DDR3 1:1 ratio? Does it exist? Does it matter?

[s74r1]

I'm pretty new to DDR3, but as I understand it with DDR and DDR2 it was usually preferable to get a 1:1 ratio of bus/fsb to ram speed. for example 400 bus and 400(800) ram on a core 2 duo would be a 1:1 ratio. Or on the older Athlon64's 300 HT speed with 300 DDR1 (600mhz) was a 1:1 ratio.

Does this exist anymore with DDR3? What would be the 1:1 ratio? Or would it be way too low? Or is there something like a 2:1 ratio that is preferable for latency reasons?

Also, if anyone can explain how slower uncore speed affect latencies and memory bandwidth that would be great too

Thanks!

 

[overshocked]

Well it still exists on the older chipsets like 775 and such. On X58 though, you need to set the memory divider (this is a lot like the 1:1 ratio) you will want to set it at 6 to get the highest CPU overclock.

 

[s74r1]

so, which memory multipliers have the best latency usually? and whats the memory performance loss of lowering uncore multi on i7's?

 

[RJR]

Your not going to see a big difference one way or another, but, here you go:

http://www.anandtech.com/memory/showdoc.aspx?i=3589

 

[s74r1]

Your not going to see a big difference one way or another, but, here you go:

http://www.anandtech.com/memory/showdoc.aspx?i=3589

Thanks, while that didn't quite answer my question, it was very informative. I can't believe how well that CAS6 DDR3-1600 scored compared to the 2000mhz in real world.


But I'm still curious which BCLK to mem ratios perform the best.

 

[RJR]

But I'm still curious which BCLK to mem ratios perform the best.

OK, lets say you have an i7 and have overclocked it to 3.8 Ghz with a bclk of 190 and 1600 ram. Your memory straps will allow 1520 Mhz or 1900 Mhz speed (once again after reading many articles, and my own system, very little gain at higher frequencies) so if your ram will let you OC it to 1900 Mhz and remain stable, great. If not, setting it at 1520 for stability purposes will NOT take much, if any, of a hit in real world performance.

As for the article, it should have answered your question because it gave both latencies and bandwidth and real life bench marks.

So, if synthetic benchmarks are important to you and you want to maximize your rams performance for said benchmarks, Everest will show you both bandwidth and latencies and you can fine tune your memory for the maximum bandwidth and minimum latency you can obtain. Once again, it won't make any difference in real world performance but if you have a few weeks to kill why not.

 

[s74r1]

OK, lets say you have an i7 and have overclocked it to 3.8 Ghz with a bclk of 190 and 1600 ram. Your memory straps will allow 1520 Mhz or 1900 Mhz speed (once again after reading many articles, and my own system, very little gain at higher frequencies) so if your ram will let you OC it to 1900 Mhz and remain stable, great. If not, setting it at 1520 for stability purposes will NOT take much, if any, of a hit in real world performance.

As for the article, it should have answered your question because it gave both latencies and bandwidth and real life bench marks.

So, if synthetic benchmarks are important to you and you want to maximize your rams performance for said benchmarks, Everest will show you both bandwidth and latencies and you can fine tune your memory for the maximum bandwidth and minimum latency you can obtain. Once again, it won't make any difference in real world performance but if you have a few weeks to kill why not.

Thanks for the info, but no the article didn't quite answer my question.

While they did test different speeds, they didn't test each speed at different BCLK's and different memory multipliers (there's different ways to achieve the same speed). But this would probably deserve it's own article, as that wasn't intended to test the memory controller's efficiency at different multis and ratios.

I'm not sure how it applies to today, but in the days of DDR and DDR2 the memory controllers would operate more efficiently at certain multipliers/ratios.

I suppose it's all moot anyways, and it's more important to get a better CPU overclock than worry about optimal ratios and multipliers, but it's something I still want to consider when trying to find that "sweet spot" for my rigs.