[rescue] point of pride in SGI R10k vs. 1Ghz x86...

[rescue at sunhelp.org]

Yeah, SGIs have a nice memory subsystem for their time, but it's not
really all that great.  I don't know what the memory bandwith on 60ns
62pin FPM simms is, but the indy/i2 use 4way interleaving to seriously
increase it, so I suspect they have something like
500-800meg/second.  this is compared to sdram at ~1gig/second and ddr at
~2gig/second.
	Nick

On Tue, 13 Nov 2001, Dave McGuire wrote:

> On November 13, Patrick Giagnocavo wrote:
> > >   I wonder, Patrick...in this case, was gimp compiled to take full
> > > advantage of the R10K processor, with a decent-for-MIPS-arch
> > > (i.e. SGI's, not GCC) compiler?  If not, it's likely that it could do
> > > even better.
> > 
> > I grabbed the binary version of Gimp 1.2 from freeware.sgi.com .  My
> > guess is that it was SGI's compiler.
> 
>   In that case, it's likely to have been compiled to run on any
> processor from the R4K up.  Compiling gimp is a bit of a pain, but you
> might see significant performance gains when compiling specifically
> for the R10K...so it might be worth the trouble!
> 
> > I think the 1MB of cache doesn't hurt, along with the higher FLOPs you
> > get.
> 
>   ...and the fact that the architecture doesn't suck. :)
> 
> > Note that the SGI's use slower 60ns 72pin SIMMs, while the Athlon has
> > 266DDR or whatever it is called.  
> > 
> > So memory latency could be higher, yet the SGI still done good.
> 
>   Slower memory, sure...but likely not a slower memory *subsystem*.
> PeeCees aren't known for their stellar memory subsystem designs.  I
> don't have hard numbers, but I'd sure be interested if anyone does.
> James L. perhaps?
> 
>       -Dave
> 
> -- 
> Dave McGuire
> St. Petersburg, FL
> _______________________________________________
> rescue maillist  -  rescue at sunhelp.org
> http://www.sunhelp.org/mailman/listinfo/rescue
>