[Trac #125] Geode GX is a poor CPU choice

[Ian Stirling]

Zarro Boogs per Child wrote:
> #125: Geode GX is a poor CPU choice

>  A simulation of Rhythmbox shows a 1.2% overall cache miss rate.  Multiply
>  this by 25 cycles and that's a 33.2% slowdown; multitasking will give you
>  another 0.7% slowdown give or take a little bit but this is negligible.
> 
>  The folks in ##electronics on Freenode seemed to be under the impression
>  that 25 cycle cache miss penalties indicate a "crippled" CPU and that this
>  thing is "dumpster-bait."  I'm no EE so I'll have to go with what they
>  came up with.

I wonder if they realise how many cycles cache miss penalties their PC 
takes...
Hint: It's nowhere near 25, and not in a good way.

Cache misses hurt in other ways.

The 'operating current read' line from the DRAM spec indicates that to 
read in a cacheline miss takes around 1W, and to do it continually takes 
about 2W. - obviously these numbers reduce proportionally if you're not 
taking cache misses at the full bandwidth of memory, and you will _very_ 
rarely find a case where you'd get no cache misses with a bigger L2 
cache, and 100% without.
(some of the bigger encryption, or more annoying media compression 
algorithms are probably closest for this).

On the flip-side.
The GX processor - and indeed most recent processors pretty much suck, 
in terms of minimum power use. This basically means that you will 
probably have to suspend-RAM very often, to get decent power savings.
This pretty much inherently means that you're losing the cache every time.
Worst case, reading back the 64K cache - 8192 lines - takes 100us.
At 100Hz context switch - this is about 1% of the time.
At maximum burst rate of the memory, it's more like 20us.

For the second generation hardware, I'd not be surprised to see more 
stuff moved away from the core, and with the machine in its basic 
configuration - nothing plugged in - state, able to do pretty much 
everything that requires no user interaction or screen repainting 
without powering up the main CPU.
For example, a MP3 audio chip, fed from the DCON, able to buffer a few 
seconds of audio.
For 3rd gen, maybe even throw away the companion chip, which has lots of 
stuff we're never going to use, and integrate DCON+USB+... in there, 
with hardware helpers designed to keep the processor off for all but 
aperiodic interrupts.
Maybe even wackyness like the DCON having enough smarts to animate 
cursors and small pixmaps if given a list of regions, and frame times.