cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

sourcery
Journeyman III

Re: Suggest Feature you want in AMD APP

Posters have already suggested faster kernel launch, have an alternative suggestion that would work for me.

Launch a kernel that stays resident in the GPU,

a) Kernel waits for a change in a specified global memory block (to be loaded by CPU)

b) Kernel processes that global memory block then write a results global memory block.(to be read by the CPU)

c) loop back to a)

I am an Opencl novice but cant see a way to do the above with synchronisation or without putting the CPU and

and GPU in a loop, which would waste processing time in both.

The proposed pipes for Opencl V2..0 would work if they were between the CPU and GPU,

get the impression they will only be between GPU kernels though.

0 Likes
nou
Exemplar

Re: Suggest Feature you want in AMD APP

Meteorhead
Challenger

Re: Suggest Feature you want in AMD APP

Thanks for the tip. Nice article. Tried out on Ubuntu Server 13.10 from a pendrive on my laptop and worked like a charm. (Apart from conflicting with the Ubuntu repo opencv installation, which depends on nvidia-319-updates that installs half of an OpenCL runtime, which effectively kills all OpenCL apps)

I hope this fix gets incorporated into the next driver and I won't have to make these changes by hand on every driver release.

0 Likes
nou
Exemplar

Re: Suggest Feature you want in AMD APP

that is a bug in package dependencies. https://bugs.launchpad.net/ubuntu/+source/opencv/+bug/1245260 just install ocl-icd-libopencl1 before installing a libopencv-dev and you should be fine.

0 Likes
subaruwrc
Journeyman III

Re: Suggest Feature you want in AMD APP

The lack of overlapping compute/full-duplex streaming has been a deal-breaker for us. We regularly deal with data/matrices that are too big to fit onto a single GPU, unfortunately attempting to 'stripe' and stream the data onto one or more GPUs is slower than simply waiting for a multi-threaded version executing on CPUs.

Another major problem has been the inability to request more than a small percentage of the total available VRAM. For example, our 6GB FirePro cards are limited to effectively ~2GB, which renders them essentially useless, especially for a $5K GPU.

The Aparapi project also has multiple issues with OpenCL that need addressing.

0 Likes
wolf0
Adept I

Re: Suggest Feature you want in AMD APP

Oh, please, please let this happen. As well as a nice, complete document detailing syntax and such for GCN ASM.

sarobi
Adept III

Re: Suggest Feature you want in AMD APP


Meteorhead wrote:



Free fglrx from the clutches of Xserver. (It might be driver related)


At the moment linux driver is very intimately related to Xserver, but HPC applications would welcome the fact if the drivers could be loaded without a graphical interface having to run in the backgroud. I am only mentioning this, because I would like to integrate OpenCL into a system which is SLC based (Scientific Linux CERN), and this OS is a MUST. It is a minimalistic Red Had distro with many useful libraries included for scientific use, but the GUI is extremely unstable, and the monolithic grid infrastructure features worker nodes with GUI-less SLC. It would simplify things if drviers could be loaded similar to NV cards. This would free VRAM also, desktop wouldn't occupy memory, plus desktop rendering wouldn't hinder the default driver. ATM default adapter (utilized for desktop rendering) has to be AMD, otherwise fglrx fails to detect AMD GPUs.



I second that, it's on the top of my wish list.

0 Likes
joej
Adept I

Re: Suggest Feature you want in AMD APP

1. Async compute (execute different kernels simultaneously)

2. Smaller latencies for small data transfers and / or compute simultaneously

3. Make OpenCL 2.0 beta available to Win7 (8 makes me so angry)

The latencies are a big problem for game dev. I'm working on real time GI and multiple times per frame i have situations like:

Rebuild a tree - only one workgroup is busy at the top level - the rest of the gpu is bored.

Please let us know what will change about that and when.

Also let me know if Mantle will offer a compute alternative that may be better suited for the <16ms problem.

Keep up good work! R9280x is more than two times faster than GTX Titan on my algorithms, Yeah!

0 Likes