MSI K8T Neo 6702 memory bandwidth sadness

  • Thread starter Thread starter Paul
  • Start date Start date
P

Paul

I've got a Athlon 64 3200+ in the above board running stock settings and my
cache & memory bandwidth and memory sub-systems performance sucks
(SisSandra). I've got 2 x 1024 MB sticks of DDR-SDRAM PC3200. My bosses 3.06
Ghz intel runs a large geo-physical simulation (lots of floating point )
that uses 500-800 MB of memory almost twice as fast. Is this normal ?

All overclocking I've tried leads to unstability :-((

Paul
 
I've got a Athlon 64 3200+ in the above board running stock settings and my
cache & memory bandwidth and memory sub-systems performance sucks
(SisSandra). I've got 2 x 1024 MB sticks of DDR-SDRAM PC3200. My bosses 3.06
Ghz intel runs a large geo-physical simulation (lots of floating point )
that uses 500-800 MB of memory almost twice as fast. Is this normal ?

All overclocking I've tried leads to unstability :-((

Paul
Click to expand...

You have a socket 754 pin Athlon 64, the 754 has a single memory channel
so it's bandwidth is only 1/2 as much as an Intel system which is using an
800 MHz FSB. The socket 939 Athlon 64s have two memory channels so the
bandwidth is the same as Intel's and the latency is much lower.
 
Thanks for that simple and easily understood explaination (not happy mind
you)

So for my next machine to run this simulation on youd recommend AMD over
INTEL

Does it need to be an FX chip BTW ?

Paul
 
Thanks for that simple and easily understood explaination (not happy mind
you)

So for my next machine to run this simulation on youd recommend AMD over
INTEL

Does it need to be an FX chip BTW ?

Paul
Click to expand...

I would definitely go for either an 939 Athlon 64 or 940 pin Opteron based
system. Both the FX and the Opteron have 1M caches vs 1/2M in the lower
priced 939 pin parts. I don't know how sensitive you application is to
cache size so it's hard to predict if the bigger cache has a large impact.
However the price of the Athlon 64FX is so high that I think a dual
Opteron system makes much more sense. A geo-physical simulation is the
sort of thing that ought to be able to run multithreaded, obviously I
don't know if the program that you use is written for multiprocessors but
if it is you will get your biggest bang from a dual Opteron. Dual Opteron
systems have a couple of large memory related advantages over a Dual Xeon.
The first is twice the system memory bandwidth because each processor has
it's own pair of memory controllers while a dual Xeon system typically
shares a single dual bus memory system. The second is more memory because
each processor has four DIMM sockets. The hypertransport bus provides a
very high bandwidth/low latency interconnect between the two processors so
they should be able to work very effeciently together.
 
Back
Top