Be Tim Cook for a day, which GPU brand for new Mac Pro?

[deconstruct60]

Yawn... And in never hit you that maybe NVidia just pulled the rug from under the competition by buying TPG?

I had missed that recent purchase .... but rug under from who????

" ... He also stressed that Nvidia would continue to work with TotalView, CAPS, Cray, Allinea, and other compiler partners, and that nothing would change in this regard in the aftermath of the PGI acquisition. ... "
http://www.theregister.co.uk/2013/07/30/nvidia_buys_the_portland_group/

Nvidia is saying that they are NOT trying to flush the large fraction of PGI's assets down the drain by pulling some myopic, bonehead move.

They are playing catch up to Intel ....

""What we don't have is a world-class HPC compiler team on the scale and with the kind of products that PGI is offering,' Ian Buck, general manager for the CUDA compiler stack at Nvidia, ... "

and

" ... "PGI has experience with ARM," says Buck, "but there is no commercial Fortran compiler available – yet." ... "

Cuda, according to your figures represent 3/4 of the market for accel/copro. The result for a quarter does not make a trend.

Chuckle there are no results for Xeon Phi prior to November 2012. Intel decided to be in the market... jumped in and took 20% with their first generation product. Including jumping to the #1 spot on the top 500 list by kicking the crap out of the biggest, baddest K20 leveraged system there was. If you think that doesn't have Nvidia running playing defensive, you are smoking something.

PGI is a way for Nvidia to make money if folks don't buy into their proprietary CUDA and their hardware. Like even they say it also fills a glarig world class software gap in their skill set. It is yet another diversification move. It isn't doubling down on narrow proprietary standards of the CUDA tarpit. PGI's approach of covering extracting parallelism wherever it exists is why they make money (or not if they can't extract the parallelism).

 

[tuxon86]

I had missed that recent purchase .... but rug under from who????

" ... He also stressed that Nvidia would continue to work with TotalView, CAPS, Cray, Allinea, and other compiler partners, and that nothing would change in this regard in the aftermath of the PGI acquisition. ... "
http://www.theregister.co.uk/2013/07/30/nvidia_buys_the_portland_group/

Nvidia is saying that they are NOT trying to flush the large fraction of PGI's assets down the drain by pulling some myopic, bonehead move.

They are playing catch up to Intel ....

""What we don't have is a world-class HPC compiler team on the scale and with the kind of products that PGI is offering,' Ian Buck, general manager for the CUDA compiler stack at Nvidia, ... "

and

" ... "PGI has experience with ARM," says Buck, "but there is no commercial Fortran compiler available – yet." ... "




Chuckle there are no results for Xeon Phi prior to November 2012. Intel decided to be in the market... jumped in and took 20% with their first generation product. Including jumping to the #1 spot on the top 500 list by kicking the crap out of the biggest, baddest K20 leveraged system there was. If you think that doesn't have Nvidia running playing defensive, you are smoking something.

PGI is a way for Nvidia to make money if folks don't buy into their proprietary CUDA and their hardware. Like even they say it also fills a glarig world class software gap in their skill set. It is yet another diversification move. It isn't doubling down on narrow proprietary standards of the CUDA tarpit. PGI's approach of covering extracting parallelism wherever it exists is why they make money (or not if they can't extract the parallelism).

We'll see... See, short and to the point.

 

[Tesselator]

Indeed, but you know that the world does not work like this. In this particular case, there must be a financial reason why have chosen to discontinue the current Mac Pro form factor. It definitely was not a personal snub at the 150 MR users that want the current shape to stay.

Tru dat...

 

[MacVidCards]

So, pretty "evenly split" at nearly 3 to 1 in favor of Nvidia over AMD. (Tongue in cheek, for those who see only words on surface)

The Fire Sale on Fire-Pros must have been pretty good.

I know some say that skipping CUDA was good, but choosing the TB tarpit in place of CUDA was an odd choice.

 

[goMac]

We'll see... See, short and to the point.

And indeed, NVidia recently announced they are building an APU based Power processor.

http://techreport.com/news/25191/ib...u-architecture-integrate-with-nvidia-gpu-tech

 

[deconstruct60]

And indeed, NVidia recently announced they are building an APU based Power processor.

Actually just IBM and Nvidia just basically said their are licensing their portions. (doesn't look like Mellanox is licensing their IP; just networking controllers chips). Who is actually an implementer is rather vague at this point.

The likely implementer using both so far is IBM. It wouldn't be surprising if next generation BlueGene implementation package to weave in a GPGPU into a POWER 8 foundation. BlueGene variants have a much better track record in HPC context than anything ARM is offering now or in the future.

Still in the same context as playing catch up to Intel which has all the pieces here ( high speed interconnect [ Cray interconnect, Infiniband/High end Ethernet ] , general purpose CPU [ Xeon ] , pure GPGPU computations engine [Xeon Phi] ) all in house.

It is an odd group. It an odd mix of licenses too. IBM is throwing next gen POWER 8 ( 22nm ) on the table and Nvidia is throwing relatively soon to be last gen tech on the table on their side (Kepler @ 28nm ).

That the open-power.org web site still isn't up yet .... "we'll see" ... indeed.