Nvidia Open Sources CUDA Platform

looks like CUDA is going open source. I Wonder where this will lead for us m17x users. here is the article I found the info from

 

this just means that Cuda will be useable on AMD as well as Intel gpus.

 

Bet nvidia gpu's will still be more expensive

Intel has gpu's besides the HD 3000/4000 ones?

 

so one less reason to opt for nVIDIA?

 

Well, least this'll lessen the gap in the everlasting war of green vs red.

 

^ green and red will now be friends?

 

BLASPHEMY!!!

No they don't... the integrated intel chips are what i was talking about

 

INCONCEIVABLE

 

there will still be a flame-war vs between the two of them for the foreseeable future. I guess the completion benefits us the consumers though

 

Solution: Use OpenCL. Then everyone is happy.

 

It's not open sourced. It ONLY, and ONLY means nVidia is using Open Source software, and will only allow SELECT PARTNERS and SELECT AMOUNTS OF CODE to be released to said developers, not unlike MS' Shared Code program. That is to ensure there is no AMD GPU port, yet still have enough flexability to make something like an ARM port.

 

This looks accurate.

However, the code they are open sourcing is the critical component... it's really not "select amounts of code", although it is "select developers".

The process goes like this:

1. you write cuda code using cuda coding guidelines and specifications (this has to be public so people can learn and write the code)

2. send your cuda code to a cuda code compiler, which converts your written instructions (readable by you) into machine instructions (readable by the computer, platform specific)

3. execute the platform specific code

the open sourced part 2... i'm not sure what else they could possibly do as far as making available. however- like you said, it's not really open source. It's "shared source". We almost certainly won't be seeing AMD support.

---

At the same time, we have OpenCL which delivers the same performance benefits, and can already run on x86 and Arm (CPUs and GPUs for both)...

 

There's already Engineering FEA software using CUDA - MSC Nastran 2012. Runs faster because of CUDA. CUDA and Physyx are the reason I only buy Nvidia vc's.

 

OpenCL is already surpassing Cuda, even though Cuda has been around much longer. No one in their right mind will start any long term software development projects today using cuda with OpenCL available. Projects that were already started before OpenCL came around, or projects that are backed by Nvidia, will continue to use Cuda.

PhysX is also pointless. Right around the time that hardware accelerated physics was possibly going to fill a niche in PC hardware requirements, ageia went through a major transition and was bought by Nvidia. Meanwhile, processor performance exploded right about the same time. Havok and bullet are cpu only and run very well. Physx is still only in a very small number of games, and an even smaller number where it's available to run on the GPU, and no cases where it is used for anything more than to tack on a special effect or two.

Meanwhile, havok is awesome, and bullet is awesome, and non-physx physics engines are being used to create crazy dynamic interactive environments (Crysis 1). Bullet is also getting OpenCL support.

 

http://www.maximumpc.com/article/columns/nvidia_it’s_time_kill_cuda

Looks like Nvidia is listening.

 

Didn't ATI (Or AMD, guess it comes under the same roof now) already have Stream? Makes me wonder why that wasn't publicized more... All I've ever really seen in the news is about CUDA.

 

They did. It was originally Brook or CTM which was much lower level than CUDA (or OpenCL, etc.). They dropped support for it (but you can still use it) and changed over to OpenCL (the Stream SDK became the APP SDK).

Also, at this point in the game, CUDA has lost its original edge since OpenCL compilers are getting better. Now, (for a valid comparison) the only reason CUDA code could/should run faster than the equivalent OpenCL code (on Nvidia hardware) is Nvidia purposefully sandbagging their OpenCL compiler since there are minimal differences when porting code.

 

that article is 3 yrs old

 

I wish Havok would re-code their engine to use OpenCL. That would end the PhysX debate once and for all.

 

this is off topic a bit but does the 6990M support opencl? 'Cause I checked that via GPU-Z and the option was not checked. Only DirectCompute 5.0.

If there is a way to make it enabled, could someone share?

 

Think Intel owns Havok or a big chunk of it so don’t think it will happen.

 

It does support it. You don't need to do anything but install an up to date driver. GPU-Z might not be reporting it correctly.

 

I don't think GPUz has ever correctly identified the mobile AMD cards as OpenCL capable. If I remember though, even the 4xxx series had basic OpenCL capabilities; everything after should have no problems with OpenCL.

 

I don't see any reason why intel should shun OpenCL.

 

thanks! I am having the latest 12.1 driver so that shouldnt be a problem.

 

Yes, just to extend. The 4xxx series had basic OpenCL capabilities with some functionality being emulated in software (such as local memory, if you don't understand this, from a high level it's basically a user program managed cache). With the 5xxx series, full OpenCL 1.1 compliance was reached and unless I'm mistaken, OpenCL 1.2 support has now been enabled with their latest SDK/driver release (Dec. 13th?).

But to conclude, don't worry about OpenCL capabilities, it has been baked into the drivers (like Nvidia had been doing for a while) for the past number of months.

As another side now, from AMD 7950 Official Documents & Specifications Leaked it seems OpenCL will be supported by WinZip 16.5 (does anyone even use WinZip anymore?).

 

Wont be that hard to get OpenCL support into 7-Zip since its open source but its already blazing fast when its using all 8 threads on my i7. :3

 

Ah, fair enough. I was wondering why I've seen nothing about it. That's probably why.

 

Would u mind sharing how you get your 7zip to use all 8 threads? I never noticed a difference between my old m11x dual core processor and my current i7.

 

OT: it depends on which compression method you choose.

 

With the GUI you can archive / extract with 8 cores when making regular .zip archives but for .7zip archives you got to use the command line.