[TT]AMD introduces heterogeneous Uniform Memory Access

[frozentundra123456]

Are you aware that all of their post 2013 products will be based around this very capability? It's not a matter of if at all, just when. HSA is the next big thing when it comes to how CPU and GPU are integrated and how they communicate in a single chip. It eliminates all sorts of bottlenecks while providing means for easier programming model. Kaveri is just first (r second if you count PS4's chip) in the line. EX core will bring even more capability:

Again, my point exactly, show me the supporting evidence, show me some data proving the adaption of HSA and the advantage it provides. I mean hard performance numbers from an unbiased review site, not some powerpoint slide from AMD marketing.

 

[USER8000]

Some documents on the AMD website mentioning the work on stacking:

http://sites.amd.com/la/Documents/TFE2011_001AMC.pdf

http://sites.amd.com/us/Documents/TFE2011_006HYN.pdf

Again, my point exactly, show me the supporting evidence, show me some data proving the adaption of HSA and the advantage it provides. I mean hard performance numbers from an unbiased review site, not some powerpoint slide from AMD marketing.

It appears the slide is not entirely accurate,as it increasing looks like the PS4 is the first hardware at stage 3:

http://techreport.com/news/24725/ps4-architect-discusses-console-custom-amd-processor

 

[R0H1T]

Again, my point exactly, show me the supporting evidence, show me some data proving the adaption of HSA and the advantage it provides. I mean hard performance numbers from an unbiased review site, not some powerpoint slide from AMD marketing.

What about OpenCL for starters ?

 

[ShintaiDK]

What about OpenCL for starters ?

OpenCL is not HSA. But HSA will most likely suffer the same fate.

Also OpenCL is now stuck in the same drawer as Cuda, DirectCompute etc.

But atleast OpenCL got a spec...

 

[ShintaiDK]

You do realise,that the PS4 uses stacked GDDR5 right?? AMD are working together with Amkor.

PS4 dont use stacked GDDR5.

 

[grimpr]

this tecnology was created since GCN shipped, it took them a very long time to finish it

probably the best slide...

I like this slide too, it shows AMDs reasoning with the software world, their no.1 priority is ease of programming, this is the most crucial so they can effectively differentiate from Intel by attracting developers.

 

[bgt]

Now we wait the rest of the Anti-AMD team to storm in here.

:biggrin::biggrin:

 

[AtenRa]

Again, my point exactly, show me the supporting evidence, show me some data proving the adaption of HSA and the advantage it provides. I mean hard performance numbers from an unbiased review site, not some powerpoint slide from AMD marketing.

This is not about a product, this is about HSA as technology. And by the way, how many Intel AVX2 products performance figures have you seen ???

 

[ShintaiDK]

This is not about a product, this is about HSA as technology. And by the way, how many Intel AVX2 products performance figures have you seen ???

AVX2 code have been generated for over a year in products. And the spec done long before that.

HSA sofar is nothing but something on a powerpoint slide. No spec, no tools. And thats what people are getting tired of.

 

[AtenRa]

AVX2 code have been generated for over a year in products. And the spec done long before that.

HSA sofar is nothing but something on a powerpoint slide. No spec, no tools. And thats what people are getting tired of.

Where are the performance figures ???

 

[piasabird]

Just show the benchmarks.

 

[Arkadrel]

Stop white knighting AMD fans.

I'll demand more from AMD - than stupid powerpoint slides that tell me nothing more than the idea in 1 sentence.

Show the performance gains of similar raw output in a dGPU or nonunified system vs a unified one.

Similarly i expect more from Haswell - than they probably will deliver.
And i'll whine about that.

hUMA is a base building block for a HSA Dream - so where the beep is the gains ? Where's the oomf?

They're just showing slides of 1 simple building block with not even some nice marketing claims?
What?
Come on.


This is the IMC all over again - the same sort of "OMFG FIRST" that claim AMD shall take charge.
It's a natural progression for ANY SoC of any uARCH.

And of course AMD is the king of IMC's now.... oh wait.

Source:
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.

Basically they can MORE than double performance this way,
compaired to current mothodes of OpenCL with CPU/GPU.

 

[CHADBOGA]

So does this mean AMD will be able to use their GPU for Floating Point operations in their CPU's?

Also, the proof will be in the eating.(i.e. some benchmarks please).

 

[Arkadrel]

Just show the benchmarks.

