Associate
- Joined
- 22 Aug 2009
- Posts
- 717
- Location
- Nottingham
Will i see much of a improvment if i overclock my 1066mhz ram am running Q9650 @ 4.15 at the mo am running 920 but the next step is 1150.
Changing ram from 920-1150
- Thread starter DarkWhisper
- Start date
- Joined
- 22 Aug 2009
- Posts
- 717
- Location
- Nottingham
What a answer
Well it's not like that much hassle and I've always found it better to test things out for yourself otherwise your not really participating are you! 
Can you get your DDR2 up to 575MHz, sounds like nice stuff, will it hit 600MHz? (DDR2-1200)
Can you get your DDR2 up to 575MHz, sounds like nice stuff, will it hit 600MHz? (DDR2-1200)
- Joined
- 22 Aug 2009
- Posts
- 717
- Location
- Nottingham
Bench testing means nothing in game play
If you read your first post again you will see there is no mention at all about gameplay or gaming
If you had typed a little more detail we may have been able to give you a better answer?
I have no idea how you use the computer, whether you enjoy benchmarking, gaming, crunching, encoding etc etc, so again some more info would be useful otherwise we have to become mindreaders!
Probably little to no differince, as you have a Q9xxx the bandwidth limitation will be the CPU's frontside bus. It doesnt matter if you have 100000mhz ram, if the front side bus is only running at 2000.
What is your FSB running at, the FSB is a 64bit wide quad pumped bus. To match the timings on a dual channel 128bit wide double pumped memory bus, the memory can be clocked as low as the same base frequency as the CPU's FSB without any significant performance loss.
IE if your FSB is 1600 (400mhz) then your memory can be as low as 400mhz (800DDR2) and the FSB is already the bottleneck. Beyond that timings is more likely to give improvement than higher speed ram.
With the new i7 processors on the other hand, the memory controller is inside the cpu, so you can clock the ram as high as the controller will allow, and the processor will recieve the benifits.
I am guessing your Q9650 is running at 460x9, so 920 on the memory (which is actually 460mhz) is a perfect bandwidth match, 25% bandwidth increase on the ram will simply be bottlenecked by the frontside bus, so unless you can go to 520x8 to achieve a similar overclock there would be little to gain. 520x8 would allow you to run the ram at 1040 while actually having sufficient bandwidth on the FSB to make use of it.
What is your FSB running at, the FSB is a 64bit wide quad pumped bus. To match the timings on a dual channel 128bit wide double pumped memory bus, the memory can be clocked as low as the same base frequency as the CPU's FSB without any significant performance loss.
IE if your FSB is 1600 (400mhz) then your memory can be as low as 400mhz (800DDR2) and the FSB is already the bottleneck. Beyond that timings is more likely to give improvement than higher speed ram.
With the new i7 processors on the other hand, the memory controller is inside the cpu, so you can clock the ram as high as the controller will allow, and the processor will recieve the benifits.
I am guessing your Q9650 is running at 460x9, so 920 on the memory (which is actually 460mhz) is a perfect bandwidth match, 25% bandwidth increase on the ram will simply be bottlenecked by the frontside bus, so unless you can go to 520x8 to achieve a similar overclock there would be little to gain. 520x8 would allow you to run the ram at 1040 while actually having sufficient bandwidth on the FSB to make use of it.
Last edited:
- Joined
- 22 Aug 2009
- Posts
- 717
- Location
- Nottingham
on cpu-z. rate fsb 1840
I don't agree with this Corasik, the FSB is not the bottleneck . . .
Are you doing this mathmetically or are you basing this on testing?
If the FSB is 400MHz (quad pumped) 1600MHz the memory will be the bottle neck if its running at just 400MHz (DDR2-800)
Effective FSB = 1600MHz
Effective MEM = 800MHz
Anyone that is running a 1.6GHz system bus with 800MHz memory speed will not get as much performance as another person running the same 1.6GHz system bus with 1200MHz Ram, I'm not sure why you are encouraging people to not speedup/overclock their ram?
I don't agree with this Corasik, the FSB is not the bottleneck . . .
Are you doing this mathmetically or are you basing this on testing?
If the FSB is 400MHz (quad pumped) 1600MHz the memory will be the bottle neck if its running at just 400MHz (DDR2-800)
Effective FSB = 1600MHz
Effective MEM = 800MHz
Anyone that is running a 1.6GHz system bus with 800MHz memory speed will not get as much performance as another person running the same 1.6GHz system bus with 1200MHz Ram, I'm not sure why you are encouraging people to not speedup/overclock their ram?
Mathamatically AND tested.
FSB is 64bits @ 1600mhz, MEM = 128bits (dual channel remember) @ 800mhz.
64 x 1600 = 102400
128 x 800 = 102400
The FSB is a bottleneck, and once you go past this the benifits just fall away. If your 800mhz ram can manage timings of 7-7-7-12, and 1600mhz ram can manage timings of 12-12-12-20 then clearly the latency on the 1600mhz will be better (waits more clock ticks, but the ticks are flying by faster).
Thing is Core 2 processors are very forgiving when it comes to memory timings, so really its just the bandwidth available on the FSB that matters. I have found that pushing the memory clocks beyond the total bandwidth on a Core 2 processor (both duo and quad) is an exercise in futility and a waste of money if it means buying faster ram. The whole point of faster ram is pushing the cpu harder and harder. As I also mentioned if the original poster can reconfigure his motherboard/cpu from 9x420 to 8x540 then he can push his ram up to 1080 and actually make use of the additional memory bandwidth.
