dual core x2. able to provide some benchmarks between single and dual channel mem int

[tristancarton]

searched around for dual core single channel performance numbers. as expected i couldn't find any.(who in their right mind would spend that much money on a cpu and only buy one stick or ram)

in a seperate thread on this forum we are talking about dual core s754s and why amd will not implement them. one of the members posted was that the memory interface in single channel would kill performance numbers.(i disagree) i know the general performance loss when running a single channel comapred to dual channel.(usually 4-8 percent) what i am wondering if these number escalate due to the dual cores?

anyone able to provide some numbers for me? not sure what benchmarks would illustrate this. maybe some photo or video editing?

[[timko]]

in a seperate thread on this forum we are talking about dual core s754s and why amd will not implement them.

Here's another reason why dual core S754 might not happen...

http://www.xtremesystems.org/forums/...ad.php?t=67471

[tristancarton]

yeah thanks timko. i know we have a bunch of threads on s754 dual cores and that they are not going to happen.

i am just trying to see how much of a killer a single channel memory interface would be for dual cores. with just one cpu it isn't a big deal.

[p0tempkin]

In theory the performance degradation by switching to single channel should escalate with dual vs. single cores because the dual cores are sharing the 128-bit memory bus (whereas a pair of single core opterons would each have their own full 128-bit memory bus). By further cutting the memory bus width in half (to 64-bit) by using a single channel interface, you might starve the cores at higher speeds. Here is AnandTech's explanation:

Historically, the Athlon 64 hasn't really been memory bandwidth bound, since the move to Socket-939, which gave it a full 128-bit wide memory bus, and more bandwidth than these CPUs could use.

With the move to dual core however, the effective memory bandwidth that each core gets is significantly reduced, as they both have to share the same 128-bit wide memory interface normally dedicated to a single processor.

I think that explains why in real world performance, Socket 939 CPUs see little real-world gain in performance with increased memory bandwidth, and why 754 show more real-world gain in performance with a similar percentage increase in bandwidth.

[Ugly n Grey]

I could test it with my 4200, but why? We already know it will degrade especially with two cores sharing a memory bus as p0tempkin said. I mean really, dual channel architectures were built for this kind of thing (two CPU's). Who wants something crippled?

[tristancarton]

ugly the reason for doing this would be to test this hypothesis. nothing more, nothing less. it has no real world value other than to see how memory bandwidth intensive dual core chips are.

[cadaveca]

i'll be sure to try when mine gets here. I don't think it's going to ahve that big of an impact in 32bit, becasue regardless of it being dual core, and dual channel, it is not dual memory controller. each core has to queue up if both cores are running with a decent load. In 64bit, when one request can fill the sinlgechannel controller fully, it should have a larger impact, but becasue of that pesky single mem controller issue, I cannot wait to see how it works out.
In an intensive 64bit situation, DDR2 may not even provide the neccesary bandwidth, and with quad cores...i guess me might finally be able to see just how efficient the AMD memory controller really is..

[aldamon]

Was this test ever done? How much of a performance hit does an X2 take while running in single channel?

[STEvil]

never seen it tested.

I would assume 10% depending on the app.

[fhpchris]

Was this test ever done? How much of a performance hit does an X2 take while running in single channel?

Only like 5% in single channel is lost to dual in my testing

[aldamon]

never seen it tested.

I would assume 10% depending on the app.

This place is benchmark crazy so I'm sure somebody here is curious. Can somebody pop out a stick of RAM or move one stick over to disable dual channel and post some quick CPU-intensive and multi-tasking benches?

I'd like to end an argument I'm having about whether or not Socket 754 could have run an X2 with reasonable performance.

I think AMD is full of crap when they say S754 couldn't support the X2.