Also, the proof will be in the eating.(i.e. some benchmarks please).

I dont think any reviewers have any HSA capable hardware + software atm.

So, all you ll get on the matter is whatever quotes you can find of amd saying gains.
Read my post above, it has performance gains in it.

 

[Enigmoid]

Highly doubt kaveri will launch in 2013. Richland as a stopgap was designed precisely because kaveri wasn't ready in time.

 

[A5]

So does this mean AMD will be able to use their GPU for Floating Point operations in their CPU's?

Also, the proof will be in the eating.(i.e. some benchmarks please).

Hardware for this is, at best, a year away. There isn't anything to run benchmarks on.

The idea with HSA is that it lets you efficiently use the GPU resources for things they are good at without all the switching overhead you get with current GPGPU solutions. It won't be a complete solution until the last section of the slide inf64 posted, though.

I imagine AMD's holy grail with this stuff is to be able to get something like OpenMP to be able to see the GPU cores when compiling. Then you get the GPGPU benefits without having to deal with a separate language/API like OpenCL.

 

[frozentundra123456]

This is not about a product, this is about HSA as technology. And by the way, how many Intel AVX2 products performance figures have you seen ???

I actually agree with you here. I feel the same about the new instructions of Haswell as about HSA and the other new AMD initiatives, there is no solid data yet showing wide (or any) one who uses them, and what the benefits will be if they are adopted.

Like the old Burger King commercial, "Show me the beef". Hopefully soon some of these new technologies will be in the hands of reviewers, and we can actually discuss real data instead of indulging in endless speculation.

 

[R0H1T]

OpenCL is not HSA. But HSA will most likely suffer the same fate.

Also OpenCL is now stuck in the same drawer as Cuda, DirectCompute etc.

But atleast OpenCL got a spec...

You think there was a product akin to Kaveri anytime before this ? OpneCL will evolve like HSA, you can't have software/programs before the hardware for it is out ! Also OpenCL is also perhaps the most common of the three you've mentioned & alot more popular as well !

 

[Olikan]

Knights corner.... easiest to code, high sustained flops and HUGE flexibility

and now kaveri... easier to code, single memory pool and low latency gpu-cpu comunication

can we call Tesla cards dead?

 

[ShintaiDK]

Knights corner.... easiest to code, high sustained flops and HUGE flexibility

and now kaveri... easier to code, single memory pool and low latency gpu-cpu comunication

can we call Tesla cards dead?

Let me know when Kaveri gets near Tesla performance. Or even, when there is a spec for it. You cant code for something nobody besides AMD knows anything about.

Telsa is anything but dead in relation to Kaveri.

 

[ShintaiDK]

You think there was a product akin to Kaveri anytime before this ? OpneCL will evolve like HSA, you can't have software/programs before the hardware for it is out ! Also OpenCL is also perhaps the most common of the three you've mentioned & alot more popular as well !

OpenCL got the exact same inherited limitations as Cuda. And we know how far Cuda reached. So lets not derail this thread further with more OpenCL nonsense and hyped hopes.

 

[Kuroimaho]

Where are the performance figures ???

Here you are, well at least for X264:

http://www.scribd.com/doc/137419114/Introduction-to-AVX2-optimizations-in-x264

 

[Phynaz]

It certainly looks like AMD is getting their act together with this stuff.

The idea of the GPU having access to paged memory is very cool. It makes sense once you think of it, but it seems nobody but AMD did think of it.

Also I'm impressed by the cache coherency.

The last thing they need to do in order to call this true HSA is to publish the GPU instruction set so that you don't have to go through a driver layer to access it.

AMD has a long history of BS Powerpoint decks, but this is one presentation I hope bears fruit.

 

[podspi]

I certainly looks like AMD is getting their act together with this stuff.

The idea of the GPU having access to paged memory is very cool. It makes sense once you think of it, but it seems nobody but AMD did think of it.

Also I'm impressed by the cache coherency.

The last thing they need to do in order to call this true HSA is to publish the GPU instruction set so that you don't have to go through a driver layer to access it.

AMD has a long history of BS Powerpoint decks, but this is one presentation I hope bears fruit.

That is the big issue isn't it? Nobody is targeting GPUs, they're targeting GPU drivers. I think one of the advantages of Intel's Phi is that you target x86.

 

[Phynaz]

Now we wait the rest of the Anti-AMD team to storm in here.

You know if you don't like it, you don't have to stir the pot.