On Wed, 28 Feb 2001, Thomas Michael Wanka wrote:

> if with "these" machines you meant Intel cpus:
> 
> I had PentiumPros with 256K (or 512K, what ever the cheaper ones 
> had) and once tested the system with 1MB chips: I found no 
> difference, probably benchmarks could show some. I have a MAD 
> K6-III in a board with 2MB external cache. The same CPU in a 
> otherwise identical board with 512K did not feel slower (both with 
> 1GB Ram). I tested Durons to CE Athlons with the same frequency, 
> the Durons felt marginal less speedy. Other observations with PIIIs 
> and Xeons are not good, as there were different environments, but I 
> would not call the Xeons faster. In the Intel world it is talked about 
> "cacheable ranges", that means depending on the cache, only parts 
> of the Ram can be cached. 

Thanks for the comments!

Consider the effectiveness of caches: the difference of a PPro with 256k
and a PPro with 1MB is not really noticeable unless you have applications
which somehow stress the cache... 256k is a good general break even
point.  Less than 256k is most likely terrible performance, and more is
probably not _much_ better.

Now, lets do the same on a risc chip, such as the ultrasparc.  Now, risc
code is generally a bit larger than cisc code.  Is 256k cache
enough?  Especially with an OS like Solaris (which they're probably put
some thought into how the kernel uses the cache), the cache size will make
a difference.  Seeing as most U10s come with 2MB cache now, I was
basically asking "is 256k enough?"

-
Jon
 --------------------------------------------------------------------
 - The opinions expressed are not necesarily those of my employer.
   "I wonder how many people actually read my .sig?"