• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD announces CUDA support

not quite, they are releasing tools that make it easier to port CUDA code to be able to run on AMD GPU's

To be clear here, HIP is not a means for AMD GPUs to run compiled CUDA programs. CUDA is and remains an NVIDIA technology. But HIP is the means for source-to-source translation, so that developers will have a far easier time targeting AMD GPUs. Given that the HPC market is one where developers are typically writing all of their own code here anyhow and tweaking it for the specific architecture it’s meant to run on, a source-to-source translation covers most of AMD’s needs right there, and retains AMD’s ability to compile CUDA code from a high level where they can better optimize that code for their GPUs.
Click to expand...
 
Well yes and no - the ability to straight up compile CUDA code to run on AMD hardware will be there - don't expect to see games that use (compiled) CUDA for compute feature working on AMD any time soon though :S
 
shankly1985 said:
What does this give amd users?
Click to expand...

Do you code for HPC systems shanks?
No? Probably nothing then.

It's a set of tools for porting CUDA code to run on AMD (firepro) cards... If you don't currently write CUDA code then its of no interest to you.

@Rroff, yeah I noticed you can run CUDA on AMD using this, but I am willing to go out on a limb and suggest that you would be better off optimising specifically for AMD still, rather than taking a big old chunk of CUDA-optimised-for-nvidia and just running that on an AMD GPU... It is certainly a good step to give CUDA-like tools if they are serious about trying to stay in a market where CUDA is currently dominant
 
Last edited:
This is about as close to "Cuda support" as nVidia are to having freesync support on the basis you can alt tab and change the refresh rate in windows according to how busy that area of the game is.

It just means there is a tool to cross compile CUDA code, we know nothing of how efficient it is (or isn't) and it could take a major performance hit, just like emulation.
 
I'm sure someone (humbug?) was saying not too long ago that everyone is dropping CUDA for OpenCL these days.
So seems an odd move for AMD to make any attempt to support CUDA now that's OpenCL is about to kill it off.
 
GoogalyMoogaly said:
I'm sure someone (humbug?) was saying not too long ago that everyone is dropping CUDA for OpenCL these days.
So seems an odd move for AMD to make any attempt to support CUDA now that's OpenCL is about to kill it off.
Click to expand...

Because people aren't dropping CUDA for OpenCL - its a fallacy based on looking at the numbers from one perspective - OpenCL is seeing rapid uptake in userbase but its not directly correlating with a consequential decrease in the CUDA userbase.
 
Rroff said:
Because people aren't dropping CUDA for OpenCL - its a fallacy based on looking at the numbers from one perspective - OpenCL is seeing rapid uptake in userbase but its not directly correlating with a consequential decrease in the CUDA userbase.
Click to expand...

Tell that to adobe as well as apple. They waved goodbye to CUDA a long time ago. Almost anyone getting into GPGPU at this point will ignore CUDA completely since it cuts your market by a large amount, not to mention the fact that new NVIDIA cards have arse double precision performance, so less people will want to buy them for scientific applications, which is most of the point of GPGPU applications.
 
Meladath said:
Tell that to adobe as well as apple. They waved goodbye to CUDA a long time ago. Almost anyone getting into GPGPU at this point will ignore CUDA completely since it cuts your market by a large amount, not to mention the fact that new NVIDIA cards have arse double precision performance, so less people will want to buy them for scientific applications, which is most of the point of GPGPU applications.
Click to expand...

The larger amount of CUDA programming is for specific applications where they already have a hardware solution so it doesn't cut the market outside of a limited number of productivity apps which make up a tiny percentage of the CUDA userbase. Industry CUDA use will be for the most part combined with Quadro or Tesla hardware that isn't as gimped for DP as the GeForce parts - though Maxwell kind of sucks for that hence why I suspect "bigger" compute heavy Pascal parts will appear earlier rather than later.

Again its a fallacy connecting a few high profile software suites switching away to that it reflects a trend within the userbase itself.
 
Last edited:
Iv made use of CUDA in a number of image processing applications (mainly robotics type stuff)

All this means is there is a simple path for people to port these types of programmes to AMD hardware and probably allow applications to have 2 code paths (one for each set of hw) without too much ugly code (tho I haven't looked at the tool set / api that AMD are really providing yet)

Edit:

Also those who think OpenCL is being adopted instead of CUDA now is quite mistaken I think, CUDA is very popular as it's much simpler to use in most cases when compared to CL. Also like gaming nvidia inject a lot of cash into its partners using CUDA
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom