Attention Overclockers

[gigabyte137]

What settings do you thing will be best for overall performance of Rybka.
This is with a Q6600...

9x333 and memory at 820mhz at 4-4-4-10 In Everest I get 55ns latency
9x333 and memory at 1020mhz at 5-5-5-12 In Everest I get 57ns latency
8x410 and memory at 820mhz at 4-4-4-10 In Everest I get 67ns latency
8x410 and memory at 1020mhz at 5-5-5-12 In Everest I get 70ns latency

I do know the 2 bootom ones are 280 mhz difference but still want to know overall which would be better for Rybka.

If you can explain your reason and dont just say the Higher Freq because of bandwith because Rybka is not all about bandwith.

Please Explain

[M ANSARI]

The third one will easily be the best performing for chess.

[gigabyte137]

weirdly enough at that setting I get the worse results in PCC with same book....It seems like the 8 multiplier effects the rams latency and makes rybka run slower.... weird...

[M ANSARI]

A good test would be to run Fritzmark on both settings. I have done that and that is why I came to my conclusion.

[gigabyte137]

fritzmark measures cpu speed not engine speed. There are more factors then cpu like memory and latency between cpu and memory...

[M ANSARI]

No that is not correct ... On my QX6700 running at 3.2 Ghz my results are 10% less than a new Q6600 I just built for my brother also at 3.2 Ghz. Both ran Fritzmark at 3.2 Ghz and both were Quads, but it seems that the Q6600 performed a litte more than 10% better due to other factors ... if Fritzmark was based only on CPU speed the results would have been the same.

[Spock]

I tried various combinations of FSB and multiplier and memoery (while keeping the CPU speed constant) and found no difference in performance in Deep Fritz 10 engine mark. So I figured high FSB and memoery speed wasn't worth striving for. But I think relying on one benchmark is bad, we need 2 or 3

Were the Q6600 and QX6700 on the same motherboards ? The P35 chipset is a bit quicker I think. Only way to test that is to use the exact same machine but to swap out the CPUs and test

[M ANSARI]

Yes the Q6600 was on a P35 motherboard. That would confirm that there is more to it than just CPU speed. I think how the memory controller deals with memory and latency has a big influence. I really never tried different Fritzmarks on the same motherboard but it would be interesting. I know for a fact that for gaming and for many benchmark tests ... an equivalent CPU speed will have more performance if it achieves it with a faster FSB.

[Spock]

I know for a fact that for gaming and for many benchmark tests ... an equivalent CPU speed will have more performance if it achieves it with a faster FSB.

I know that as well - but for the Fritz enginemark it makes little or no difference... And trouble is the Fritz enginemark can vary on susbsequent runs. The 1st run and 2nd run can give different numbers anyway. It's not that reliable

[Bryan Hofmann]

What settings do you thing will be best for overall performance of Rybka.
This is with a Q6600...

9x333 and memory at 820mhz at 4-4-4-10 In Everest I get 55ns latency
9x333 and memory at 1020mhz at 5-5-5-12 In Everest I get 57ns latency
8x410 and memory at 820mhz at 4-4-4-10 In Everest I get 67ns latency
8x410 and memory at 1020mhz at 5-5-5-12 In Everest I get 70ns latency

I do know the 2 bootom ones are 280 mhz difference but still want to know overall which would be better for Rybka.

If you can explain your reason and dont just say the Higher Freq because of bandwith because Rybka is not all about bandwith.

Please Explain

Front side bus preformance increases are measured in single digits, and single digit preformance increases mean nothing in terms of ELO. In other words the difference between 820 and 1020 would be less then 4% and this would mean nothing in increase strength/ELO.


Bryan