AMD Unveils its Heterogeneous Uniform Memory Access (hUMA) Technology

[vickybat]

Like i said in an earlier thread, HSA is growing at a faster pace with the advent of next-gen consoles. Amd is next in line to join nvidia and finally announces its intentions on bringing unified memory to pc market. GPU's have immense computing power and if used properly, can yield massive benefits, changing the face of computing.

Unified memory access is going to impact gpgpu computing positively leading to more efficient compute API's and both cpu and gpu can work in harmony with extremely low latency issues. Gpu will finally have direct access to main memory and will have plenty of data for parallel execution and cpu too can access complex macro operations directly from VRAM as they will act as one now.

HSA is upon us.

AMD Unveils its Heterogeneous Uniform Memory Access (hUMA) Technology

 

[RCuber]

THIS is the boost we in performance we are looking for!!! , I just like the innovation and the courage by AMD in taking up new challenges. AMD acquiring ATI was the best move ever in the computer industry.

 

[harshilsharma63]

Great job AMD, great job. I doubt if it'll be a success as like most of AMD's great techs, there will be the issue of lack of exploitation by software.

 

[NoasArcAngel]

THIS is the boost we in performance we are looking for!!! , I just like the innovation and the courage by AMD in taking up new challenges. AMD acquiring ATI was the best move ever in the computer industry.

yes like facebook going public. If amd would not have bought ATI, i think they would still be competing with intel

the boost in performance will only come when intel and amd shift to hsa and so does nvidia. This means a singular architecture and greater optimization for the same hardware and therefore better performance.

 

[kg11sgbg]

THIS is the boost we in performance we are looking for!!! , I just like the innovation and the courage by AMD in taking up new challenges. AMD acquiring ATI was the best move ever in the computer industry.

+1,same for me also.
In fact I am WAITING for the HSA based(KAVERI) laptops,even if I have to wait more...........

Great job AMD, great job. I doubt if it'll be a success as like most of AMD's great techs, there will be the issue of lack of exploitation by software.

That is one point worth for consideration.Hopefully developers will take care for that,a WISH again.

AMD and Partners to Discuss the Future of Heterogeneous Computing This November - X-bit labs

It could imply that arrangements are starting and software developers are preparing for their worth.

 

[rajnusker]

When is this happening?

 

[harshilsharma63]

When is this happening?

NOthing much, just some blowing with in Intel's eyes.

 

[Sainatarajan]

Now Intel will be up with a newer one.... I like the competition here...

 

[Cilus]

Actually designing the PS4 and XBOX Next architecture is the best thing for AMD in recent days. There they got the funding to implement things they have been planning for years.
Currently CPU performance level is basically reaching its thresholds and almost all the optimizations to boost its single threaded and IPC performance have been used. We need some new concepts here, concept to utilize each componets of our PC to act as an untified resource to handle the tasks. AMD's idea of fusion is to bring CPU and GPU under a single umbrella, not only in hardware implementation but also in the programming and design concepts.

If you go though the slides then you will find out that due to the unified access over the system physical memory and Virtual Memory, we can ignore latencies for identification of type of work (whether it will run better on CPU or GPU) then 1st redirecting the thread to the corresponding module and then copying the data to that module's local memory.
In HUMA (NUMA for heterogeneous Architecture), as all the units share a single memory, programmers can directly assign the code blocks or processes to the respective modules like CPU or GPU. Also, here redirecting is not needed and simple interrupt controls can guaarnty the consistency of the data.

 

[Tech_Wiz]

Ok....most of it went 2 feet above my head

 

[RCuber]

@Cilus: where is the like button when you need one

 

[sumonpathak]

aaand cilus is on fire today....

 

[harshilsharma63]

aaand cilus is on fire today....

NO offence for Cilus, but read CUDA docs and thoughts like these will come to your mind too.

 

[anirbandd]

awesome thing this is... AMD +1

 

[NVIDIAGeek]

Now, tell me when they're gonna release the components required for this? I don't want it to be "coming soon™" or in 2015....

 

[vickybat]

^^ Nvidia's HUMA implementation will be in their maxwell gpu's while amd probably ( certainly) implement huma for GCN 2.0. Expect both in first quarter 2014.

 

[NVIDIAGeek]

^So next year at this time? I think I'm gonna stick to I have now then.

 

[sumonpathak]

NO offence for Cilus, but read CUDA docs and thoughts like these will come to your mind too.

actually i did...the comment was more of an appreciation...everything i know about CPU architecture is due to him.

"dont know why you decide to pick out my comment"

 

[noob]

Whatever it is, how much % performance gain and power savings are we talking here ?

 

[harshilsharma63]

Whatever it is, how much % performance gain and power savings are we talking here ?

That strictly depends upon the implementation. If the unified hardware is to decide which processing unit to allot to which code, then it depends on the hardware itself. If the hardware just provides a platform for devs to and leaves it on them to decide about the processing unit, then there may be a lower performance gain over the former method.