[rescue] Windows XP to be released on the Sun Blade 100

[James Lockwood]

On Sun, 11 Nov 2001, George Adkins wrote:

> Umm, the same thing as on all the Ultrasparc "i's" and "e's", it's not a
> complete ultrasparc processor.  Things have been left out, for example, the
> SMP stuff has been chopped, and the caches tend to be smaller, (when they

The IIi and IIe don't have stuff "left out" so much as "put in".  They
have significantly more glue on the CPU module (PCI bussing), but it is
less complex glue than what is (traditionally) incorporated onto the
UltraSparc-I/II motherboards for SMP concurrency.

"i" stands for "integrated".  Compare the Ultra AX and AXi if you want a
case study in motherboard simplification.

> have caches... I don't consider 128K or 256K to be worth counting as a 'L2

AFAIK no IIi has shipped with 128KB of cache.  The most common
permutations I have seen are 256KB, 512KB and 2MB.

> cache', those are the size of L1 caches.  Putting them in between the L1
> cache and the main system RAM does not make them a _real_ L2 cache, it just
> makes them a buffer.

Bottom line, for a uniprocessor application with reasonable I/O the IIi is
quite comparable to the II.  Cache is a separate issue, the "average"
cache sizes of the UltraSparc-II and IIi modules are probably reasonably
similar.

Whether the II/IIi are "fast enough" is another matter entirely.  They
are both underperformers in todays CPU market.

There can be significant benefits to keeping L1 small (32K range) and
having a smallish L2 (256KB) over having just a large L1.  The tradeoffs
here are not trivial.

-James