

AMD revealed an updated roadmap for its future GPUs. The company has been criticized for being slow out of the gate with its upcoming Polaris launch, but if the roadmap is any indication, we’re about to see an acceleration in the company’s pace.

We already know that AMD plans to launch Polaris in the second half of this year. The company said that Polaris is expected to deliver up to 2.5x performance per watt, but the first products will be targeted at notebook computers. We don’t know when Polaris will hit the high-end desktop market, but the slide for AMD's roadmap seems to indicate that it could be short-lived if it does. 

The slide indicates that in the early months of 2017, we’ll see the first Vega GPUs equipped with HBM2 memory. The graph doesn’t map the curve, but given Vega’s position above Polaris, we can infer 3.5x to 4x performance per watt compared to 28nm GPUs.

Furthermore, AMD’s slide indicates that Navi, the successor to Vega, will offer even better efficiency and is expected to launch in the beginning of 2018. The slide indicates that Navi will employ some unknown “next gen memory” technology. The slide also states that Navi will feature scalability, but without context, it’s hard to infer what that would mean.

This is simply a roadmap, and as such is subject to change, but it gives us a little glimpse at what AMD is planning for the coming years.

Follow Kevin Carbotte @pumcypuhoy. Follow us on Facebook, Google+, RSS, Twitter and YouTube

Kevin Carbotte

 Kevin Carbotte is a contributing writer for Tom's Hardware who primarily covers VR and AR hardware. He has been writing for us for more than four years. 

See more GPUs News
60 Comments Comment from the forums
  • InvalidError
    I believe it is easy enough to figure out what "scalability" means: HBM pushed the memory interface width to the limits of what is practical to manufacture, HBM3 or whatever comes around after HBM2 will push data rates to the limits of what is electrically practical too. If you cannot push the bandwidth between a processor and its memory much further, the only direction left is parallel chips: high bandwidth interconnects between chips with an architecture optimized for distributed load sharing.
    Reply
  • junkeymonkey
    '' AMD Roadmap Hints At Three GPU Launches By 2018 ''

    so does this mean AMD's motto went from ''next year'' to now in ''2 years '' ?? lol......

    I was reading a article of amd and I laughed cause every third sentence it had next year bla bla bla next year ,ect ....... lol

    how bout something today or at least this year ??

    AMD, Enabling today.Inspiring tomorrow ,next year
    Reply
  • turkey3_scratch
    Junkeymonkey, I'd say AMD is not the only company getting slower. Intel keeps holding off Broadwell-E, they are now on a 2.5 year tick tock cycle. Nvidia's Pascal might release after Polaris cards, and what was initially thought of as Pascal cards in 2015, we're now looking realistically at late 2016. Every PC company seems to be getting slower, not just AMD.

    It also looks like now that HBM2 won't be featured in Polaris, bummer. I think Pascal is strictly using GDDR5X, maybe 1 HBM card.

    No way in heck do I expect Vega GPUs to be released early 2017. Expect late 2017, or early 2018.
    Reply
  • ern88
    I really want something this year that would be the same price of what I paid for my Sapphire HD7950 vapor x. And will give me about 50% plus performance of that card.
    Reply
  • voodoochicken
    I'm just bumming with on-board graphics of my z170 board I just got until SOMETHING gets released.
    Reply
  • junkeymonkey
    ''Intel keeps holding off Broadwell-E, they are now on a 2.5 year tick tock cycle. Nvidia's Pascal might release after Polaris cards,'

    well don't you got the great skylake and slap that Maxwell behind it ?? its not like AMD is pushing them. they will release things once they see its time to got more fresh money from you .. dang man they got to make there money and get all they can from you on what they offer you now first
    Reply
  • turkey3_scratch
    17665239 said:
    I really want something this year that would be the same price of what I paid for my Sapphire HD7950 vapor x. And will give me about 50% plus performance of that card.

    With a die shrink by a multiple of 2 we should be expecting hopefully 150% more performance IMO.
    Reply
  • heinlein
    "The company said that Polaris is expected to deliver up to 2.5x performance per watt, but the first products will be targeted at notebook computers."

    Unless a VR capable Polaris is available when I get my pre-ordered Oculus I will be buying a Nvidia card. I know others feel the same. Don't blow this opportunity AMD,
    Reply
  • turkey3_scratch
    I expect Polaris to come before Pascal.
    Reply
  • kcarbotte
    17665145 said:
    I believe it is easy enough to figure out what "scalability" means: HBM pushed the memory interface width to the limits of what is practical to manufacture, HBM3 or whatever comes around after HBM2 will push data rates to the limits of what is electrically practical too. If you cannot push the bandwidth between a processor and its memory much further, the only direction left is parallel chips: high bandwidth interconnects between chips with an architecture optimized for distributed load sharing.

    And you might be on to something, but I'm not about to put that in a news post. We're not in the business of rampant speculation. We take what's given to us and interpret what we have. estimating the performance number of a graph without labels is a far cry from making broad assumptions.
    Reply