NVIDIA GeForce GTX 980 Dominates With OpenCL On Linux

[efikkan]

I don't get your argument.
This sounds like you're blaming AMD for NVidia not having OpenCL >= 1.2 support.

Then I suggest you read my post again.

We all know it's just a political thing NVidia does not provide support for newer OpenCL version.
And this is bad. No matter how great their OpenCL 1.1 support is.

And who's more than a bit biased now? Nvidia claim support when they have verified the implementation is working very well, while AMD tends to be a bit more trigger happy when it comes to announcing support. You should also know Nvidia is for a fact one of the largest contributors to OpenCL 1.1 and 2.0.

[opensource]

So Maxwell pretty much nullifies having to buy AMD anymore for OpenCL?

That's interesting indeed. Also fast in Hashcat benchmarks. And of course works with Blender

GTX 980 is not the card to get, the gtx970 is(price/performance point of view).

Agreed.

[entropy]

And who's more than a bit biased now? Nvidia claim support when they have verified the implementation is working very well,
while AMD tends to be a bit more trigger happy when it comes to announcing support.
You should also know Nvidia is for a fact one of the largest contributors to OpenCL 1.1 and 2.0.

Please enlighten me.
Why does Nvidia still not supports OpenCL 1.2 and 2.0 with a driver implementation?
Lack of engineering power? Not passing QA? Hmm, what's left..?

Don't tell me Nvidia is not hindering competition for CUDA.

BTW, OpenCL 1.2 was officially announced in Nov. 2011.
Quite a bit of time to get something out - if that was the goal.

[Sdar]

Please enlighten me.
Why does Nvidia still not supports OpenCL 1.2 and 2.0 with a driver implementation?
Lack of engineering power? Not passing QA? Hmm, what's left..?

Don't tell me Nvidia is not hindering competition for CUDA.

BTW, OpenCL 1.2 was officially announced in Nov. 2011.
Quite a bit of time to get something out - if that was the goal.

On guru 3d TJ stated 1.2 support on 344.11 driver:


I'm not sure if it's true, but Nvidia 1.1 Opencl is still better than AMD Opencl 2.0 when your kernel is large.

[entropy]

On guru 3d TJ stated 1.2 support on 344.11 driver:


I'm not sure if it's true, but Nvidia 1.1 Opencl is still better than AMD Opencl 2.0 when your kernel is large.

If that is true, than it must have been introduced for not too long.
Step in the right direction. Very appreciated.

[entropy]

Hmm, the Caps Viewer screenshot on
http://www.geeks3d.com/20140919/nvid...x-970-support/
seems to tell a different story.

[cutterjohn]

Nvidia does in fact offer better OpenCL-support, not only performance wise, but also in standards compliance. AMD does however advertise higher version support, but there is a big difference in claiming support and actually providing support. This is also true for OpenGL on AMD, they still struggle with standards compliance.

...but nvidia used to almost completely gimp double precision on consumer cards, which they are apparenty no longer doing...

I could never figure that out as, who would do serious work w/o ECC memory or spend the time to take care of it in sw -> perf hit anyways...

ATI/AMD: drivers: they're all halfassed, advertising crap that borked, claimed to fix borkedness, but not really(and oh adding 5 or 6 new borked things), etc. which was the only reason that I swore off ATI.


N.B. I don't give a rat's --- if it has OSS drivers or not, just that it actually works VERY close to advertised. And for you OSSheads driver whingers, wake me up when ATI/AMD coughs up their Verilog/VHDL mdels.

[EDIT]
OpenGL: In theory, yes ATI claims to be more "standards" compliant, but see above, i.e. it's a borkennessfest, and now it's just not their sw. Their hw is looking pretty sad this gen as well, and this time it can't be explained away by craptastic drivers.
[/EDIT]

[tuke81]

On guru 3d TJ stated 1.2 support on 344.11 drive...

Not quite sure, but listed intel hd4600 supports opencl1.2.

Although techpowerupdb says gtx 980 to support opencl 1.2. But i.e. compubench says opencl_c_version OpenCL C 1.1.

[GT220]

Since people need to be reminded every thread. From a Luxrender/Luxmark developer.

http://www.luxrender.net/forum/viewtopic.php?f=34&t=11009

This is the result with Intel OpenCL CPU device:
4.5 seconds is a quite reasonable time.

This is the result with AMD OpenCL CPU device:
I don't know how much takes to compile the kernel because I gave up after 15 minutes.
I mean ... seriously ... C compilers for CPUs are a 40 years old technology.

http://www.luxrender.net/forum/viewtopic.php?f=34&t=11009&start=10

AMD GPU compiler is far far worse than the CPU one. My surprise here was in the finding that even CPU compiler has problems.

AMD OpenCL is pure garbage.

[cutterjohn]

Well this was supposed to be edit2 but the 5m rule bit me on my posterior:
[EDIT2]
As to opencl versions, I'd be pretty well satisfied that, that is likely driven by what their professional grade(gpgpu users who intentionally buy for that purpose) want/require, whereas ATI/AMD just spews out everything since they have to, and then we just hope that most of the ATI pieces work that we get something mostly functional as I don't belive that ATI has separate professional driver development team like nvidia does -> opencl probably really comes from that group.

I mean, lets get down to brass tacks here, nvidia is like the Intel of CPUs fully funding with adequate staffing of their serious efforts, meanwhile ATI/AMD just cobbles crap together on a shoestring, continuously trying to overreach rather than paring back and get their basic shit right FIRST. AMD is like a bunch of amateurs running a company IMNHO, meanwhile nvidia is run like a professional growing concern(mostly they have some rough edges too but nothing ridiculous like AMD).

There are still two things that I need to see from AMD: competitive(or even better more efficient) CPU core designs, and just getting their GPU side act together FINALLY! Hell even this round of hw design is apparently pretty awfuln whereas in the past I always thought that THAT WAS their STRONG point, i.e. GPU hw design.
[/EDIT2]