• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

[TT]AMD introduces heterogeneous Uniform Memory Access

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
I don't get why anyone wouldn't be excited by this development. Even if it is all hype-machine, if they get people to buy into the hype then that means Intel suddenly has to try and answer that. A return to competition in the marketplace would be welcome by me. Why not you?
 
(1) GCC will be used, for FOSS. LLVM would have been better (CLang will probably crash trying to build CPU+GPU, but still... 😉), of course. Or, since only Catalysts will use OpenCL right, you can get performance with buginess and random X features not working right (AMD Windows drivers I like...Linux, not so much), or just 2D w/o OpenCL.
(2) Linux will have another regression, right in time for new distro releases (sadly, this actually happens a lot with power consumption--they're still a bit too server-centric), not to be fixed in any commonly-used distro for months. By the time it's fixed, and Intel CPUs gain just as much, claims that it was hampering AMD more will be quietly forgotten about.
(3) MS and Intel conspired to buy off some AMD's BoD, into making bad decisions?

🙂 I use Linux as much as Windows, and it's all just as "bad" there, too.

Don't get too serious, y'all, OK?
 
Warmonger said:
This is far from a architecture change. HSA is a huge stick that AMD is about to wap Intel with. If the software community takes in HSAIL (its already being looked over via C++ extensions), than a measly $150 A10-7800k will over throw the $1,000 i7 3960x in number crunching. Making even the best Intel has to offer at a enthusiast level look like garbage. You know what the on-die iGPU is capable of? The one in the Kaveri A10 will consist of GCN cores and AMD is talking 1 teraflop. This is a master plan AMD had up their sleeve ever since they bought ATI. They can't out perform Intel on the x86 level, but they can still product APU's that can out perform them. It all just hasn't unraveled until now.
Click to expand...

Even if this were true, only code that is so parallel that running it on a GPU makes sense will profit from this. The problem with this is that in any application this will always be a small part and even if that part is 10x times faster the overall gains will be much less. (see the link to x264 and AVX2 in this thread.)

Even worse most applications will not profit from this in any way and most importantly not a browser. So any gains will be seen in niche-products...like facial recognition that seems to be the favorite example used by AMD. video encoding is another thing but there fixed-function hardware is just a lot faster and already available even in smartphones.

Or what I'm saying even if it turns out to be so great as you claim (which I highly doubt) only a handful of niche-applications could actually profit from it.
 
Maybe this is the reason for AMD's stock skyrocket today, though everyone else seems to be thinking it's from rumors of intel buying them out? possibility...
 
Beavermatic said:
Maybe this is the reason for AMD's stock skyrocket today, though everyone else seems to be thinking it's from rumors of intel buying them out? possibility...
Click to expand...

You think AMD's stock gets a bump everytime they pre-release more powerpoint screenies of vaporware? AMD only wishes that were true 😀
 
Idontcare said:
We all know that when it inevitably arrives, later than we were led to expect as well as sporting less features than we were led to expect, the natural course of argumentation will be that it only appears to be a failboat because the apps weren't properly optimized for HSA due to (pick any combination):

(1) Intel compiler being used (gcc should be used)
(2) MS thread scheduler being used (Linux should be used)
(3) insert favorite conspiracy theory of back-room unethical tactics which must have been responsible for stifling AMD's innovation (illegal rebates, showroom girls, the yeti, Intel fanboys, free beenies, etc).

So just sit back and enjoy the show, provided you don't get bored with how formulaic and scripted it may feel at times (like Hollywood movies, the AMD fanboy argument is) because you know it is coming. Give it time.
Click to expand...

I will write some remarks about HSA but first let me say that I completely agree with you!

(1) Everyone knows that FCC decision of settling charges of anticompetitive conduct against Intel biased compiler was completely unfair!

We know that gcc already includes optimizations for Steamroller and future versions of gcc will include extensions for the new HSA technology. But would not be more fair if the intel compiler was used to test HSA? Then all of us would obtain the trivial conclusion of how HSA generates no benefit when HSA is ignored by the Intel compiler.

Unless you convince me that Intel, which is not a member of the HSA foundation, will provide HSA support in its compiler.

The FCC would investigate gcc and other fair compilers by optimizing for both Intel and AMD. That is very unfair for Intel. Compilers would optimize only for intel.

(2) Yes Microsoft recognizing a problem with AMD bulldozer/piledriver modules and trying to fix it by releasing patches was again unfair.

(3) Agree again. All those commissions and courts finding that Intel had violated competition laws in many countries are clearly part of some elaborated conspiracy theory against Intel.

End irony.

Now serious... I find interesting how so many people here do not still get that HSA is no longer an AMD-alone project...

http://hsafoundation.com/
 
1. Why would the Federal Communications Commision be involved with Intel.

2. What IDC was saying went completely over your head.
 
The J said:
The impression I got was that he was talking about discrete GPUs specifically since their memory has to be accessed over the PCIe bus. A different quote mentions being able to choose either DDR3 or GDDR5. I wasn't sure if that meant you can mix the two. I could envision a scenario in which you'd buy, for example, a motherboard with 2GB of GDDR5 soldered to it and then add extra DDR3 like normal. The amount of GDDR5 onboard would make for a nice product differentiator, which the motherboard manufacturers would probably like.
Click to expand...

You just gave me a flashback to the Socket 7 days when the L2 cache came with the motherboard.
 
AMDs most valuable asset isn't their brightest engineer. It's whoever creates their powerpoints. Someone hire that guy ASAP. Just imagine what kind of sales can be had with the grand claims made within, if it actually applied to a company that delivered on their promises 😀
 
I think this is a great idea and an important accomplishment for AMD to have finally worked this out. What worries me is if it will actually be used to its full potential.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top