core 2 duo and P4 3.2 E

"Fishface" wrote in message
...
Ed Medlin wrote:

When set to 2 cores/no HT memory performance went up from
14896mb/s read to 16827mb/s.....write up to 13939mb/s.

But- I mean't Super-Pi, since you weren't bandwidth limited as I was
with more than one core. Since I shaved 2.5 s by eliminating the
threading overhead, I wondered if you would see a similar gain by
eliminating two threads, as your CPU utilization was 50% to my 25%.

Super Pi performance did not change at all. CPU utilization was also
unchanged. I don't understand.....:-) Even set to 1 core, no change. This is
with HT also turned off. There must be some boost from the lack of the
traditional FSB and direct memory control on-die.


Ed

Ed Medlin wrote:

"Fishface" wrote in message
...
Ed Medlin wrote:

When set to 2 cores/no HT memory performance went up from
14896mb/s read to 16827mb/s.....write up to 13939mb/s.

But- I mean't Super-Pi, since you weren't bandwidth limited as I was
with more than one core. Since I shaved 2.5 s by eliminating the
threading overhead, I wondered if you would see a similar gain by
eliminating two threads, as your CPU utilization was 50% to my 25%.

Super Pi performance did not change at all. CPU utilization was also
unchanged. I don't understand.....:-) Even set to 1 core, no change.
This is with HT also turned off. There must be some boost from the lack
of the traditional FSB and direct memory control on-die.


Ed

How much cache does your processor have ? How much memory
does SuperPI use for the calculation ?

In a quick glance here, I thought SuperPI for 1 million
digits, uses 8MB of RAM. Consequently, for benching with
a processor with a large cache, the number of digits
has to be bumped up. Check the Task Manager performance
monitor and see how much the memory occupancy jumps up,
when SuperPI begins to run.

I can use 1 million digits here, because the cache on
mine is tiny :-)

Paul

"Paul" wrote in message
...
Ed Medlin wrote:

"Fishface" wrote in message
...
Ed Medlin wrote:

When set to 2 cores/no HT memory performance went up from
14896mb/s read to 16827mb/s.....write up to 13939mb/s.

But- I mean't Super-Pi, since you weren't bandwidth limited as I was
with more than one core. Since I shaved 2.5 s by eliminating the
threading overhead, I wondered if you would see a similar gain by
eliminating two threads, as your CPU utilization was 50% to my 25%.

Super Pi performance did not change at all. CPU utilization was also
unchanged. I don't understand.....:-) Even set to 1 core, no change. This
is with HT also turned off. There must be some boost from the lack of the
traditional FSB and direct memory control on-die.


Ed

How much cache does your processor have ? How much memory
does SuperPI use for the calculation ?

In a quick glance here, I thought SuperPI for 1 million
digits, uses 8MB of RAM. Consequently, for benching with
a processor with a large cache, the number of digits
has to be bumped up. Check the Task Manager performance
monitor and see how much the memory occupancy jumps up,
when SuperPI begins to run.

I can use 1 million digits here, because the cache on
mine is tiny :-)

Paul

The Nehalem has a completely different architechture as far as the cache
goes. L1 is 32kb per core ECC on die/full speed. L2 is 256kb per core, ECC,
on die/full speed. L3 is 8mb, on die ECC full speed. Running SuperPi at 1M
digits, memory usage goes from 1.24GB to 1.25GB....virtually nothing. Total
memory in Task Manager shows as 6134MB including Cached of 3557. SuperPi
doesn't seem to use very much of my 6Gbs of physical memory at all when
running any number of digits. CPU usage goes down to 25-30% with HT turned
off, which is a bit confusing. Turning off cores doesn't seem to change
anything. Vista 64 uses 1.24GBs to run at idle with all the bells and
whistles.

Ed