[Beowulf] NVIDIA buying PGI

Joe Landman landman at scalableinformatics.com
Mon Aug 5 06:47:12 PDT 2013


Would have replied earlier, but spent Thursday night -> Sunday 
recovering from an injury due to sparring.  I've learned that I have 
intercostal muscles, and I can injure them (/sigh).

On 08/01/2013 05:39 PM, "C. Bergström" wrote:
> On 08/ 2/13 04:21 AM, Joe Landman wrote:

[...]

>> Ok, quickie poll on Beowulf:
>>
>> a) whom is running Intel and/or AMD kit?
>> b) whom is considering running new AMD kit?
>> c) whom would reconsider if PG end-of-lifed (which I think is Dave's
>> speculation) the AMD targeted compiler?
>> d) whom doesn't really care about PG's eol status
> d), but I'm biased ;)
>
> No need for this conversation to be heated - It's clear this is a
> strategic acquisition meant to strengthen their (NV) market position and
> has the *potential* to define future programming models.
>
> More importantly - do you care if PG doesn't support OpenMP 4? What
> really makes them unique enough for people to care?

I should point out, as it probably didn't come across, that I am a huge 
fan of the PG compiler suite.  I've been a paying customer in the past, 
and somewhat regret that we don't have enough projects that could make 
use of it so that I could justify the cost.  Same issue with the other 
non-free compilers.  Gcc and related handle most of my compilation these 
days, simply because I don't do so much of that any more.

This said, I think its important to separate the facts from the emotion, 
and I wasn't getting heated about this conversation.  Replying with 
incredulity that someone whom I highly respect would strongly suggest 
something that sounds ... well ....

Facts are we don't know PG's plans.  We do know that NVidia is getting 
into ARM in a big way.  And that PG has some very cool technologies to 
make a single code base across accelerators and substrate computing 
(where you plug the accelerators into).  So one might reasonably suppose 
that PG will *increase* their ARM support, especially for the NVidia 
variants.

Facts are we don't know NVidia's plans for PG.  I think they see CUDA as 
being a potentially limiting factor for code adoption (violates 
everyone's "avoid vendor lockin" mantra), and needed to get better at 
saying "hey it just works, try it with this fancy compiler, toss these 
directives in it, and your code will run like normal on normal systems, 
and accelerated on accelerated systems."

Basically I am saying this was a smart move on NVidia's part, a smart 
tie up for PG, and generally speaking, I believe it to be good, at least 
on its face, for the industry.

For those whom aren't aware, I am a huge proponent of accelerators and 
accelerator technology, and I loved PG's OpenMP like approach prior to 
OpenMP addressing the issue.  I am also an OpenMP fan.

Does this acquisition reduce customer choice?  Not as of yet, and its 
not obvious it will.  It could, and companies could decide to stop 
supporting (insert random platform here) for (insert random reason 
here), which "reduces choice in market" due to business considerations 
(cost of continuing those operations, acquisition, strategic focus, 
...).  Is removal of product support anti-competitive?  Only when the 
product is a dominant player in the market.  I think it would be hard to 
argue that PG was dominant.  They are good, of that there is no doubt. 
But I don't think they've been spectacularly profitable.  Tools 
companies rarely are.

Basically, I am saying I simply don't see it, even with the most 
generous reading of the situation ... I simply cannot imagine that this 
is anticompetitive in any way shape or form.

Moreover, as Intel owns/controls MIC, and a processor platform set, and 
a compiler, if anything, a regulator would have a very hard time arguing 
that NVidia cannot gird itself the way its competitors do.

I like PG, and I'd love to have a few more projects where I could use 
them (same of Intel, PathScale, etc.).   I am not worried about "loss of 
choice" in the market.  I don't think anyone should.


-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
        http://scalableinformatics.com/siflash
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615


More information about the Beowulf mailing list