impact of Athlon's slower front-side-bus (FSB)

impact of Athlon's slower front-side-bus (FSB)

[joe briggs]

Gentlemen - 

Can anyone provide arguments, evidence, or guidance regarding the followng:

The fastest AMD single processor Athlon XP is 3200 with 400 Mhz FSB.
The fastest AMD dual processor Athlon MP is 2800 but with only 266 Mhz FSB.

So, for a multimedia application, which platform would be faster?  How does 
the much slower FSB of the dual processor impact its ability to grab and 
crunch.  Does its onboard cache make the slower speed FSB less important?  

Also, does a dual processor platform distribute the interrupt loading as well 
as process loading?  I my systems I have between 1 and 8 frame identical 
frame grabbers.  Would the interrupt processing of these devices be 
distributed evenly on the dual processor platforms?
-- 
Joe Briggs
Briggs Media Systems
105 Burnsen Ave.
Manchester NH 01304 USA
TEL/FAX 603-232-3115 MOBILE 603-493-2386
www.briggsmedia.com

Re: impact of Athlon's slower front-side-bus (FSB)

[Alan Cox]

On Llu, 2003-06-02 at 14:47, joe briggs wrote:
> The fastest AMD single processor Athlon XP is 3200 with 400 Mhz FSB.
> The fastest AMD dual processor Athlon MP is 2800 but with only 266 Mhz FSB.
> 
> So, for a multimedia application, which platform would be faster?  How does 
> the much slower FSB of the dual processor impact its ability to grab and 
> crunch.  Does its onboard cache make the slower speed FSB less important?  

Its really hard to tell. The 3200 has a bigger cache too if I remember
rightly. If you are planning on buying big boxes for this you might want
to ask the vendor if you can do a test run or two.

> Also, does a dual processor platform distribute the interrupt loading as well 
> as process loading?  I my systems I have between 1 and 8 frame identical 
> frame grabbers.  Would the interrupt processing of these devices be 
> distributed evenly on the dual processor platforms?

Yes. You would probably want to tie different cards/encoders to
different processors and the IRQ to the same one. You can do this via
/proc and with the -ac or most vendor trees (and 2.5) you can tie
processes to CPUs with syscalls

Re: impact of Athlon's slower front-side-bus (FSB)

[joe briggs]

> Yes. You would probably want to tie different cards/encoders to
> different processors and the IRQ to the same one. You can do this via
> /proc and with the -ac or most vendor trees (and 2.5) you can tie
> processes to CPUs with syscalls

Can I do this with the 2.4.19 kernel (debian)?  The cards in question are quad 
bt878 frame grabbers. How specifically can I tie a particular bt878 to a 
particular processor on the dual athlon platform?  

One last question, given the slow FSB and the fact that 2 uP's are groping for 
the same memory space and that each bt878 is dma'ing its data to memory, is 
the SMP still a better idea than uni-processor?

Thanks for all of the help!
-- 
Joe Briggs
Briggs Media Systems
105 Burnsen Ave.
Manchester NH 01304 USA
TEL/FAX 603-232-3115 MOBILE 603-493-2386
www.briggsmedia.com

Re: impact of Athlon's slower front-side-bus (FSB)

[Doug McNaught]

joe briggs <jbriggs@briggsmedia.com> writes:

> Can I do this with the 2.4.19 kernel (debian)?  The cards in question are quad 
> bt878 frame grabbers. How specifically can I tie a particular bt878 to a 
> particular processor on the dual athlon platform?  

I don't think so but building a kernel package with an -ac kernel (or
any other version) is dead easy on Debian--don't let that stop you.

> One last question, given the slow FSB and the fact that 2 uP's are
> groping for the same memory space and that each bt878 is dma'ing its
> data to memory, is the SMP still a better idea than uni-processor?

If there's any way to actually test both configurations, I'd do
so--there are enough variables here that random handwaving arguments
aren't going to be really useful.

-Doug

OT: Re: impact of Athlon's slower front-side-bus (FSB)

[William Gallafent]

On Mon, 2 Jun 2003, Alan Cox wrote:

> On Llu, 2003-06-02 at 14:47, joe briggs wrote:
> > The fastest AMD single processor Athlon XP is 3200 with 400 Mhz FSB.
> > The fastest AMD dual processor Athlon MP is 2800 but with only 266 Mhz FSB.
> >
> > So, for a multimedia application, which platform would be faster?  How does
> > the much slower FSB of the dual processor impact its ability to grab and
> > crunch.  Does its onboard cache make the slower speed FSB less important?
>
> Its really hard to tell. The 3200 has a bigger cache too if I remember
> rightly.

Yes, 512KB vs 256KB. The top "Barton" MP (2800+?) also has 512KB L2 onboard,
and uses the same core as the top Athlon XPs, but sticks to 266MHz FSB since
this is what's supported by 760MP(X) chipset.
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26426.PDF

-- 
Bill Gallafent.