Обсуждение: Re: [PERFORM] Projecting currentdb to more users

From AMD's suit against Intel. Perhaps relevant to some PG/AMD issues.

"...125. Intel has designed its compiler purposely to degrade performance when a program
is run on an AMD platform. To achieve this, Intel designed the compiler to compile code
along several alternate code paths. Some paths are executed when the program runs on an Intel
platform and others are executed when the program is operated on a computer with an AMD
microprocessor. (The choice of code path is determined when the program is started, using a
feature known as "CPUID" which identifies the computer's microprocessor.) By design, the
code paths were not created equally. If the program detects a "Genuine Intel" microprocessor,
it executes a fully optimized code path and operates with the maximum efficiency. However,
if the program detects an "Authentic AMD" microprocessor, it executes a different code path
that will degrade the program's performance or cause it to crash..."

On Tue, 2005-07-12 at 13:24, Mohan, Ross wrote:
> From AMD's suit against Intel. Perhaps relevant to some PG/AMD issues.
>
> "...125. Intel has designed its compiler purposely to degrade performance when a program
> is run on an AMD platform. To achieve this, Intel designed the compiler to compile code
> along several alternate code paths. Some paths are executed when the program runs on an Intel
> platform and others are executed when the program is operated on a computer with an AMD
> microprocessor. (The choice of code path is determined when the program is started, using a
> feature known as "CPUID" which identifies the computer's microprocessor.) By design, the
> code paths were not created equally. If the program detects a "Genuine Intel" microprocessor,
> it executes a fully optimized code path and operates with the maximum efficiency. However,
> if the program detects an "Authentic AMD" microprocessor, it executes a different code path
> that will degrade the program's performance or cause it to crash..."

Well, this is, right now, just AMD's supposition about Intel's
behaviour, I'm not sure one way or the other if Intel IS doing this.
Being a big, money hungry company, I wouldn't be surprised if they are,
but I don't think it would affect postgresql for most people, since they
would be using the gcc compiler.

On Tue, Jul 12, 2005 at 01:41:14PM -0500, Scott Marlowe wrote:
> On Tue, 2005-07-12 at 13:24, Mohan, Ross wrote:
> > From AMD's suit against Intel. Perhaps relevant to some PG/AMD issues.
> Well, this is, right now, just AMD's supposition about Intel's
> behaviour, I'm not sure one way or the other if Intel IS doing this.

I think its more a case of AMD now having solid evidence to back
up the claims.

This discovery, and that fact that you could get round it by
toggling some flags, was being discussed on various HPC mailing
lists around about the beginning of this year.

    -Mark

On Tue, 2005-07-12 at 15:06, Mark Rae wrote:
> On Tue, Jul 12, 2005 at 01:41:14PM -0500, Scott Marlowe wrote:
> > On Tue, 2005-07-12 at 13:24, Mohan, Ross wrote:
> > > From AMD's suit against Intel. Perhaps relevant to some PG/AMD issues.
> > Well, this is, right now, just AMD's supposition about Intel's
> > behaviour, I'm not sure one way or the other if Intel IS doing this.
>
> I think its more a case of AMD now having solid evidence to back
> up the claims.
>
> This discovery, and that fact that you could get round it by
> toggling some flags, was being discussed on various HPC mailing
> lists around about the beginning of this year.

Wow!  That's pretty fascinating.  So, is the evidence pretty
overwhelming that this was not simple incompetence, but real malice?

I could see either one being a cause of this issue, and wouldn't really
be surprised by either one.

On Tue, Jul 12, 2005 at 03:11:35PM -0500, Scott Marlowe wrote:
> On Tue, 2005-07-12 at 15:06, Mark Rae wrote:
> > I think its more a case of AMD now having solid evidence to back
> > up the claims.
>
> Wow!  That's pretty fascinating.  So, is the evidence pretty
> overwhelming that this was not simple incompetence, but real malice?


I suppose that depends on the exact nature of the 'check'.

As far as I was aware it was more a case of 'I don't recognise this
processor, so I'll do it the slow but safe way'.

However from what AMD are claiming, it seems to be more of a
'Its an AMD processor so I'll be deliberately slow and buggy'


Having said that, I have tried compiling PG with the intel compiler
in the past, and haven't noticed any real difference. But in a database
there isn't much scope for vectorization and pipelining
compared with numerical code, which is where the Intel compiler
makes the greatest difference.

    -Mark