Different bandwidths?

[Firestrider]

Me and my roommate have the same RAM: Crucial Ballistix DDR2-800, both with timings 5-5-5-15-20-2T and in dual-channel. I'm running mine at 472.5Mhz and he is running his at 500Mhz. Yet when we run a SiSoftSandra benchmark to find the bandwidth why do I get 9.4GB/s and he gets 8.33GB/s? I'm guessing because of the length of the interconnect between the RAM, northbridge, and CPU, but why such a degradation?

[thideras]

Same RAM, yes. What motherboard, processor, etc. As strange as this sounds, memory performance can be reduced by other components. Unless you have the exact same hardware running the same speed, you can't compare them.

[meionm]

CPU, motherboard will affect it. Some mobos have memory enhencment feature that will produce large difference when being on and none to little in real life.

[Firestrider]

He has a E8400@4.0Ghz and GA-P35-DS3L. My CPU and motherboard are listed in my signature.

[thideras]

AMD's have been known to have faster RAM bandwidth because there is no "northbridge" that adds latency (delay) to access the memory.

[Firestrider]

Thanks for clearing that up. I have a couple more questions:

When the memory on the video card gets full does it offload the extra data onto the system memory? If so how can I benchmark this interconnect?

I'm aware that a faster core clock speed on a video card would mean better pixel/texture/vertex fillrate usually increasing FPS in a game, and a faster clock speed on a processor would mean more instructions per second and floating-point operations per second for calculations in a program, but what does a faster bandwidth on system and video memory increase in real-world performance?

[thideras]

When the memory on the video card gets full does it offload the extra data onto the system memory? If so how can I benchmark this interconnect?Yes. There is no reason to benchmark that...

I'm aware that a faster core clock speed on a video card would mean better pixel/texture/vertex fillrate usually increasing FPS in a game, and a faster clock speed on a processor would mean more instructions per second and floating-point operations per second for calculations in a program, but what does a faster bandwidth on system and video memory increase in real-world performance?More bandwidth means it can just transfer that much more data. If it was a faster clock speed (bus speed), the latency goes down, meaning that the rest of the computer doesn't have to wait as long for the video card to be ready to accept data.

[Firestrider]

So how can I find bottlenecks in my system?

[thideras]

So how can I find bottlenecks in my system?Well, that isn't a bottleneck unless you are using all your video card's RAM. At that point you need to turn down AA/details/resolution. If you mean CPU/GPU bottleneck, decrease either one by a little bit and if FPS doesn't go down, the other is the bottleneck ;)

[Firestrider]

Can the system memory ever bottleneck something?

[thideras]

Can the system memory ever bottleneck something?Honestly, I can't think of anything unless you store a program or OS completely in memory...the only things that can really saturate it is memory tests that are supposed to saturate it :beer:

[Firestrider]

So is there any point in overclocking RAM other than to keep up with the bus speed from overclocking the CPU? No matter how much bandwidth RAM it won't be a bottleneck in the system?

[thideras]

So is there any point in overclocking RAM other than to keep up with the bus speed from overclocking the CPU? No matter how much bandwidth RAM it won't be a bottleneck in the system?Well, things like SuperPi use memory quite a bit ;)