Printable View
Just looking at the two model of 64bit Athlons, and was wondering if someone might explain something to me. What are the limitations incurred with the two different Integrated Memory Controllers on the two different chips (Athlon64 3200 & FX-51)...The 3200 is singlechannel 64bit while the FX-51 is dualchannel 128bit...
Any explanations for me?
Many Thanks
Dual memory controllers effectively means that each module can be written/read to/from at the same time, as a 64-bit memory controller is dedicated to each. This means that on paper you get twice the memory bandwidth, and in reality it does make a difference. On an Abit NF7-M nforce2 board using the onboard geforce4 MX graphics, the score of 4000 marks in 3dmark2001SE was achieved with the RAM modules in the wrong slots.
Once they were shifted over and DualDDR was enabled, the score jumped to 6000 marks. Of course, this dramatic leap won't be evident in all systems.
Anyone know any more about this? I remember reading a respected hardware site that affirmed that in all the tests that they had done the difference between single and dual channel performance was in the order of 1-2%, with a whole page devoted to dualchannel in like a 12 page write-up .
Were the earliest A64 only single channel? As far as I'm sure, Everything 939 is dualchannel. how much performance is gained with dualchannel now that it's integrated? is the difference over 10% In real-world benchmarks?
Anyone got some wise words on how the on board memory controller responds to overclocking? All I've heard is that the the number one limit on A64 platform is the speed of the RAM.
It certainly won't; the onboard graphics on an NF7-M, whilst they are based on a GF4MX core use the systems RAM for memory. Doubling the bandwidth, therefore, will make a HUGE difference in circumstances like that, but significantly less isn systems which are not using integrated graphics dependent upon the systems's RAM.Quote:
Originally Posted by DaftpuNk
I've seen various estimates for the effect of dual-channel memory outside of those cases, and they do vary between 1-5% on overall system performance. The difference will be especially noticeable, however, in any memory intensive tasks.
Oh, and the above is correct; S754 A64s are single channel only, 939s are dual.
for A64 systems there isnt a *huge* advantage
yep, a64's hardly benefit from dual channel.
As mentioned it ranges from 2% to around 5% better.
which isnt alot, at all.
Im happy with my single channel 3400+ anyway.
Main issue is dual channel only boosts bandwidth not latency, and latency is what really kills performance with RAM.
Bandwidth is more important in a few cases, but normal use and games will see much more benifit from low total latency than from high bandwidth.
Yeah my CAS0.5 EDO pwns my DDR3200 dual channel ;)Quote:
Originally Posted by Butcher
Real latency and memory timings are two very different things, though timings do influence latencies.
The sytem in my sig has 39ns total latency to main memory. I take this exact ram and run it at the exact same speed/timings in an Athlon XP system and I get well over double that.
You also need enough badwidth. 60MB/s with 1ns latency is going tobe really really slow for memory. Just as 5000MB/s is going to be slow (for memory) if the latency is in the miliseconds.