Running RAM @ faster then 1:1 seems like a waste

[ghost recon88]

So I know the CPU is connected to the chipset and RAM via the FSB, therefore is there any technical and real-world performance increase by running the RAM at faster then 1:1? I mean the data transfer is bottlenecked by the FSB of the CPU, so therefore if your FSB is at 500MHz and the RAM @ 600MHz, its still only going to be transferring data @ 500MHz. I've personally benched and seen no difference, yet people claim that running faster then 1:1 does yield better results. Personally I'm starting to believe its just an e-peen thing that they can run their RAM super fast. Anyone else have facts or can chime into this?

 

[imported_wired247]

You've pretty much got it.

Certain algorithms (compression for example?) IIRC do ram->ram calculations, so in those instances you may gain a little.

 

[ghost recon88]

Thanks, now to show this thread to my arrogant friend

 

[JustaGeek]

The memory bandwidth is higher with higher frequency, but ex. my computer feels generally smoother with lower frequency and 1:1.

I used to run my RAM at rated 800MHz, now at 667MHz.

 

[lopri]

Higher bandwidth or lower latency will both contribute to the performance, albeit small. It's because FSB subsystem isn't perfect just like everything else. If the FSB 'misses' data in a cycle, it will have to look after the next available data. That's when bandwidth/latency comes into play.

 

[flexy]

i do think its faster running mem higher. But you have a point. I might set mine down to FSB-Speed and turn down latencies and check how this would look like using everest benchmark.

 

[JustaGeek]

On my computer:

800MHz - Everest Read ~8900
667MHz - Everest Read ~8400

 

[sutahz]

Originally posted by: JustaGeek
On my computer:

800MHz - Everest Read ~8900
667MHz - Everest Read ~8400

Purely synthetic, little real world relevence. I doubt its a good analogy but I just thought it up so I'll share it. Its like having a VW motor that can rev at 12,000RPM (just the motor sitting on a mount at VW labs) and having a Ferrari motor that can rev at 10,600RPM (again just the motor on a mount in Ferrari's lab). Is the VW motor better then the Ferrari motor? If so, you should buy a VW and race it, you're for sure to win right?

Higher is not a guarantee of being better. Lets say I had a voodoo3 w/ a 850MHz core speed. Will it beat your 8800GTX w/ a 750MHz core speed? I may have a Voodoo3 to sell you if so.

 

[flexy]

well i did some benchies...fricking X38 doesnt like some straps at all....my FSB is at 442 and ram 5:6 at 530. SO i set my board to 1:1 and set memory to FSB.
Everest benchmarks are a little better with the faster memory, but whats important is that the faster memory also translates into better latency.
With 530(1060) Mhz i have 57.4ns latency and 442 (884) Mhz i had 64.something. It would be interesting if i would've been able to runs sticks cas 4-4-4- at 442 to compare to 530 CAS 5-6-6-..but this didnt work. SO: Yes..faster mem translates into a little performance and better latency, but nothing too extreme.

 

[JustaGeek]

Originally posted by: sutahz

Originally posted by: JustaGeek
On my computer:

800MHz - Everest Read ~8900
667MHz - Everest Read ~8400

Purely synthetic, little real world relevence. I doubt its a good analogy but I just thought it up so I'll share it. Its like having a VW motor that can rev at 12,000RPM (just the motor sitting on a mount at VW labs) and having a Ferrari motor that can rev at 10,600RPM (again just the motor on a mount in Ferrari's lab). Is the VW motor better then the Ferrari motor? If so, you should buy a VW and race it, you're for sure to win right?

Higher is not a guarantee of being better. Lets say I had a voodoo3 w/ a 850MHz core speed. Will it beat your 8800GTX w/ a 750MHz core speed? I may have a Voodoo3 to sell you if so.

Good try, but RPM is not everything. Engine size, horspower, ignition, fuel intake etc.

Here we are talking about an almost identical product, where you can ONLY control timings and frequency, and play with voltage.

I have concluded that the difference is minimal, hence I am running it at 667MHz.

But the "speed freaks" trying to beat a SuperPi record will not agree - the times are decisevely better with higher RAM frequencies.

But the real world relevance is, as you've said, almost nil.

 

[n7]

I don't know how many times i've seen people asking why run faster than 1:1...

As i've mentioned in reply many times, while the difference is usually very small, performance for most applications is slightly better with faster memory.

Obviously, there are a number of factors, & for some things 1:1 can be just as good or better, but as a general rule bandwidth > timings.

It doesn't necessarily mean you should go spend twice as much for faster RAM...i wouldn't.

 

[Throckmorton]

Originally posted by: flexy
well i did some benchies...fricking X38 doesnt like some straps at all....my FSB is at 442 and ram 5:6 at 530. SO i set my board to 1:1 and set memory to FSB.
Everest benchmarks are a little better with the faster memory, but whats important is that the faster memory also translates into better latency.
With 530(1060) Mhz i have 57.4ns latency and 442 (884) Mhz i had 64.something. It would be interesting if i would've been able to runs sticks cas 4-4-4- at 442 to compare to 530 CAS 5-6-6-..but this didnt work. SO: Yes..faster mem translates into a little performance and better latency, but nothing too extreme.

As I understand, it actually increases real latency because the clocks aren't synced. Data has to wait for the next clock of the RAM to go through. It's not like the FSB does the 667mhz, and then stops while the RAM does its extra 133mhz. They are going at the same time, and are out of sync..