Archives Discussions

[oscarbarenys1]

Hi,

mainly here I'm providing jointly what I think could be good samples to be present in stream sdk showing how to use some advanced features of latest hardware (sorry some may have been asked for in some other questions by me):

*dual dma cayman usage example which show a nice perf improvement over a single dma card as evergreen

*a example showing concurrent kernels in cayman

*a example showing how to use zero copy (cpu mem acessible from gpu)

*microbench showing fusion improved bandwith between gpu and cpu vs discrete cards over pcie and also another microbench showing improved/reduced latency between cpu gpu improvements may be using some host sync between kernels..

*openvideo decode

*fusion efficient unified mem usage (maybe can be equal to zero copy sample)

*some example showing how to use fetch to lds to improvements in perf..

Thanks.

[moozoo]

I'd be happy if they just updated the "AMD Accelerated Parallel Processing OpenCL™ Programming Guide" to include Cayman information.

I'd really like to know the details behind this item in the Samples Release notes for 2.3

"Increased error tolerance due to reduced accuracy of SIN and COS instructions on the AMD
Radeon™ HD 6970 and AMD Radeon™ HD 6950 relative to earlier AMD GPUs"

My understanding is that GPGPU teams in both Nvidia and AMD have much smaller resources than the hardware and driver teams and as such are often well behind them in the release of information and samples.

I'm also wondering how much general interest in GPGPU computing might be waning as I've not seen any postings about 69xx GPGPU performance (other than anandtech SmallLuxGPU testing).

[ljbade]

Originally posted by: oscarbarenys1 Hi,

*openvideo decode

Check the page at http://developer.amd.com/gpu/ATIStreamSDK/samples/Pages/default.aspx as they have added a OpenVideo Decode sample in the last week.