As announced some time ago on the forum, the move of the linux kernel towards HSA is now entering a final phase:
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=5cc9ed4b9a7ac579362ccebac67f7a4cdb36de06
This commit for kernel 3.16 speaks for itself really...
Don't buy Asus mobos with locked IOMMU or Intel k-parts without VT-d guys!
I am looking forward to hsa, though I need to research deeper research on hsa. I do think my laptop supports it, dell latituded d530 with a core2 duo t7250
I'm still waiting for the reason why high end PCs dont have a socket 1150 K CPU in it.
I have a 4770 and vmware flies on it. I couldn't give a frosty onion on linux compatibility, unless steamOS becomes what it promises.
There's no point in buying a K knowing it's a crippled but unlocked multiplier cpu unless you want to try your luck and overclock with proper (water) cooling. Before this K nonsense every CPU had a downwards unlocked multiplier and overclocking was done through FSB and playing with memory ratio pairings and timings. BCLK (base clock) overclocking is minimal and not as fun as back in the old days. OCing sucks and shouldn't be supported in the current state.
Hmm, interesting this should be quiet an improvement. On another note with all of these Computex videos, it makes me sad how their marketing points are pieces of crappy software that's clearly going to be abandoned, it's a shame really; and with the Ausustor what kind of unholy abomination of a UI this that! Some of the apps are clearley closed source and would never be ported to linux but the NAS also has XBMC on it. i'm confused now.
Very Cool. So glad I went with a non-K i5 for my gaming rig and just dropped one of the latest AMD quad-core APU's into my latest build.
I'm already blown away by the efficiency of Linux. (Yeah I'm a new Linux user...lol) Last night I tried throwing all kinds of things at the Athlon 5350 all at once and it never maxed out (CPU usage) nor slowed down. On top of that it was barely using 2GB ram. Very eager to see how it runs with HSA.
This will be great in five years when developers catch up. By that time hardware will be worth investing in, but until then it's just a good idea.
I really like what AMD is doing with technology these days, HSA and Mantle seem like very good things to come, more so on Linux than Windows.
Just want AMD to get on the chromebook train, imagine a low end APU on Chrome OS, using an ARM CPU to keep costs and have HSA enabled (Possible on ARM?) and maybe BIG.little cores to help with battery life, would be rather amazing :)
Zoltan, Why dont you explain what this means to the average user who does not keep them self "up to date" with the latest on the Linux front. BTW 4790k does not have the same limitations that the 4770k had imposed.
I am also very confused. As fas as i thought to know, HSA was only available on AMD CPUs. Since when does this apply to intel and why is HSA bad news for the K processors?
Means nothing for the general user.
I'm sure others could explain it better than I and in more detail but what it basically boils down to is this:
Certain compute tasks can be performed MUCH faster using GPU cores rather than CPU cores. But right now we're not tapping into that potential. HSA (heterogeneous systems architecture) basically will allow for the use of the on-board iGPU cores to be used for such tasks/processes, thus greatly increasing the speed at which those tasks are completed. While those tasks are off-loaded to the GPU cores, the CPU cores are free to handle other tasks, significantly reducing CPU overhead in the process. This gives "multi-tasking" a whole new meaning.
A necessary part of how this works is the ability for the GPU cores to have full, direct access to the memory. I'm not 100% sure on this, but I believe IOMMU (input/output memory management unit) and VT-d (virtualization technology for directed I/O) has something to do with making this work on Intel CPUs, while the newest AMD APUs already have this architecture built in.
In a nutshell, it means much faster processing/computing.
Well this is clear to me. I am currently working with openCL on dedicated graphics cards. What is not clear is how intel iGPUs can profit from HSA, because as far as I know they do not support native HSA. In AMD APUs, every bit of the system RAM can be accessed by the GPU units and the CPU at the same time. In Intel (at least thats what I think) there is n 'area' in your RAM dedicated for the iGPU, which is not so easily accessible by the CPU and vice versa. Could as well be that I am wrong here. Someone please enlighten me :-)
Where do the K vs non-K parts come into play here?
Well, as I said, believe this is where IOMMU and VT-d comes into play for allowing Intel to take advantage of HSA.
The non-K parts have VT-d and the K parts do not. Thus the K parts will not be able to take advantage of this.
Well the new K parts have VT-d now so its not really a problem anymore.
Source: Intel Ark
http://ark.intel.com/products/80807/Intel-Core-i7-4790K-Processor-8M-Cache-up-to-4_40-GHz?ui=BIG
HSA is supported on ARM devices
http://www.hsafoundation.com/
(Check founders)
I'm just wondering something.
When you are doing bigger HSA processing (something that can't fit into cache) it need to access it from the main memory. AMDs APUs are already suffering from low-frequency bandwidth, so having some GCN cores taking bandwidth will result in even more bottleneck.
Another reason why AMD will benefit from DDR4.
I had heard that. But people will still be buying 4670K's and 4770K's, especially if they are on sale now that their replacements are surfacing.
Seems like Intel has taken a page from AMD's book lately. Don't remove features from parts that cost a premium and allow overclocking on all mainstream boards.
Well from Broadwell onward we should see the K parts continue to retain the virtualization extensions from now on, hopefully.
Also its nice to know that Intel does listen to feedback
Sure they are restricted by the lower-frequency memory bandwidth in gaming. But we're not talking about gaming, we're talking about other tasks that don't necessarily need super-fast memory. So that really isn't too much of an issue.
DDR4 will benefit everything - especially APUs.