GPU RAID

[5950x]

Gpu raid has been announced, and there are two different types. The only requirements are as follows: Volta architecture and 2 or more cards in SLI (crossfire is not supported [as if it would be]) 

Raid 0: the VRAM is configured into 2 or more seperate caches similar to regular raid 0 for an effective doubling of vram speed. 

Raid 1: the VRAM is configred into a redundant array effectively making it less vulnerable to errors (such errors are computational miscalculations that can actually result in, believe it or not, dead pixels in your monitors)

This are the only configurations that have been anniunced, but more types are promised, although gpu raid cards may be required

The raid cards will come in many forms, such as: pcie 8x, USB 3.1, U.2, and thunderbolt 3

This announcement is revolutionary in the way of graphics computing, thank yo. u for your attention to the topic.

[NumLock21]

Link?

And is 5950X some cpu Intel didn't tell us about?

 

[5950x]

3 minutes ago, NumLock21 said:

And is 5950X some cpu Intel didn't tell us about?

 

Yes, knight's landing

[NumLock21]

3 minutes ago, 5950x said:

Yes, knight's landing

You mean Knight's falling, cause that model don't exist.

And link to article

[5950x]

 

5 minutes ago, NumLock21 said:

You mean Knight's falling, cause that model don't exist.

https://www.nextplatform.com/2016/06/20/intel-knights-landing-yields-big-bang-buck-jump/

[NumLock21]

Just now, 5950x said:

 

https://www.nextplatform.com/2016/06/20/intel-knights-landing-yields-big-bang-buck-jump/

Did a search on that link, 5950x, does not exist.

[5950x]

3 minutes ago, NumLock21 said:

Did a search on that link, 5950x, does not exist.

High end server software isn't always well documented

[Mira Yurizaki]

The only Intel 5950X is the Core-i7 5950X. No Knights Landing has a model number in the 5000 range. Also GPU RAID is an old idea: http://ieeexplore.ieee.org/document/5599249/

 

And at this point, I don't even understand what's being said here.

32 minutes ago, 5950x said:

Raid 0: the VRAM is configured into 2 or more seperate caches similar to regular raid 0 for an effective doubling of vram speed. 

On the board itself, VRAM already works like this. VRAM is essentially a scaled up memory RAID 0 system. If you're talking about getting two video cards together to have their VRAM pool combined, then that's a different story. However, the speed won't be doubled unless they have a bus connecting the two that is capable of handling the bandwidth (which no interconnect bus has that capability so...)

 

Quote

Raid 1: the VRAM is configred into a redundant array effectively making it less vulnerable to errors (such errors are computational miscalculations that can actually result in, believe it or not, dead pixels in your monitors)

Which okay, but this is what ECC memory is for, and is what's on workstation cards. I mean, other than reducing overhead time to correct the error, I don't really see a benefit of this because you're asking HPCs to chop their memory by at least half to solve a problem that may not really help them for the cost (i.e., what kind of gain do  you get using ECC + RAID vs. just ECC?)

 

And of course, I'd really like a link to this as well.