hUMA's new trick; standardised task dispatch packets

Subject: General Tech | October 23, 2013 - 10:53 AM |
Tagged: amd, hUMA

hUMA's currently well know trick, a shared memory space which both the CPU and GPU can access without penalty is only the first of its revealed optimizations.  The Register talks today about another way in which this new architecture allows the CPU and GPU equal treatment, standardized task queues and dispatch packets which avoid dealing with a kernel level driver to assign tasks.  With hUMA the GPU is able to shedule tasks for the CPU directly.  That would allow any application that was designed to hUMA standards to have its various tasks assigned to the proper processor without needing extra coding.  This not only makes it cheaper and quicker to design apps but would allow all hUMA apps to take advantage of the specialized abilities of both the CPU and GPU at no cost.

View Full Size

"The upcoming chips will utilise a technique AMD calls Heterogeneous Queuing (hQ). This new approach puts the GPU on an equal footing with the CPU: no longer will the graphics engine have to wait for the central processor to tell it what to do."

Here is some more Tech News from around the web:

Tech Talk

 

Source: The Register
October 23, 2013 | 01:34 PM - Posted by Anonymous (not verified)

I wish AMD would just go ahead and put an x86 cpu on their descrete GPUs with this HSA ability, then games could run completely on the descrete CPU/GPU. or run mostly in the CPU/GPU combo, using the fat GPU bus and DDR5 memory, with the laptop's/desktop's CPU there for extra boost when needed! With this hUMA the laptop/computer intrigrated CPU would also be able to dispatch tasks to both the descrete GPU/CPU for accelerated general purpose workloads also! PCIe is fast, but a descrete CPU/GPU combo on the same die shareing a 256 bit bus(or larger), and DDR5 memory would be a hardware optimized gaming console on a card! AMD's Intigrated CPU/GPUs are ok on laptops, but maybe for the desktop the ability to add an extra descrete CPU/GPU combo, hardware optimized for gaming would sell.

October 23, 2013 | 03:10 PM - Posted by Coupe

With everyone moving to SoC you want to move the CPU to a discrete interface. What are the benefits?

October 23, 2013 | 05:08 PM - Posted by Anonymous (not verified)

Not moving the main CPU to a descrete card, but adding a CPU to the descrete GPU! The computer's main CPU remains on the motherboard, and can over PCIe as always send/recieve tasks and such to the CPU/GPU descrete card, with all the PCI overhead, of course! The CPU on the CPU/GPU card whould be directly connected to the GPU over a fat GPU sized bus, that is able to directly address the GDDR5 memory! Say the card has 512 bit bus, well the memory controller could fetch 8, 64 bit data/instrucions from GDDR5 memory per half cycle(DDR), a big plus for keeping the CPU fead! The CPU could also feed the GPU results directly over the fat bus, cutting out the PCIe overhead and such, that would add latency, that would be there if the motherboard CPU needed to interact with the GPU on the descrete card! An extra x86 CPU pared with the GPU on a custom card could have all the optimized hardware tweaks that a deicated gaming console would have, optimizations that could not be done on a CPU that was ment for general purpose computing, the general purpose tasks would be the responsability of the main CPU on the computers motherboard!

October 23, 2013 | 02:59 PM - Posted by Anonymous (not verified)

If this works, sounds good.

To me, sounds prone to lock up or crashing.

How is it determined which has priority for a task?

A little more info please.

October 23, 2013 | 05:31 PM - Posted by Anonymous (not verified)

With the same hardware and software algorithms that they use for their APUs of course, only the APU would be on the descrete APU that resides on the descrete card an APU with a 512 bit bus, monster GPU and GDDR5 memory, a customizied gaming CPU/GPU mashup!

October 23, 2013 | 05:12 PM - Posted by Anonymous (not verified)

edit: deicated to dedicated.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.