Announcement

Collapse
No announcement yet.

Report: Ryzen "Raven Ridge" APU Not Using HBM2 Memory

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    According to tests it still performs way better than any of Intel integrated GPUs (comparisons to dedicated chips is stupid) so I guess the HBM may have been unnecessary, but it is none the less disappointing to see this sticking to regular DDR4. Let's hope they have HBM-equipped models at some point in the pipeline and at least there's the recently unveiled MCM (multi chip module) product with Intel (thou it apparently uses a Polaris-based GPU).

    Comment


    • #12
      No idea why we are suddenly talking about HBM in a Vega APU. It was never supposed to have HBM. However, I completely agree with the single-channel memory being very unfortunate. What were they thinking? Hopefully a memory upgrade will be possible.

      I've been using Raven Ridge for quite a while now (an engineering sample though). The graphics support is there, but the full driver should be in kernel 4.15. For Mesa, you can choose between 17.1, 17.2, and 17.3. Go with 17.3 if you have the choice. LLVM 5.0 or later is required.

      There is the Intel CPU with an AMD GPU with HBM in a laptop coming up. It's big in the news and should satisfy the high-end segment.

      If you want the best integrated graphics with a Zen CPU, Raven is the choice.
      Last edited by marek; 23 November 2017, 09:28 AM.

      Comment


      • #13
        Thanks Marek for the update. Can you share at least how is RR APU GPU performance stacks vs the green/blue mobile solutions.

        Comment


        • #14
          Give me a Raven Ridge up to 45W (or until it saturates DDR4-2666) in a laptop. 15W? Meh.

          Comment


          • #15
            I have an Asus N56JN, with an 840M, and a 4th gen i7-4710HQ, it regularly thermally throttles itself. It's 15-47W for the CPU and another 15-25W for the 840M. If AMD can do better performance in 15W, that's a huge win compared to 30-75W.

            I'd buy a dual-channel version any day. It shouldn't matter that it's dedicated DDR4 for cache. AMD APUs before this didn't even have L3, much less L4 (if it isn't shared).

            Comment


            • #16
              Originally posted by marek View Post
              No idea why we are suddenly talking about HBM in a Vega APU. It was never supposed to have HBM. However, I completely agree with the single-channel memory being very unfortunate. What were they thinking? Hopefully a memory upgrade will be possible.
              Possibly, because a while ago there was talk about a compute focused "server APU" with many CPU cores plus GPU plus on package HBM. IIRC, that was purely speculative, though. Other than that, I've never heard of a APU with HBM. It might be a good idea in theory but such a device would have to compete with higher end dedicated solutions to justify the added cost.

              Comment


              • #17
                A poorly researched and/or disingenuous article. This is a 15W APU (let's be generous and suggest that it can cTDP up to 25W), it was never going to use anything apart from standard DDR4 system memory. It isn't going to outperform a discrete GPU that uses 20-30W on its own. The aim is to have better graphical performance than other 15W integrated GPUs (i.e., Intel), and so far that does seem to be the case, although benchmarks are thin on the ground overall. The single channel memory is a problem, as that may hold back the GPU in this case - other laptops are coming with dual-channel.

                Comment


                • #18
                  Originally posted by eydee View Post
                  When they demoed it, and showed playing a video instead of 3D graphics, people already suspected that something is fishy. Same as introducing Vega in CF, in a game with low requirements, without a fps counter.
                  There is nothing fishy about it, people were just holding out their usual too high hopes for an AMD product...

                  Bristol Ridge got mostly a "meh" reaction from most people, even though it was quite an impressive feat of engineering compared to its predecessors... but everybody was expecting a miracle, which was obviously unrealistic.

                  It's the same with Raven Ridge... there is a good reason for HBM2 only being included on top-shelf and enterprise focused products: it is scarce, and an exclusive new technology for which a lot of costumers are willing to pay a hefty premium. RR on the other hand is targeted at the value-oriented consumer market, where it simply would not be competitive with HBM2.
                  Also, the 512 and 640 shader IGPs in Raven Ridge will still be entry level performers, regardless of HBM2 being included or not. At best, you could get RX550-level performance with some HBM2. On the other hand, I think it can be hailed as a huge success if Vega 10 with dual channel DDR4 can perform close(~20-30% worse) to an MX150. I would expect at that point AMD would call it a day as well, and go off celebrating.

                  Based on previous experiences (28nm AMD IGPs vs 14nm Intel IGPs vs 28nm mobile nVidia chips), I do expect Vega 10 to outperform Intel IGPs by a huge margin. Also, based on the way Ryzen CPUs scale with voltage and clock frequency, I expect really good performance-per-Watt numbers.
                  The same goes for Vega 8/10: desktop Vega cards are operated on pretty much the edge of their efficiency, while the Vega IGPs target more of an optimal clock frequency. For instance you can increase power consumption on V56/V64 by hundreds of Watts even with some mild overclocking, while you can drop it by 20-30% by sacrificing ~5% of performance/frequency... imagine what a 20% drop in clock rates would do to Vega's efficiency! That's where these IGPs are headed.

                  Also, even previous AMD IGPs have reached roughly the same perf./W level as Intel's, even though they were manufactured on 28nm compared to Intel's 14nm. Now the playing field is more even in this respect.

                  Tl;dr: There is nothing wrong with Raven Ridge or Vega 8/10. The only problem is that people expect ~RX560-performance, even though RX550-level performance would be a huge step forward for IGPs and AMD.
                  I think at some later point during this APU generation's lifecycle AMD will make some high-end APUs with HBM2, similar to Intel's Iris Pro products.
                  Last edited by OneBitUser; 23 November 2017, 09:59 AM.

                  Comment


                  • #19
                    Originally posted by marek View Post
                    No idea why we are suddenly talking about HBM in a Vega APU.
                    Because the Intel+Vega APU will have HBM2. That means I'll have to buy Intel once again: not a very smart move from AMD.
                    ## VGA ##
                    AMD: X1950XTX, HD3870, HD5870
                    Intel: GMA45, HD3000 (Core i5 2500K)

                    Comment


                    • #20
                      1) It was clear a LONG time ago that it will use system memory only (though it's supposed to be shared, not dedicated so we should avoid double copy).. What made you think there would be any HBM2 model?
                      2) The HP Envy will have dual channel memory configuration.. They changed it after media backlash. See http://store.hp.com/us/en/pdp/hp-env...ouch-1za07av-1

                      Comment

                      Working...
                      X