discuss-gnuradio
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Discuss-gnuradio] AMD HSA and digital signal processing


From: Tom Rondeau
Subject: Re: [Discuss-gnuradio] AMD HSA and digital signal processing
Date: Thu, 21 Apr 2016 17:16:59 -0400

On Thu, Apr 21, 2016 at 2:38 PM, Piotr Krysik <address@hidden> wrote:
Hi all,

Lately I came across something called HSA (Heterogeneous System Architecture) that is used in new AMD APU's (CPU's with integrated graphics processor). In general it enables CPU and GPU cores to share the same memory space. Exchange of data between them is therefore a matter of sending a pointer.

Many times I've heard that bottleneck of data transfer between CPU and GPU is main obstacle for application of massive GPU's processing power for signal processing. HSA seems to be the solution.

The idea is great but what surprises me is that although AMD processors with HSA are present for over two years I haven't heard of anyone using it.

GNU Radio could potentially benefit from HSA. But have anyone on the list used it? If yes - what was your experience?

P.S. I've sent the same e-mail few hours earlier but as of now it hasn't appeared on the list. I'm sorry in advance if the post is doubled.

--
Best Regards,
Piotr Krysik


Piotr,

Yes, I'm aware of the HSA efforts, and yes, there is some possibly significant advantages for GNU Radio here to make use of these kinds of co-processors.

One reason you haven't seen much discussion of this is because you have it slightly wrong about this being present in AMD processors for 2 years. While AMD has had the concepts there for a while, the HSA Foundation only recently certified their architecture, and AMD's processor Carrizo that came out last year (and has recently started showing up in off the shelf systems) is the first with the official HSA architecture. This info is straight from the HSA Foundation's president from a conference in mid March.

It can be difficult to get too excited about a technology like that which is still evolving and limited to a small subset of processors.

But it would be interesting now, for instance, to see someone look into experimenting with a "volk-hsa" that starts to see about taking advantage of the concept.

Tom


reply via email to

[Prev in Thread] Current Thread [Next in Thread]