Do we run RAMS at rated FSB or real FSB?

[invazn]

Hi guys, just a question regarding rated fsb and real fsb.

Example, if i were to have my quad core cpu at 266x9=2.4Ghz with a rated fsb of 1066(266x4 quad-pumped), do i set my rams to run at 266x2=533(assuming 1:1 ratio) or do i set it to 1066 with a 1:2 ratio? This quad pumped thing is getting me a litte confused...

 

[n7]

Generally speaking, the higher, the better.

But you won't likely see any real world difference. DDR2-533 is the minimum required, that's all.

Any DDR2-800 to DDR2-1066 is fine for OCing unless you are aiming for extremely high FSBs or RAM speeds.

 

[invazn]

So its fine to set it to 533 even though the rated fsb is 1066? Wouldn't the rams be bottle-necking the cpu then?

 

[chizow]

Originally posted by: invazn
Hi guys, just a question regarding rated fsb and real fsb.

Example, if i were to have my quad core cpu at 266x9=2.4Ghz with a rated fsb of 1066(266x4 quad-pumped), do i set my rams to run at 266x2=533(assuming 1:1 ratio) or do i set it to 1066 with a 1:2 ratio? This quad pumped thing is getting me a litte confused...

Your math is right, and as n7 stated higher memory frequency is typically better although it can lead to instability for not much of a real world benefit. Your memory's rated speed just gives you a maximum threshold for which you should expect stable operation, so if you're troubleshooting or looking for a stable OC you should stay below the rated speed.

 

[n7]

Originally posted by: invazn
So its fine to set it to 533 even though the rated fsb is 1066? Wouldn't the rams be bottle-necking the cpu then?

Not really.

RAM speed is determined by what ratio is used, so assuming Intel w/ a CPU @ 266 FSB (ignore quad-pumped rating).

We have:
1:1 (266 MHz aka DDR2-533)
5:6 (320 MHz aka DDR2-640)
4:5 (333 MHz aka DDR2-667)
And many more all the way up to 1:2 (533 MHz aka DDR2-1067).

There is more than enough bandwidth @ 1:1 for DDR2-based AMD & Intel systems, but having more (higher ratios) does slightly improve some things...mostly only noticeable in benchmarking though.

This is different with DDR3 & now i7 + DDR3, where 1:1 is no longer a realistic option, much higher is ideal.

 

[Denithor]

Originally posted by: n7
RAM speed is determined by what ratio is used, so assuming Intel w/ a CPU @ 266 FSB (ignore quad-pumped rating).

We have:
1:1 (266 MHz aka DDR2-533)
5:6 (320 MHz aka DDR2-640)
4:5 (333 MHz aka DDR2-667)
And many more all the way up to 1:2 (533 MHz aka DDR2-1067).

There is more than enough bandwidth @ 1:1 for DDR2-based AMD & Intel systems, but having more (higher ratios) does slightly improve some things...mostly only noticeable in benchmarking though.

Great summary.

This is different with DDR3 & now i7 + DDR3, where 1:1 is no longer a realistic option, much higher is ideal.

I'm actually curious about this. Is the fsb for i7 133MHz or something else? If so, does memory really run at that much of a multiplier versus the fsb? It still doesn't seem to make much difference, from the i7 memory comparison AT saw minor differences (at best) between DDR3-1066 and DDR3-1600.

 

[Lonyo]

Originally posted by: Denithor

Originally posted by: n7
RAM speed is determined by what ratio is used, so assuming Intel w/ a CPU @ 266 FSB (ignore quad-pumped rating).

We have:
1:1 (266 MHz aka DDR2-533)
5:6 (320 MHz aka DDR2-640)
4:5 (333 MHz aka DDR2-667)
And many more all the way up to 1:2 (533 MHz aka DDR2-1067).

There is more than enough bandwidth @ 1:1 for DDR2-based AMD & Intel systems, but having more (higher ratios) does slightly improve some things...mostly only noticeable in benchmarking though.

Great summary.

This is different with DDR3 & now i7 + DDR3, where 1:1 is no longer a realistic option, much higher is ideal.

I'm actually curious about this. Is the fsb for i7 133MHz or something else? If so, does memory really run at that much of a multiplier versus the fsb? It still doesn't seem to make much difference, from the i7 memory comparison AT saw minor differences (at best) between DDR3-1066 and DDR3-1600.

Well the fact that dual channel vs triple channel was a minimal difference (on AT, in real tests, not bandwidth tests) means that any speed difference would logically also be minimal.
Aren't there set multipliers for Nehalem anyway? Like Core is 20x (for 920) while uncore is bx FSB and RAM is cx FSB, so they all have their own multis?

 

[Denithor]

Originally posted by: Lonyo
Well the fact that dual channel vs triple channel was a minimal difference (on AT, in real tests, not bandwidth tests) means that any speed difference would logically also be minimal.

Exactly.

Aren't there set multipliers for Nehalem anyway? Like Core is 20x (for 920) while uncore is bx FSB and RAM is cx FSB, so they all have their own multis?

That's what I'm trying to figure out. Either way though, faster memory speed doesn't seem to be required/do anything really on this architecture (like we have seen previously on K8/K10 with IMC & C2D/C2Q with large L2 cache).

 

[invazn]

So when some people say that running rams at 1:1 is a better option, it is for stability reasons only? In terms of performance, it is still better to use a fsb:ram ratio to match the rated FSB of your cpu? In this case... 1333?

 

[Cogman]

Please, it is RAM in the singular, not RAMS (there is no S) RAM stands for Random Access Memory, so saying Random Access Memories doesn't really make a lot of sense.

As for performance, run your ram at what you can, pushing it won't gain you anything as ram speeds are generally not the bottleneck or most applications.