AMD have released a tool which converts Nvidia's CUDA code to portable C++ code which works on both Nvidia and AMD cards!

[patrickjp93]

15 minutes ago, Joe_MacDougall said:

True. I mean they are putting Radeon tech in new Intel CPUs apparently.

That's a rumor with no source. It's just an OC3D (OCN?) editor claiming it.

[Joe_MacDougall]

Just now, patrickjp93 said:

That's a rumor with no source. It's just an OC3D (OCN?) editor claiming it.

Yeah. Apparently.

[ThinkWithPortals]

56 minutes ago, patrickjp93 said:

Barcelona and Mullins say hi.

I feel like I should know what you're talking about, but I'm afraid I don't. What do you mean?

[Drak3]

7 hours ago, Joe_MacDougall said:

I know but what I'm saying is that the US might have similar competition laws?

They do, but enforcement isn't consistent, and more times than not, they still act as one company.

And although I'll probably get flak for this, I don't think government has any right to split up a company only because it became a monopoly.

[Joe_MacDougall]

35 minutes ago, Drak3 said:

They do, but enforcement isn't consistent, and more times than not, they still act as one company.

And although I'll probably get flak for this, I don't think government has any right to split up a company only because it became a monopoly.

I think that they should probably actively encourage new competition to appear but splitting a company is sketchy.

[Drak3]

Just now, Joe_MacDougall said:

I think that they should probably actively encourage new competition to appear but splitting a company is sketchy.

I agree. Splitting the company doesn't really work either. They just assume two monopolies that don't interfere with each other and often act as one company anyways.

[Curufinwe_wins]

2 hours ago, patrickjp93 said:

That's a rumor with no source. It's just an OC3D (OCN?) editor claiming it.

Which for the record, it could be nothing more than free sync implementation. 

[patrickjp93]

2 hours ago, ThinkWithPortals said:

I feel like I should know what you're talking about, but I'm afraid I don't. What do you mean?

AMD caught flat-out lying and suffering legal consequences.

 

58 minutes ago, Curufinwe_wins said:

Which for the record, it could be nothing more than free sync implementation. 

I doubt it. AMD's FreeSync is just a front-end name for Adaptive Sync on AMD graphics technologies. Intel can implement that on its own.

[RexinOridle]

Nvidia is good and all that but they can shove it up their "places". Proprietary shit is screwing up games for quite some time now.

[Albatross]

I don't understand really what this means but it sounds beneficial?

[leadeater]

17 hours ago, dalekphalm said:

For a few reasons:

1. Whether we want to believe it or not, CUDA seems to be a more efficient GPGPU Compute language, which leads into:

2. CUDA is pretty much the "go-to" Compute language, and finally

3. CUDA is exclusive to NVIDIA - and they have not given AMD licenses required to compute CUDA code on AMD hardware

 

So AMD has found a good work around: A code translator that takes raw CUDA source code, and converts that code into raw C++ source code, which can then be compiled into an OpenCL application/service/whatever.

Part of the reason why CUDA is more efficient than OpenCL is due to it's popularity. More people use it and so it naturally evolves to a better product over time because of it. If we were to compare the very first iteration of CUDA to what OpenCL is now I can't say which would better but it would be closer than current CUDA.

 

What OpenCL needs to get better is people to use it, fine tune it, submit feature improvements etc.

 

A code translator is a very good move to get people to step in to the realm of OpenCL, if you then only have to optimize what has been converted that can be much less work than doing all of it from scratch. But if there isn't good support from AMD, the hardware people need, partnerships in place with server manufacturers then the whole effort is moot as people wouldn't move to OpenCL as why would you if you are going to stay on Nvidia hardware.

[AresKrieger]

If Nvida controls CUDA they can counter this they so chose to, plus I highly doubt it will have a 1 to 1 efficiency so why would you want to get a Radeon card if you planned to use CUDA, it isn't going to be the choice made by companies who use the technology hence who is this trying to sell products to?

[AlwaysFSX]

9 hours ago, Prysin said:

no. Because AMD isnt the one running or modifying the code. The tool is used by people who already owns the license. All AMD allowed them to do, is to "quickly" rewrite the code they already wrote themselves.

