[x264-devel] Re: Speed-up method

[Tomas Carnecky]

Guillaume POIRIER wrote:
> On 3/3/07, Tomas Carnecky <tom at dbservice.com> wrote:
>> PCI-E has a very big bandwidth, even for GPU->CPU transport (maximum
>> with my 7800 is ~1150 MB/s).
> 
> But latency is very poor (compared to a HT link (AMD) or shared L2/FSB
> communication(intel) ). That can easily kill all speed-ups you gained
> from the GPU. You then need to let the GPU do a lot of work without
> having the data going back and forth between the CPU and GPU, which
> requires serious threading effort (it's not garanteed that x264's
> threading model maps too well on heterogenous archs like CPU+GPU).
> 

I was more thinking about letting the GPU do _all_ the work :P. In HDTV 
resolution (1920x1280, progressive), one picture is 3.6MB, so it would 
be possible to send ~310 frames/s to the card. That's more than enough.
The host application would then only read the raw bitstream from the card.

And btw, the GPU does int<->float conversions all the time. Maybe the 
GPU is faster than the CPU even when dealing with integers.

tom

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html