Having just left the stage at AMD’s financial analyst day is CEO Dr. Lisa Su, who was on stage to present an update on AMD’s computing and graphic business. As AMD has already previously discussed their technology roadmaps over the next two years earlier in this presentation, we’ll jump right into the new material.

Not mentioned in AMD’s GPU roadmap but now being mentioned by Dr. Su is confirmation that AMD will be launching new desktop GPUs this quarter. AMD is not saying much about these new products quite yet, though based on their description it does sound like we’re looking at high-performance products (and for anyone asking, the picture of the card is a placeholder; AMD doesn’t want to show any pictures of the real product quite yet). These new products will support DirectX 12, though I will caution against confusing that with Feature Level 12_x support until we know more.

Meanwhile the big news here is that these forthcoming GPUs will be the first AMD GPUs to support High Bandwidth Memory. AMD’s GPU roadmap coyly labels this as a 2016 technology, but in fact it is coming to GPUs in 2015. The advantage of going with HBM at this time is that it will allow AMD to greatly increase their memory bandwidth capabilities while bringing down power consumption. Coupled with the fact that any new GPU from AMD should also include AMD’s latest color compression technology, and the implication is that the effective increase in memory bandwidth should be quite large. For AMD, they see this as being one of the keys of delivering better 4K performance along with better VR performance.

In the process AMD has also confirmed that these HBM-equipped GPUs will allow them to experiment with new form factors. By placing the memory on the same package as the GPU, AMD will be able to save space and produce smaller cards, which will allow them to produce designs other than the traditional large 10”+ cards that are typical of high-end video cards. AMD competitor NVIDIA has been working on HBM as well and has already shown off a test vehicle for one such card design, so we have reason to expect that AMD will be capable of something similar.


With apologies to AMD: NVIDIA’s Pascal Test Vehicle, An Example Of A Smaller, Non-Traditional Video Card Design

Finally, while talking about HBM on GPUs, AMD is also strongly hinting that they intend to bring HBM to other products as well. Given their product portfolio, we consider this to be a pretty transparent hint that the company wants to build HBM-equipped APUs. AMD’s APUs have traditionally struggled to reach peak performance due to their lack of memory bandwidth – 128-bit DDR3 only goes so far – so HBM would be a natural extension to APUs.

Comments Locked

146 Comments

View All Comments

  • Zefeh - Wednesday, May 6, 2015 - link

    If you haven't been keeping track of information on AMD's 390X, then you would be sorely mistaken. The HBM memory is giving the card around 640GB/s bandwidth at 1.25Ghz - That's DOUBLE that of the 290X 320Gb/s running GDDR5 at 7Ghz. Also, Benchmark rumor mill has the 390X running just above the Titan X in performance at ~$700 price point. Add in the fact that they are going to make a dual-GPU card using this chip is amazing. Watch out, because AMD has a HUGE powerhouse coming in thanks for HBM!
  • chizow - Wednesday, May 6, 2015 - link

    It's actually a full 1024GB/s of bandwidth, but bandwidth alone means nothing. Not much point in putting 20" racing tires on a hamster, for example. But keep hope alive! AMD needs it!
  • MisterAnon - Thursday, May 7, 2015 - link

    @chizow

    Considering a last generation part like the R9 290 matches Nvidia's current 970 in performance your comparison makes no sense.

    It just makes you look like a mad fanboy, but I don't blame you. I wouldn't be able to live with myself without being in perpetual denial either if I wasted money on a Titan X (lol).
  • chizow - Thursday, May 7, 2015 - link

    It actually makes plenty of sense, since the 290 is slower than the 970 and has more bandwidth already. So yeah, adding more bandwidth to a part that doesn't need it or can't make use of it, to any non-idiot, would be like putting more lipstick on a pig. But its OK, I'll let you AMD fanboys learn and be disappointed the hard way.

    And of course you woudln't be able to live with yourself, AMD fanboys are generally tech bottomfeeders, so yeah I know you wouldn't be able to wrap your mind around spending top dollar for the best components, and that likely won't change given the bargain perspective that is clearly most important to you.
  • MisterAnon - Thursday, May 7, 2015 - link

    You are one sad nvidia fanboy. Did his comment about you wasting money on a Titan X really trigger you that hard? Hit a little too deep?

    The performance of a R9 290 from 2013 is just about even with a 970 at 1080p and even greater at higher resolutions despite being a generation behind. If you actually wasted money on a Titan X I can see why you're so mad considering 400 dollar cards are about to come out that essentially match it in value and performance. Ouch.
  • chizow - Thursday, May 7, 2015 - link

    290 is slower and also uses nearly 2x the power but yeah, it is an older, inefficient part. The main benefit of the 970 was that it cut pricing on existing parts, including the 290/X and brought that level of performance down to a $330 price point. Big win for everyone, even AMD fanboys like yourself. How many 290/X do you own btw?

    I mean for $1000 you could buy maybe 3 or 4 bargain basement 290/X and Criss-Cross-Bonfire your way to victory! LMAO no thanks, the thought of dealing with CrossFire drivers and profiles that never get updated.... I'll pass, give me one of the best please: Titan X.
  • testbug00 - Thursday, May 7, 2015 - link

    290 uses about 70-80 more watts than a 970. Or, about 40-50% more. On average gaming load. Peak power while gaming is slightly higher difference.

    IF you run a power virus or heavy FP64 stuff, the 290 can approach 2x+ the power draw of the 970... But, it is also a lot faster than 2x in FP64.
  • chizow - Thursday, May 7, 2015 - link

    No, its more like 70% if you use 135W for the 970, so yeah nearly double. And the 970 is faster than the 290, closer to the 290X in performance but when it comes to power it really is double.
  • testbug00 - Friday, May 8, 2015 - link

    er, from what I see, under a gaming load the power draw from the 970 is closer to 150-160. The 290 closer to 235-245. That's about a 50% increase. Power viruses is 160-170 versus 310-320.

    A 980 uses approx 175-185. 290x about 255-265. Those are under gaming load.

    However, it is still a HUGE DIFFERENCE in LARGE favor of Nvidia. Good for their engineers (well, and their margins/stock price.)
  • chizow - Friday, May 8, 2015 - link

    Yeah again, you can spin the numbers however you like, but at the end of the day, 970 uses almost half the power of the 290, and is still faster than it. Pretty damn amazing!

Log in

Don't have an account? Sign up now