Advanced search

Message boards : Number crunching : The Radeon can officially host NVIDIA CUDA code

Author Message
Profile Zarck
Send message
Joined: 16 Aug 08
Posts: 145
Credit: 328,473,995
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45268 - Posted: 15 Nov 2016 | 15:56:47 UTC

http://www.anandtech.com/show/10831/amd-sc16-rocm-13-released-boltzmann-realized?_ga=1.109822469.1049115754.1472464840

@+
*_*
____________

[VENETO] boboviz
Send message
Joined: 10 Sep 10
Posts: 158
Credit: 388,132
RAC: 0
Level

Scientific publications
wat
Message 45303 - Posted: 17 Nov 2016 | 17:28:29 UTC - in response to Message 45268.

AMD has made good progress on the HIP front as well. HIP itself as it 1.0, and to demonstrate their progress, AMD is touting how easy it was to port the CAFFE deep learning framework from CUDA to HIP. Using the HIPify Tools for automatic conversion, AMD was able to translate 99.6% of the code automatically. This is admittedly less than 100% – AMD has been upfront before that they may always be playing catch-up and won’t necessarily be able to replicate NVIDIA driver features 1-for-1 – but it’s significant progress. Ultimately AMD was able to finish the feature-complete port in less than 4 days for a 55K line project


To be a beta version, it's great...

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45349 - Posted: 21 Nov 2016 | 12:52:39 UTC - in response to Message 45303.
Last modified: 21 Nov 2016 | 12:53:35 UTC

It's moved from dev to beta but it's likely that porting from CUDA to OpenCL/GL/HIP/CAFFE... will always result in some performance loss; NV design their GPU's to run CUDA while AMD are coding to make their GPU's compatible/porting via a toolkit from CUDA into their code to utilize their non-CUDA GPU designs.

CUDA is also in continuous development, so when trying to emulate CUDA, AMD will always be playing catch up.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

[VENETO] boboviz
Send message
Joined: 10 Sep 10
Posts: 158
Credit: 388,132
RAC: 0
Level

Scientific publications
wat
Message 45392 - Posted: 25 Nov 2016 | 10:45:32 UTC - in response to Message 45349.

CUDA is also in continuous development, so when trying to emulate CUDA, AMD will always be playing catch up.


The solution is to write an opencl native app for gpugrid :-P

3de64piB5uZAS6SUNt1GFDU9d...
Avatar
Send message
Joined: 20 Apr 15
Posts: 285
Credit: 1,102,216,607
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwat
Message 45393 - Posted: 25 Nov 2016 | 11:58:12 UTC - in response to Message 45392.
Last modified: 25 Nov 2016 | 11:59:36 UTC

CUDA is also in continuous development, so when trying to emulate CUDA, AMD will always be playing catch up.


The solution is to write an opencl native app for gpugrid :-P


+1
AMD Vega ante portas...
____________
I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

Profile Zarck
Send message
Joined: 16 Aug 08
Posts: 145
Credit: 328,473,995
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45723 - Posted: 14 Dec 2016 | 12:59:31 UTC

https://developer.amd.com/wordpress/media/2012/09/7637-HIP-Datasheet-V1_4-US-Letter.pdf

@+
*_*
____________

Profile Zarck
Send message
Joined: 16 Aug 08
Posts: 145
Credit: 328,473,995
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45738 - Posted: 15 Dec 2016 | 8:57:58 UTC

https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP

@+
*_*

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 511
Credit: 4,617,042,755
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 45741 - Posted: 15 Dec 2016 | 14:29:39 UTC - in response to Message 45349.

It's moved from dev to beta but it's likely that porting from CUDA to OpenCL/GL/HIP/CAFFE... will always result in some performance loss; NV design their GPU's to run CUDA while AMD are coding to make their GPU's compatible/porting via a toolkit from CUDA into their code to utilize their non-CUDA GPU designs.

CUDA is also in continuous development, so when trying to emulate CUDA, AMD will always be playing catch up.


It may be slower to emulate something but you would be gaining the potential power of all the AMD cards.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45750 - Posted: 15 Dec 2016 | 20:16:42 UTC - in response to Message 45741.
Last modified: 15 Dec 2016 | 20:18:25 UTC

At present there are a lot of people attached, not much work & POEM has ended. There isn't enough work to go around, and there are significant download problems. Developing a new app isn't the priority. New experiments + fixing the download problems (if possible) are the priorities. After that might be honing the existing app, project management and publications.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Post to thread

Message boards : Number crunching : The Radeon can officially host NVIDIA CUDA code

//