Bargain Hardware Thread

[petrw1]

...and the rest of you too.

[ldesnogu]

Originally Posted by ewmayer

Oliver's || builds of my Mlucas code in generic-C mode (cf. here) indicate that pthread support is all one needs. (Unless Intel is using MPI to support pthreads - no clue about that, or even if it makes any sense, I've not used MPI, at least not knowingly.)

I'm afraid I don't know any more than you do. But Intel clearly mentions the $2949 cluster edition on this page.

So I take it GCC cannot generate machine code for Phi?

gcc vectorizer isn't up to the task according to Intel, so it's only used to compile the Linux kernel for Phi. Look at the note at the bottom of the above page:

Our changes to the GCC tool chain, available as of June 2012, allow it to build the coprocessor’s Linux environment, including our drivers, for the Intel(R) Xeon Phi(tm) Coprocessor. The changes do not include support for vector instructions and related optimization improvements. GCC for Intel(R) Xeon Phi(tm) is really only for building the kernel and related tools; it is not for building applications. Using GCC to build an application for Intel Xeon Phi Coprocessor will most often result in low performance code due its current inability to vectorize for the new Knights Corner vector instructions. Future changes to give full usage of Knights Corner vector instructions would require work on the GCC vectorizer to utilize those instructions’ masking capabilities.

So I'm afraid the reduced price Phi is basically of no use unless you're ready to spend another $3k :(

[ewmayer]

I pinged Oliver about the compiler issue - his reply:

To my knowledge it is still the case that only the Intel compiler is capable of generating code for the Phi.

Are you sure that the non-supported non-commercial version of the compiler isn't available anymore? I don't know because have a "gold-license" for the software which has no (technical) restrictions at all so I don't care about the non-supported non-commercial version.

About the 31S1P: Keep in mind that

- it requires proper cooling
- to my knowledge it won't run on most consumer boards (Xeon Phi need explicit support from BIOS!)

On the other hand the price for the 31S1P is hot... really HOT. This is what Intel needs... a CHEAP variant of Xeon Phi for experiments. I mean every entry level nvidia card for 50 US$ can do CUDA.. not really fast but you can try.

I wonder if a "designated builder" model would be viable for our potential "local enthusiasts club" here - approach 1 or 2 people with the needed compile tools about their willingness to do-build/shoot-back-executable, then do whatever code development you need locally (using GCC) and when it seems ready for a Phi build, shoot code tarball to builder.

But first we need to settle the issue as to whether the non-com version of ICC is/is-not still available.

[ewmayer]

Originally Posted by ewmayer

I wonder if a "designated builder" model would be viable for our potential "local enthusiasts club" here - approach 1 or 2 people with the needed compile tools about their willingness to do-build/shoot-back-executable, then do whatever code development you need locally (using GCC) and when it seems ready for a Phi build, shoot code tarball to builder.

Or better, someone with the needed tools willing to provide guest accounts for remote logins.

But first we need to settle the issue as to whether the non-com version of ICC is/is-not still available.

Non-Commercial Software Development | Intel

Intel C++ Compiler | Wikipedia

Mt main specific interest here is in the coming AVX512 step in the evolution of Xeon Phi, but I also have a general interest in the emerging manycore paradigm, which IMO (and this was Intel and AMD's own doing, or better non-doing) has been overly centered on nVidia. Competition here can only be good: nVidia's efforts have forced mighty Intel to race and try to catch up, and Intel's efforts will (hopefully) lead nVidia to make their software tools less reliant on proprietary API-hooks - as I've said elsewhere, one thing I think Intel got right is the "take your standards-compliant multithreaded code, let our compiler map it to our custom hardware" use model. Imagine how "portable manycore" code of (say) 5-10 years out would look if each of a half-dozen major manycore vendors required their own custom parallel-coding API to be used.

[LaurV]

I am still in for one, in case someone collects the money to buy a carton box of them. I may be throwing away ~130 bucks (plus taxes) and just store it in the cabinet until its time will come, but well, there is a small risk I can afford. It may be that I could use it in the future, even learn something from it trying to write small pieces of code. It may be not...

[kladner]

Refurb sale at New Egg.

[petrw1]

===== I HAVE =======
I currently have a 4-port KVM ... Keyboard/Video/Mouse.
4 PC's ... 1 Keyboard / 1 Mouse / 1 Video Monitor

I plug it into 4 PCs and at the push of a button I can control which of the 4 is controlled by Keyboard/Mouse and which is displayed on my monitor.

======== I WISH I HAD ======

HOWEVER...now I am dreaming and envisioning a device (hardware/software/firmware) that can cut my monitor into 4 quadrants and display the screens for all 4 PCs at the same time.

AND....depending where on the screen I move my mouse will determine which PC interacts with the Keyboard/Mouse.

I realize I would need a big monitor with good resolution...and very good Glasses :)

Has anyone ever heard of such a thing?
Or is this just a lot of wishful thinking?

[axn]

Originally Posted by petrw1

======== I WISH I HAD ======

HOWEVER...now I am dreaming and envisioning a device (hardware/software/firmware) that can cut my monitor into 4 quadrants and display the screens for all 4 PCs at the same time.

AND....depending where on the screen I move my mouse will determine which PC interacts with the Keyboard/Mouse.

I realize I would need a big monitor with good resolution...and very good Glasses :)

Has anyone ever heard of such a thing?
Or is this just a lot of wishful thinking?

Four remote desktop / VNC windows? Wouldn't that take care of our requirement?

[kracker]

Originally Posted by axn

Four remote desktop / VNC windows? Wouldn't that take care of our requirement?

+1
TeamViewer FTW.

[TheMawn]

Two GTX 580's apparently never overclocked, waterblocks AND original coolers included, bidding at $100 US and buyout at $250 US. I've never strayed into used parts before (I do look occasionally though) but this deal is tempting me... I don't know if I can manage the extra heat into the liquid cooling system or the noise of the air coolers though.

[Mark Rose]

Originally Posted by TheMawn

Two GTX 580's apparently never overclocked, waterblocks AND original coolers included, bidding at $100 US and buyout at $250 US. I've never strayed into used parts before (I do look occasionally though) but this deal is tempting me... I don't know if I can manage the extra heat into the liquid cooling system or the noise of the air coolers though.

Each GTX 580 will dump 200-250 watts into your coolant loop at stock clocks. That's a lot of radiator to add. I've never done water cooling because it's way cheaper to run on air.

I picked up a GTX 580 for $75 CAN on Saturday, off Kijiji. That's about $60 US at the moment. I had an available power supply that could handle it, so it was too tempting to not buy. I see another for sale at $80 CAN OBO, and this one on eBay for $72 CAN shipping included.