 

unlike Nvidia who tried to make that X86 translator in hardware. And that is something else

Okay this is an explanation I understand, thanks.

[patrickjp93]

1 hour ago, AresKrieger said:

If Nvida controls CUDA they can counter this they so chose to, plus I highly doubt it will have a 1 to 1 efficiency so why would you want to get a Radeon card if you planned to use CUDA, it isn't going to be the choice made by companies who use the technology hence who is this trying to sell products to?

If I could compile every Java program in the world to C++, I would, because C++ is faster, more lightweight, and more stable.

 

Language matters both in programmer productivity and program performance. It's exactly what AMD is acknowledging with this. CUDA is easy to program in compared to OpenCL, even OpenCL 2.1.

[Space Reptile]

16 hours ago, AlwaysFSX said:

Isn't that illegal without AMD having a license?

there was talk about them having acquired the cuda license

 

amd is a clever girl , time for stuff to get ported over

 

AMD FLEX DEMO WHEN

[Citadelen]

18 hours ago, AlwaysFSX said:

-snip-

AMD isn't using CUDA, they've just released a porting tool. Nothing illegal there.

[AresKrieger]

7 hours ago, patrickjp93 said:

If I could compile every Java program in the world to C++, I would, because C++ is faster, more lightweight, and more stable.

 

Language matters both in programmer productivity and program performance. It's exactly what AMD is acknowledging with this. CUDA is easy to program in compared to OpenCL, even OpenCL 2.1.

I'm aware C++ is great however I'm not seeing how this will translate into additional money for amd atm, maybe their are enough people who like the concept of converting to C++ but from my understanding CUDA development is more so used in a corporate/engineering setting.

[The Benjamins]

1 minute ago, AresKrieger said:

I'm aware C++ is great however I'm not seeing how this will translate into additional money for amd atm, maybe their are enough people who like the concept of converting to C++ but from my understanding CUDA development is more so used in a corporate/engineering setting.

It means the custom code made on CUDA can now easily be used on any hardware, which means that a company can buy AMD hardware now when they get money for more hardware or upgrade versus needing to account re-writing code into the cost of switching. this all relies on that the AMD counterpart card has a better price/performance. It basically give companies the option to buy AMD if they have a lot of code made for CUDA.

[manikyath]

23 hours ago, AlwaysFSX said:

Isn't that illegal without AMD having a license?

i wouldnt see why it would be.

[SpaceGhostC2C]

14 hours ago, Joe_MacDougall said:

I think that they should probably actively encourage new competition to appear but splitting a company is sketchy.

It depends on the nature of the market. You'll find company splitting or, much more often, rejections of merger proposals (i.e., two companies becoming one) in markets that are (close to) "natural monopolies". That is, markets were the fixed investment cost is so large that the market cannot have more than one (or n) profitable companies. Usual examples are land telephone, electricity, etc, where the cost of building the infrastructure to supply all costumers is so high than no other company could build its own network and expect to stay alive (unless it becomes a monopoly itself). The trend in recent years, especially in landlines, has been to force the original monopolist to rent the infrastructure to any company willing to at a fixed, regulated price (so it can't keep competition away by saying "OK, you can use it, but every minute costs... one million dollars!"). That way, multiple operators can exist, although there still is a single infrastructure provider.

[Hayabusa1989]

1 hour ago, SpaceGhostC2C said:

(so it can't keep competition away by saying "OK, you can use it, but every minute costs... one million dollars!")

The UK has finally started to crackdown on OpenReach for this who rent the infrastructure out to ISP's but charge extreme amounts for other companies to use the ducting to provide their own lines. Whilst this is still an issue at least now Openreach are being forced further away from their Parent ISP BT

[Hayabusa1989]

and on the CUDA point, good for AMD ! But like Apple, Nvidia always find a way to make something else that seems essential to the future of humanity further down the line and make that proprietary

[vorticalbox]

On ‎15‎/‎12‎/‎2016 at 3:35 PM, AlwaysFSX said:

Isn't that illegal without AMD having a license?

how so? It translates c++ into cuda nothing illegal about it. If licence to use cuda is needed than that would fall to the user.

[phidn]

Did someone benchmark this?