VRW - AMD Intel APUs Open Source GPGPU Apps and Performance

[Dribble]

People here forget that the vast majority of hardware will be mobile the coming years. Face/voice recognition (no need for passwords), GPU accelerated browsing, Gaming, Office/financial apps, video/photo editing, scientific apps (chemistry, physics, medical, Oil, seismic etc) all those can be GPU accelerated.

Your mobile phone/Laptop and Tablet will know you (voice/face recognition), you will be able to edit large 4K photos/videos on your mobile devices, more censors will be integrated in to mobile devices that will need GPU acceleration. Gaming will also benefit.
Mobile workstations will also benefit for people on the go, designers, architects, lab simulations, financials and more. Yes Servers will also be there but you will be able to have a small office workstation at your fingertips all day long everywhere you are, office, on your car, on the plane, on the boat (space ??? yea it will take more time for that unfortunately).

It is why everyone in the industry invest on iGPUs.

That's not how mobile does it. If there is some specific thing that needs accelerated they just add a custom bit of hardware to the soc to do it, they do not use the gpu.

 

[sm625]

There are very good reasons to not have biometrics replace passwords. What do you do when someone steals a password? You change it. What do you do when someone steals a biometric?

Not to mention fingerprints can and have already been hacked using just a simple phone camera. It is only a matter of time before retina scans are also hacked. The only reason they are pushing biometrics so hard is because they are closet Nazis who know damn well that biometrics offer you no security whatsover if they have both you and your device in custody. They love that. Biometrics offers zero security vs a password. At least with a password they have to waterboard you. But the real killer about passwords is that you can always code in a 2nd password that, when entered, completely wipes the device. "I cannot take any more torture, I will reveal my password. It is "snookies212". When they enter "snookies212" the device wipes itself. Ooops... yeah Nazis hate that. So they push biometrics at all their tech conferences, in tech literature, and in pop culture. They want badly for the general public to move away from passwords. And of course apple users are only too happy to oblige.

 

[Essence_of_War]

Hmm HSA kinda solves that, by directing what compute units do what computation. Serial tasks to the cpu and parallel ones to the gpu.

HSA, if working/implemented properly, should prevent you from having to manually copy objects back and forth because you're using shared address spaces. That doesn't actually write OpenCL code for you, and it doesn't make your problem intrinsically more parallel.

 

[Phynaz]

Hmm HSA kinda solves that, by directing what compute units do what computation. Serial tasks to the cpu and parallel ones to the gpu.

HSA doesn't go that, the programmer still has do it. And there's literally zero commercial tools for programmers today.

 

[monstercameron]

HSA doesn't go that, the programmer still has do it. And there's literally zero commercial tools for programmers today.

Tools aren't production ready but are being actively developed. Besides what I mentioned was a design goal of HSA, and is one of the differentiating factors over opencl.

 

[Essence_of_War]

Besides what I mentioned was a design goal of HSA, and is one of the differentiating factors over opencl.

It sounds to me like you're talking about HSA like it's supposed to be a replacement for OpenCL, and I'm pretty sure that is just not the case. Are we having a miscommunication?

 

[AtenRa]

HSA doesn't go that, the programmer still has do it. And there's literally zero commercial tools for programmers today.

http://developer.amd.com/resources/...hat-is-heterogeneous-system-architecture-hsa/

The HSA team at AMD analyzed the performance of Haar Face Detect, a commonly used multi-stage video analysis algorithm used to identify faces in a video stream. The team compared a CPU/GPU implementation in OpenCL™ against an HSA implementation. The HSA version seamlessly shares data between CPU and GPU, without memory copies or cache flushes because it assigns each part of the workload to the most appropriate processor with minimal dispatch overhead. The net result was a 2.3x relative performance gain at a 2.4x reduced power level*. This level of performance is not possible using only multicore CPU, only GPU, or even combined CPU and GPU with today’s driver model. Just as important, it is done using simple extensions to C++, not a totally different programming model.

Also, HSA developer tools are available in the link bellow.
http://www.hsafoundation.com/hsa-developer-tools/

 

[AtenRa]

It sounds to me like you're talking about HSA like it's supposed to be a replacement for OpenCL, and I'm pretty sure that is just not the case. Are we having a miscommunication?

No, he means that through HSA, OpenCL(or any other app/program etc) will become even more efficient. See my quote above.

 

[mrmt]

Tools aren't production ready but are being actively developed. Besides what I mentioned was a design goal of HSA, and is one of the differentiating factors over opencl.

What makes you think AMD will be capable of develop these tools? Developers that bought AMD entry on OpenCL in circa 2011 were burned.

 

[monstercameron]

What makes you think AMD will be capable of develop these tools? Developers that bought AMD entry on OpenCL in circa 2011 were burned.

Don't be difficult, amd isn't just one man twiddling his thumbs. It is so pointless replying a spin master. It's like your are claiming that AMDs really smart engineers can not build tools for their hardware.

 

[geoxile]

It sounds to me like you're talking about HSA like it's supposed to be a replacement for OpenCL, and I'm pretty sure that is just not the case. Are we having a miscommunication?

My understanding is that there is an openCL compiler that can produce HSAIL, so it's complementary in that case. But the foundation also produced a C++ AMP compiler that can produce HSAIL, so developers can avoid writing OpenCL when applicable and use some other language to produce HSAIL.

 

[ShintaiDK]

Don't be difficult, amd isn't just one man twiddling his thumbs. It is so pointless replying a spin master. It's like your are claiming that AMDs really smart engineers can not build tools for their hardware.

But 4 years later, where is the tools? Talking about spin, HSA have been the greatest thing how long now? And so far it ended up where?

