Maybe We Don't Understand the implications of OCL and HSA

[monstercameron]

found theses benchmarks

you can see how the a10-6800k performs in a straight up cpu benchmark, not bad. The Intel parts do out perform it.

now if we look at the benchmark with the a10 6.8k having enabled avx and opencl acceleration, it outperforms the i7-3960x -in cpu transcoding.

This has me thinking that maybe we [by that I mean I] dont fully comprehend what a fully hsa compliant apu is capable of. Although it wont matter if devs dont take advantage of this.

I havent found any other benchmarks like this so its pretty much a cherrypick. In any case what do you think are the implications of HSA?

 

[ShintaiDK]

Try transcode using quicksync...

HSA, CUDA, OpenCL etc is nothing but a tiny niche that constantly gets hyped beyond reality.

 

[monstercameron]

Try transcode using quicksync...

HSA, CUDA, OpenCL etc is nothing but a tiny niche that constantly gets hyped beyond reality.

isn’t quicksync dedicated TO video encoding, then we can compare that to VCE in amd gpus, but those benchmarks are for compute[gpu&cpu] not dedicated hardware.

 

[Erenhardt]

This has me thinking that maybe we [by that I mean I] dont fully comprehend what a fully hsa compliant apu is capable of. Although it wont matter if devs dont take advantage of this.

And Intel have enough $$$ to ensure no one major will optimize for it, as long as AMD is on top.

 

[mrmt]

And Intel have enough $$$ to ensure no one major will optimize for it, as long as AMD is on top.

Proof of that?

 

[Erenhardt]

Proof of that?

Ok, you got me. Intel doesn't have enough $$$ to ensure no one major will optimize for it, as long as AMD is on top.

 

[ShintaiDK]

Ok, you got me. Intel doesn't have enough $$$ to ensure no one major will optimize for it, as long as AMD is on top.

So no proof?

 

[monstercameron]

Didn't take long for this to devolve into a conspiracy thread.

 

[mavere]

Video encoding is kind of a bad example if only because a tiny sliver of fixed-function silicon (e.g. QuickSync) can perform the same task significantly quicker while using less power.

 

[ShintaiDK]

isn’t quicksync dedicated TO video encoding, then we can compare that to VCE in amd gpus, but those benchmarks are for compute[gpu&cpu] not dedicated hardware.

You also forget quality of the output. Specially with transcode.

But the underline is, compute is a niche and we dont really have more apps using compute today than CUDA had for ages. We now just sit with HSA, CUDA, OpenCL instead, but the ship is still in dock.

 

[monstercameron]

You also forget quality of the output. Specially with transcode.

But the underline is, compute is a niche and we dont really have more apps using compute today than CUDA had for ages. We now just sit with HSA, CUDA, OpenCL instead, but the ship is still in dock.

I understand your point, maybe I should have been clearer, I wanted to see what guys thought HSA could give us, not necessarily what it is giving us now.

with shared memory, what benefits could we see, other than game physics?

aside: does anyone know any programs that have added ocl acceleration so we can see before and after?

 

[mrmt]

I understand your point, maybe I should have been clearer, I wanted to see what guys thought HSA could give us, not necessarily what it is giving us now.

Where's Haswell AVX + OpenCL score?

 

[monstercameron]

Where's Haswell AVX + OpenCL score?

good question...

 

[mrmt]

Ok, you got me. Intel doesn't have enough $$$ to ensure no one major will optimize for it, as long as AMD is on top.

Maybe you could try the other way around:

Who is going to pay thousands of hours in salaries to optimize for AMD.

 

[Jovec]

It's been what, almost 4 years with AMD's HSA potential? AMD as a company didn't, and still doesn't, have the resources or market share to push HSA and OpenCL in any meaningful way. Llano should have launched with an AMD provided encoding OpenCL app or library. Now Intel has GT3e beating AMD's IGP, and the HD4000 line is comparable in performance and can easily be extended to match whatever GPGPU performance AMD can offer should Intel wish to allocate die space.

