Article: AMD's Mobile Strategy
By: Dan Fay (daniel.fay.delete@this.gmail.com), January 4, 2012 6:57 pm
Room: Moderated Discussions
Ricardo B (ricardo.b@xxxxx.xx) on 1/4/12 wrote:
---------------------------
>Dan Fay (daniel.fay@gmail.com) on 1/4/12 wrote:
>---------------------------
>>What I could see as possibly more realistic is for AMD to provide a fast, local
>>DRAM memory in the neighborhood of 256-512MB (perhaps something like GDDR5) on its
>>processors to improve embedded GPU performance. This memory could then be repurposed
>>on servers and/or HPC systems as a high-speed, local, general-purpose memory.
>
>On die?
I was thinking either on p (I guess, hypothetically, the extra channel(s) give you more memory capacity too).
ackage or die-stacked.
>And again, it's access time aren't that better so it doesn't help that much with latency bound single thread workloads.
True, but if it's essentially "free" (i.e., the NRE is already amortized from their Fusion products), then it might be okay. Bandwidth must still matter, otherwise SNB-E wouldn't have a quad-channel memory interface.
>And in terms of GPU performance, AMD has no problems.
They can always do better, however. Moreover, the lower burden of not having to drive long, multidrop traces might help out a bit with power consumption.
---------------------------
>Dan Fay (daniel.fay@gmail.com) on 1/4/12 wrote:
>---------------------------
>>What I could see as possibly more realistic is for AMD to provide a fast, local
>>DRAM memory in the neighborhood of 256-512MB (perhaps something like GDDR5) on its
>>processors to improve embedded GPU performance. This memory could then be repurposed
>>on servers and/or HPC systems as a high-speed, local, general-purpose memory.
>
>On die?
I was thinking either on p (I guess, hypothetically, the extra channel(s) give you more memory capacity too).
ackage or die-stacked.
>And again, it's access time aren't that better so it doesn't help that much with latency bound single thread workloads.
True, but if it's essentially "free" (i.e., the NRE is already amortized from their Fusion products), then it might be okay. Bandwidth must still matter, otherwise SNB-E wouldn't have a quad-channel memory interface.
>And in terms of GPU performance, AMD has no problems.
They can always do better, however. Moreover, the lower burden of not having to drive long, multidrop traces might help out a bit with power consumption.