No announcement yet.

Intel Announces CPU With HBM2 Memory & AMD Graphics

  • Filter
  • Time
  • Show
Clear All
new posts

  • #71
    Originally posted by Marc Driftmeyer View Post
    People seem to be oblivious of the Toshiba memory consortium purchase. SK Hynix and Apple [especially Apple] invested in getting that business for $18.1 Billion. The licensing between members for their portfolio and availability expansion of HBM2 seems rather timely concerning Apple's all-in with Vega and HBM2.

    More than anything, this is Intel licensing IP from AMD to keep Apple from dumping them completely, outside of LTE modem for wireless needs. This agreement most likely doesn't even exist w/o Zen/Vega/HBM2 and Apple iMac Pro and future Mac Pro having options other than Intel.

    Ultimately, with Thunderbolt opening up Apple can dump Intel all-together in 2018 and go Zen+/Zen2 Vega/Navi with their own custom boards adding TBolt 3 and never looking back.

    AMD wins either way. SK Hynix needs Apple's business, AMD needs Apple's business, Intel's leveraged the hell out of Apple's business for 10 years. Samsung just lost out to LG for Apple OLED in 2018 and beyond: a reason the CEO of Samsung retired and conceded the future more dim than their latest quarter benefiting from Apple's iPhone X contract and much more. Samsung jacked up all the prices on HBM2 as SK Hynix had not the resources to provide large quantities.

    AMD suffered, gaming suffered and only Miners got the bits available, at inflated prices.

    After the Toshiba deal, within 45 days we have all this news, AMD's GPGPU lines are MSRP again, availability for all is common and everyone is peeing themselves over this Intel announcement?

    The big deal is how AMD, SK Hynix, TSMC, Apple and anyone not Intel and Samsung fair, never mind Nvidia still a one trick pony no one cares about beyond their demoed AI smart systems and Gaming.

    Intel isn't worried about Nvidia taking them on in CPU market. They are worried Apple's ready for a divorce and AMD enjoys the double dating.
    Hope I can no longer boil an egg on AMD's CPUs . Glad to see some hardcore engineering leadership at AMD's helm.
    Last edited by MartinN; 07 November 2017, 03:43 PM.


    • #72
      I don't think, these chips will cannibalize Risen Mobile sales. As suggested before, the target market for these chips is vastly different. There are two clear indicators for this at the moment. First of all, manufacturing three chips and putting them on some kind of interconnect - no matter how cheap it might be - will always be a lot more expensive than an APU SoC like Ryzen Mobile. The only reason to go the multi-chip route is when the chips would otherwise reach ridiculously huge die sizes as is the case with EPYC where yields suffer greatly. Small mobile chips don't fall into this category. Also, HBM2 is not cheap. So, it is safe to assume that this Intel-AMD-Frankenstein will be an expensive product.
      The second and somewhat more obvious reason is that we know the CU count of both products. Ryzen Mobile will feature 8 or 10 CUs. The newly announced Intel-AMD-thingies are apparently going to include 24 CUs. That is another beast entirely.
      As it seems, Ryzen Mobile will be quite competitive in both CPU and GPU department. The only real worry is that system manufacturers will keep AMD's chips in low end systems with sub-par build quality and single channel memory.
      From my point of view, AMD just managed to tap into an additional piece of the market which earns them some extra money and also increases their relevance in developer mindsets. At the same time, it leaves Nvidia out in the cold for high end mobile systems that would otherwise have included an Intel CPU and some Nvidia mobile GPU. Seems like a good move to me.


      • #73
        Originally posted by arunbupathy View Post
        Whaaaat!!!!???? An AMD Radeon GPU on an Intel CPU?! But surely, I'am dreaming, right? I can't believe it is true.

        While it looks like it could bring AMD more visibility in the market, this also sounds so much confusing because they already have their own high performance CPUs which they could pair the Radeons with.

        But anything that is good for AMD is welcome by me! We need competition. I really hope that this is the fighting chance that AMD deserves. (Although, it was the same Intel who had fucked them over in the past.)

        I hope Lisa Su was right in saying that AMD are really back and that they are just getting started. I guess Raja Koduri too deserves praise here for the hard work RTG has put in to make this possible.

        Woohoo!! Great times to be living in as a PC enthusiast!
        Looks like he has left RTG!


        • #74
          Originally posted by Mabhatter View Post

          AMD doesn’t own Fab anymore so price isn’t as drastic a problem if they get a consistent royalty paycheck. And 3-7 years from one design and a few speed bumps is easy money.
          I'm really not certain it's that simple. Whether AMD owns the fab tech or GF owns it doesn't matter really. If problems with yield happen, as they have, it costs -much- more this way.


          • #75
            Hell has frozen over for one more time. Oh, wait, there was an early warning about upcoming climate change this time.


            • #76
              Originally posted by vito View Post
              Hopefully this makes Intel abandon their OpenCL implementation (i.e. Beigenet) in favor or something that works on Radeon (and Intel) GPUs. Perhaps it is time for Mesa Clover to be resurrected?

              Lack of decent and easily accessible OpenCL implementation on Radeon cards is the only reason why we still use Nvidia for number crunching at work. Getting ROCm to work on up to date distros is PITA.

              My hope is that this changes in the near future - hopefully once the kernel 4.15 is out, ROCm will work in userspace so it is easy to install on up to date distros like Fedora. Until then, the test Vega cards we got as a proof of concept are collecting dust in the bin.

              What do you use instead of OpenCL then?