HSA isnt going to save AMD and change everything. Rather the opposite as sales show. All the HSA resources are wasted. Even worse, taken from more important projects.

 

[monstercameron]

But 4 years later, where is the tools? talking about spin, HSA have been the greatest thing how long now? And so far it ended up like what?

On github/bitbucket, and open source for the most part. Oh come on shintai we both know it takes time to implement these thing...I mean where are all the avx/2 enabled software?

 

[ShintaiDK]

On github/bitbucket, and open source for the most part. Oh come on shintai we both know it takes time to implement these thing...I mean where are all the avx/2 enabled software?

Plenty of AVX and AVX2 software around. Its so common you dont even notice. HSA? Not so much.

Why? Tools.

 

[AtenRa]

The first fully enabled HSA hardware was just released (Carrizo), things will only get faster now. Especially when Samsung, Qualcomm and all the others will start to produce HSA enabled hardware/apps.

 

[ShintaiDK]

The first fully enabled HSA hardware was just released (Carrizo), things will only get faster now. Especially when Samsung, Qualcomm and all the others will start to produce HSA enabled hardware/apps.

Will they? Can you show what products they commited to develop with HSA? Or is it more on the hope they may will basis?

 

[Nothingness]

Plenty of AVX and AVX2 software around.

Really? Would you have a list?

 

[ShintaiDK]

Really? Would you have a list?

AVX code have been produced since 2010. And auto vectorization means you dont even have to code for AVX to use AVX.

https://software.intel.com/sites/de...pThr_-_Using_AVX_Without_Writing_AVX_Code.pdf

The list would be endless. You could just as well ask for a list of SSE software.

 

[Phynaz]

http://developer.amd.com/resources/...hat-is-heterogeneous-system-architecture-hsa/

From your link:

applications can create data structures in a single unified address space and can initiate work items on the hardware most appropriate for a given task.

Applications. Not the hardware. The programmer still has to write the application. It doesn't magically become HSA by setting a switch somewhere.

 

[monstercameron]

AVX code have been produced since 2010. And auto vectorization means you dont even have to code for AVX to use AVX.

https://software.intel.com/sites/de...pThr_-_Using_AVX_Without_Writing_AVX_Code.pdf

The list would be endless. You could just as well ask for a list of SSE software.

Nice cop out. I shouldn't have expected any less from you

 

[ShintaiDK]

Few examples just with AVX2.

Linux RAID since 2012 supporting AVX2.
https://git.kernel.org/cgit/linux/k.../?id=2c935842bdb46f5f557426feb4d2bdfdad1aa5f9

Grid 2 game even contains dedicated grid2_avx.exe.

Dirt Showdown with AVX2.
http://www.pcgameshardware.de/Dirt-.../Dirt-Showdown-Benchmarks-CPU-und-GPU-886671/

FFmpeg is another with AVX2 support.

PicoHTTPParser with AVX2 support.
https://blog.cloudflare.com/improving-picohttpparser-further-with-avx2/

Cyberlink products with AVX2 support.
http://www.cyberlink.com/products/powerdirector-ultra/spec_en_EU.html?&r=1

HVEC/H.265 encoding with AVX2:
http://www.tomshardware.com/reviews/x265-hevc-encoder,3565-2.html

x.264 with AVX2:
https://mailman.videolan.org/pipermail/x264-devel/2013-April/010044.html

Various Adobe products with AVX2, example:
http://blogs.adobe.com/lightroomjournal/2014/10/1366.html

Again, all about the tools. Something Intel supplies and support while AMD...well...much to be desired.

 

[Phynaz]

On github/bitbucket, and open source for the most part. Oh come on shintai we both know it takes time to implement these thing...I mean where are all the avx/2 enabled software?

Number one dev environment in the industry - Microsoft. AMD HSA support?

Intel. AMD HSA Support?

Oracle (Java)?

Portland Group?

Adobe?

Three years on and nothing.

 

[Nothingness]

AVX code have been produced since 2010. And auto vectorization means you dont even have to code for AVX to use AVX.

https://software.intel.com/sites/de...pThr_-_Using_AVX_Without_Writing_AVX_Code.pdf

So I guess AVX-512 is widely used given that intrinsics and icc support for it have been available for months :biggreen:

The list would be endless. You could just as well ask for a list of SSE software.

And icc is used by whom exactly? Your claim needs some more substance.

If the use of AVX was so widespread do you really think Intel would castrate a huge proportion of their CPU? Or is it that AVX/2 speed impact is minor in most cases? Or major only in cases where GPGPU works well too, that is code and 3d?

 

[Nothingness]

This came while I was writing my previous answer

Few examples:

Linux RAID since 2012 supporting AVX2.
https://git.kernel.org/cgit/linux/k.../?id=2c935842bdb46f5f557426feb4d2bdfdad1aa5f9

OK.

Grid 2 game even contains dedicated grid2_avx.exe.

3D? GPU are good at that.

FFmpeg is another with AVX and AVX2 support.

Codec. GPU are good at that cf. h265 on Intel iGPU.

PicoHTTPParser with AVX2 support.

Now that's a good example

Don't get me wrong, I like AVX a lot (and use it a lot), I am just not sure it's that used

 

[mrmt]

Don't be difficult, amd isn't just one man twiddling his thumbs. It is so pointless replying a spin master. It's like your are claiming that AMDs really smart engineers can not build tools for their hardware.

I'm not claiming anything about AMD engineering, I'm claiming things about their management teams. AMD incompetent management simply cannot adequately fund and manage their engineering teams in order to build good software tools.