Announcement

Collapse
No announcement yet.

DDR3-2400MHz On AMD's A10 Kaveri With Kingston's HyperX Beast

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

  • #31
    Originally posted by bridgman View Post
    Ahh, you think profoundWHALE was talking about single/dual rank rather than single/dual channel, ie saying that 2133 dual rank could outperform 2400 single rank with the same number of channels ? If so, that could make sense, at least on certain workloads -- seems that the Kaveri memory controller is able to interleave multiple memory banks on the same channel. AFAIK using 2 channels still makes the biggest difference though...
    It's posts like this that remind me I still know very little about everything and that my pursuit of knowledge will end in defeat.

    Comment


    • #32
      This stuff is a lot easier to understand if you grew up on mainframes where 4-way or 8-way interleaving of core memory was the only way for it to keep up with the CPU. The cool thing is that CPUs and GPUs are finally getting fast enough and wide enough that the same techniques are starting to be required again.

      You kids with your semiconductor RAM and flash drives have things so damn easy

      There is an exponential-growth thing to knowledge that makes it worth pursuing at all times.
      Last edited by bridgman; 03 June 2014, 05:22 PM.
      Test signature

      Comment


      • #33
        Originally posted by Kivada View Post
        Thats not what it means. The sticks themselves aren't dual channel, they are independent of that. the memory controller is what dictates how many memory channels are available. Most chipsets for the last decade have been dual channel, where if you have 2 sticks of ram with the same capacity, CAS and speed installed in the proper slots the controller will split up the data being split up and sent to the individual channels for increased bandwidth.

        It's the same thing you see with GPUs and their bit rate, the increased bitrates are more channels being added to the GPU's memory controller. On low end GPUs you'll see various memory setups on the same chip, these are because the manufacturers are allowed allot of leniency in what they use, most of them support 128-bit GDDR1-2-3-5 but can be found in a setup as low as 32 bit, which is caused by the manufacturer only installing higher capacity memory chips on one channel to cut costs.

        Now that said, AMD had previously stated that increasing the number of memory channels on a CPU only system results in increased latency and generally ends up as a wash performance wise. With a GPU in the mix I've got no idea if quad channel system ram would be of help for the APU as the increased bandwidth may help the GPU side but hurt the CPU side on latency.
        am a newbie, or rather a rusty oldbie, but given amdS multi core cpuS have little effect?, would scrapping say 2 x CPUs, allow room on the apu for a 2nd GPU?

        crossfire igp for processing down the track?

        Comment


        • #34
          Originally posted by msroadkill612 View Post
          am a newbie, or rather a rusty oldbie, but given amdS multi core cpuS have little effect?, would scrapping say 2 x CPUs, allow room on the apu for a 2nd GPU?

          crossfire igp for processing down the track?
          What would be the point in having two gpus on one die? It's much less efficient than one bigger gpu, and if you have the space for two, you can easily put one big.

          Comment


          • #35
            Originally posted by curaga View Post
            What would be the point in having two gpus on one die? It's much less efficient than one bigger gpu, and if you have the space for two, you can easily put one big.
            true. I suppose I was thinking of simply replicating what they are doing what they pretty much do now - crossfiring the apu gpu with an r7 discrete gpu. only on die. The cheap version has one igp neutered maybe?

            agree tho. You are right.

            Do you agree tho, that amd multi core cpuS have little effect? Threading isnt much good is it?

            Maybe the space could be better used by better igpS?

            Comment


            • #36
              Depends on what you do. For my use, many cores and good threading is essential.

              Comment

              Working...
              X