[Leaplist] Do *NOT* upgrade nVidia to latest 196 drivers ...

[Dan Cherry]

Bryan J. Smith wrote:
> I've already seen a report on another board about systems
> hanging in 2 minutes when running the "nvidia" driver.
> Considering nVidia's binary objects are unified across all
> platforms, it's very likely the same culprit:  
> 
> http://blogs.zdnet.com/hardware/?p=7551&tag=content;col1  
> 
> Although it's been yanked, several 3rd parties may have
> built the driver.  I offer my "educated guess" below on what
> is happening based on reports.
> 
> --- Forward ---
> 
> No, it's very likely _related_!
> 
> nVidia uses an _unified_, binary kernel memory-GPU object plus
> user-space objects for all Oses, with only select interface
> differences being the difference for the OSes.  So if it affects
> one OS, it's very likely to affect _all_ OSes.  It's the main
> way nVidia avoids all IP and porting issues, and has for the
> last decade.
> 
> CPUs and GPUs over the last 5+ years will automatically scale
> down performance based on thermal/environmental conditions.
> Since these are nominal operations -- especially in GPUs -- they
> will not be audited (unless debugging is turned up).  In this
> case, the thermals are quickly going through the roof, so the
> GPU starts shutting down units to the point of a hang.
> 
> Sounds like a faulty set of conditionals in the nVidia control
> systems logic, one that is wrapped up in those unified objects
> (and doesn't need to be).  If there was ever a reason for
> nVidia to consider striping the binary object down to only the
> essential IP/ABI portions, and open source everything else,
> this is one.
> 
> Then again, I've seen open source drivers dork things like this
> before as well -- everything from DACs and scan ranges to control
> of the fan and output.  Years of dealing with Intel has taught
> me that it's no less likely to happen on open source -- although
> it does show how the unified object in nVidia's world can quickly
> mess up _all_ OSes in short order.  ;)
> 
> -- Bryan
> 
> P.S.  At least in Linux, you can bring the OSes up without the
> nVidia driver and X at all.  In Windows, even safe mode may not
> launch Windows without the driver.  This leaves people in a
> circular state that cannot be solved.  Oh boy, this was a major
> oversight!
> 
> 
>  -- 
> Bryan J  Smith           Professional, Technical Annoyance 
> Linked Profile:         http://www.linkedin.com/in/bjsmith 
> ---------------------------------------------------------- 
> Red Hat:  That 'other' American software company built on
> open customer selection of options and value, instead of
> controlled distribution channels of forced bundle and lock
> 
> 
Thanks for the heads up!
Dan

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.