[wplug] It's all about the company focus

[Bryan J Smith]

Understand my "picking on this" is not to be "right" or to boast anything.
It is from years of helping clients with integration projects.
And that comes down to market focus, which affects everything.

Where is Intel's focus?
Where is nVidia's focus?
Where is AMD's focus?

Intel's is simple.

90% of Intel marketshare is selling a new PC to home Windows
users every 2-3 years. Those home PCs are fully integrated designs.
They push everything, even their own mainboard, to PC OEMs.

What does that mean to Linux?

Well, it means both good and bad. The good is that with 50%+ of the
PCs being these "stock" designs, once drivers are developed, they work
for so many boards.

The bad?  With Intel so focused on Wintel bare-simple
PCs, their engineers have a nasty
habit of rev'ing things without
considering how compatible the changes. are with existing drivers.
I've personally see nagging issues with newer revisions of the ICH7, let
alone non-support of ICH8, ICH9, etc... perpherials, including "false positive"
identification of a chipset revision.
I.e., it loads the driver, and then the problems start.

Unless you have the latest kernel, and sometimes not even then,
you run into issues. Furthermore, timing and output changes on the
video are a continuing issue, often not found in X11 until products ship.
Again, "Linux is an afterthought" at Intel, and I feel for the Linux
support team at Intel because of this continually "non-consideration."

It does not surprise me one bit that Dell started shipping Ububtu.
RHEL and SLES/NLD, despite updates/backports, just cannot
"keep up" with Intel chipset and GPU rev's, especially their older X.
They must ship a leading edge release, and Fedora does not have 
commercial support, so Conocial's
Ubtuntu LTS became the obvious.

Furthermore, Broadcom gets bashed for its proprietary WLAN firmware,
but I don't see the same happening to Intel when they have the same thing.
It's a FCC requirement every since someone took the Prism ASIC
and started using it well out-of-band.

Now let's look at nVidia.

nVidia's most profitable and staple venture is Quadro and nForce Pro.
They sell, including their own reference systems now, to the high-end.
And that right there is 90% Linux, a complete 180 from Wintel.
That means Linux support is _immediate_ in the "critical path"
of any new nForce Pro and
Quadro product - of which nVidia provides as much hardware
backward compatibility as possible.

Their consumer GeForce and nForce products are the same logic,
so they result in the same levels of compatibility and support.
They GPL everything they can, put guys on the MIT 2D drivers.
nVidia is always well ahead of ATI in new product support for MIT 2D.

The 3D approach is the same they've had ever since their full
disclosure and source code release for XFree86 3.3 - no sense in
possible legal issues for substandard support anyway.
GPUs double in performance every 9-12 months, major GPUs ISAs change
every 18-24 months.  CPUs double every 18-24, and major ISA
changes don't happen for 6-9 years.

More relevant to my point ...

I will have a client that orders Intel workstations from Dell
allegedly "certified" with RHEL or SLES/NLD - bam! Issues on first boot.
There is a reason why Intel started supporting AMD, and it wasn't cost.
It was the fact that Opteron-nVidia solutions "just work" out-of-the-box.

I regularly told my last client in New York to return the Xeon-Intel
Dell boxen and order Opteron-AMD ones. Hesitant at first, they'd
get just one and, sure enough, I was correct. And even performance
of the Opteron solution ended up
being superior because they are
hitting the 3+3-issue ALU and FPU
of the Opteron, and not running SSE or other inprecise math used in gaming.
I believe Dell now offers nVidia-based chipset Intel solutions too,
which is another way they are addressing the "Linux is an
afterthough" issue with Intel's "commodity" hardware designs.

There is a reason why I have a Turion x2 with nForce/GeForce
and not a Core 2 Duo with ICH/MCH+IGP - on many levels.
Yes, I don't like the Broadcom firmware load issue either, but
I'd still have an Intel firmware load issue as well with Intel.

AMD is the new kid on the GPU block and nothing could make me happier.
AMD clearly bought ATI because they _know_ the "money" is at the
"high-end" integrated solution, and they had to buy a GPU vendor to get
GPUs on their HTX and away from PCIe. AMD donimates supercomputing
because of the Infiniband HTX option, which blasted Intel PCI-X options
(yes, PCIe closes some gap, but there is still the platform/coherency issues).

And that is why AMD is "getting serious" about Linux now.
Because the space they want is Linux-dominated.
We still don't have even an HT-integrated GPU design yet
(the G690 is not it, but ATI's final chipset design, with various
Linux incompatibility issues), but I can't wait until we see the first one.

And you can be sure that nVidia, despite ATI being owned by AMD,
is working with AMD to release a HTX GPU product.
Although I wouldn't rule out a possible HT-native "commodity" chipset
solution from nVidia, it's really a matter of marketshare and how much
they could gain over the AMD-ATI solutons which will dominate.

These are things most people just don't consider, but the _do_
explain _everything_. ;)

--  
Bryan J Smith - mailto:b.j.smith at ieee.org  
http://thebs413.blogspot.com  
Sent via BlackBerry from T-Mobile