AMD has never been short on ideas, but unfortunately ideas, even tech ideas, are a dime a dozen. Not much even in Haswell is a bleeding edge idea. It's all about execution and implementation and Intel just beats AMD in those areas.

 

[beginner99]

Where's Haswell AVX + OpenCL score?

exactly...

And as already mentioned there is also QuickSync and for some people quality matters and cpu encodes (x264) are still just better.

And lets be honest, this is even more niche than gaming...I think I have trans coded like less than 10 videos and I'm not an average person.

 

[Vesku]

You also forget quality of the output. Specially with transcode.

But the underline is, compute is a niche and we dont really have more apps using compute today than CUDA had for ages. We now just sit with HSA, CUDA, OpenCL instead, but the ship is still in dock.

Yep, and since there is no AMD APUs in the upcoming Xbox and Playstation there won't be much of a reason for developers to work with HSA and OpenCL... oh wait a second.

 

[ShintaiDK]

Yep, and since there is no AMD APUs in the upcoming Xbox and Playstation there won't be much of a reason for developers to work with OpenCL... oh wait a second.

By that logic everything now should support 6+ threads. Because Xbox360 and PS3 did.

HSA has gone nowhere. OpenCL is just a copy of CUDA in terms of apps.

So what will radically change?

 

[monstercameron]

exactly...

And as already mentioned there is also QuickSync and for some people quality matters and cpu encodes (x264) are still just better.

And lets be honest, this is even more niche than gaming...I think I have trans coded like less than 10 videos and I'm not an average person.

that was just an example, not the point. the point was to compare the performance of a high performance cpu core versus an hsa style setup.

the a10-6800k [$150] outperforms the i7-3960x [$1000+] by using optimized software that uses both cpu and igpu, not arguing if fixed function hardware is faster, amd, intel, nvidia etc all have fixed function video encoders. Saying that this is useless because fixed function hardware is faster, is like saying lets replace all general purpose hardware with fixed functions blocks. <sarcasm>fixed function firefox browser for everybody</sarcasm>

 

[Cerb]

I havent found any other benchmarks like this so its pretty much a cherrypick. In any case what do you think are the implications of HSA?

Not much. They blew hot air for too long. Any such victories like this are short-lived, as Intel supports everything they do, now, plus has AVX2-supporting CPUs out the door.

 

[Sequences123]

By that logic everything now should support 6+ threads. Because Xbox360 and PS3 did.

I'm not sure if you've done threading work before, but its hard (at least I find it hard). There are overheads (from both the software and hardware perspective) to consider when doing multithreaded programming as well, so more might not necessarily be better. As far as I can tell, threading nowadays is much better than it was 10 years ago.

With respect to highly multithreaded applications, don't graphics cores count? Its my understanding that CUDA/OpenCL are offered as API to allow for massively parallel (compute intense) programs.

 

[monstercameron]

Not much. They blew hot air for too long. Any such victories like this are short-lived, as Intel supports everything they do, now, plus has AVX2-supporting CPUs out the door.

this isnt necessarily about amd v. intel, rather apu v cpu, also if we are talking about the future, there is always kaveri that has steamroller cores with avx2 support and gcn core with massive ocl perf boost over vliw4...so lets keep the flame baiting to a minimum, please.

 

[AtenRa]

aside: does anyone know any programs that have added ocl acceleration so we can see before and after?

http://openclnews.com/apps

 

[NTMBK]

It's going to be an interesting few years. We're already seeing Intel letting you use what is basically a GPU as the main processor (Xeon Phi in a socket). Has anyone seen good OpenCL benchmarks of the Phi yet?

 

[Vesku]

By that logic everything now should support 6+ threads. Because Xbox360 and PS3 did.

HSA has gone nowhere. OpenCL is just a copy of CUDA in terms of apps.

So what will radically change?

Xbox 360 and PS3 weren't x86, I'd expect given the next gen consoles we will see more games and programs get more effect from having access to more than 2 threads.