Last edited:
Yes, 1840FSB is 420mhz quad pumped. Its the difference between clock speed and number of data transfers available
.The FSB is 64 bit though, and the memory is 128bit, so the the bandwidth is identical on both the CPU and Memory bus.
Wayne.. But just for the record... If someone had a motherboard with a broken memory slot, you can get away with running single channel configuration instead, but then you really would need to get 800mhz (ddr1600) to keep up with a 400mhz (1600) fsb.
Also as I believe I mentioned already, with the i7 the game is different, as the FSB limitation no longer applys
Also as I believe I mentioned already, with the i7 the game is different, as the FSB limitation no longer applys
You are doing the maths but not the testing I think, I'm certain the system speeds up as the memory frequency increases, what you appear to be saying (with calculator in hand) is that the system has reached maximum speed with the memory running [1:1; sync with the FSB when actually all my first hand experience says otherwise . . .
As much as engineers cannot make the bumble bee fly on paper it flys anyway, so does a Core 2 system with memeory running faster than the FSB, go figure!
Corasik, I'm glad you have knowledge of the maths, thats cool because I'm not great with that stuff, I'm just more hands on practical type of guy. I don't know what kind of testing you did to backup your figures but my own experience has been different to yours . . . .
I won't be happy until I've got a stable 333MHz-FSB (1333MHz System Bus) running on the Core 2 Duo platform with some DDR2 running at 666MHz CL5 (DDR2-1333), according to your calculator thats a wasted effort but according to my own experience I would expect to see gains!
Theory vs Practical, doesn't always match up!
Can you expand on that statement please, I'm sure there is something your trying to tell me but that I am not understanding!
DDR2-800 (CAS7) 17.5ns 6,400 MB/s
DDR2-1600 (CAS12) 15.0ns 12,800 MB/s
Can you expand on that statement please, I'm sure there is something your trying to tell me but that I am not understanding!
DDR2-800 (CAS7) 17.5ns 6,400 MB/s
DDR2-1600 (CAS12) 15.0ns 12,800 MB/s
DDR2-800 would have to be at least CAS6 to have a chance to keep up with DDR2-1600 with CAS12 because of the latency difference. (the higher the clock speed the less time required to wait for those wait states to pass).. I would guess that the speed of the ram in ns is the of the biggest factors when it comes to how tight the timings can be.
The MB/S is completely beside the point, as if for example you have a 333FSB (not overclocked) then you only have 5,300MB/s bandwidth between the northbridge and the CPU anyway.
My own system seems to run best with DDR2/667 (CAS4), would like to try some CAS3, but couldnt be bothered in the end to pay the premium. I have the FSB overclocked from 266 to 333, and the ram at 333. It works just fine, and I have tried ram @ 666, but it made no difference except in a few synthetic benchmarks.
Hello again Corasik and Merry Winter Solstice to you!
I'd like to try and get to the bottom of this thread and if possible come out of it where both of us are not going to offer conflicting advice on OcUK, I respect you as an old forum user and am aware you understand the mathmetics better than I, but this subject has been of particular interest to me for over a year and I have piously devoted a large chunk of my Geek Time-Allowance farting about trying to get my head around everything Intel® AGTL+ . . . therefore I am most perplexed why your calculator is at odds with my hands-on testing?
I think the subject matter is ridiculously complex when you drill down into it and I'm sure there is no one single person that understands all the in's and out's intimately! so I am hoping between the two of us we can get this straightened out as IMHO there is clearly some other factors at play which perhaps are not being considered . .
tRD & NBCC
I didn't know much myself about tRD (aka Performance Level) myself until I read this article
AnandTech ASUS ROG Rampage Formula Why we were wrong about the Intel X48
The Clock Crossing Procedure and MCH Read Delay (tRD)
Found that a very interesting read and it seemed to imply that running your memory asynchronously was preferable to synchronously [1:1] on an Intel® Core™2 platform . . . . so that's what kickstarted me to have nose into it and found once I got the memory running faster and with good timings/tRD the whole PC seemed to become more responsive/snappy compared to how it had been previously when running [1:1] sync . . .
So I'm sure that once you factor tRD in to the equation the argument for fast RAM becomes a bit more solid . . . . and then there is the second consideration which you have not yet mentioned and that's NBCC!
NBCC (North Bridge Core Clock) is the true frequency that is veiled behind what we know as FSB and its workings are not common knowledge . . . . having some basic understanding of how it works leads me to believe that your figures may be Skewed . . . In your calculations you are using the figure 400 when working out the FSB bandwidth for a 400MHz-FSB (1600MHz System Bus) right?
So basically the NBCC frequency is generated from the following equation . . . . which I don't know how to express mathmatically
You take the native CPU multiplier and you divide it by the set CPU multiplier and you get the NBCC Multiplier . . . . . You then multiply the NBCC multi with the set FSB and you get your NBCC MHz figure
Example E8400 @ 3.6GHz (9x400)
CPU Native Multi = 9
CPU Set Multi = 9
Set FSB = 400
9/9= 1
1 x 400 = 400MHz NBCC
Using a screenshot of CPU-z and Sisoft-Sandra to illustrate the above:
The CPU-z screenshot says we have a 400MHz-FSB and so does SiSoft Sandra which also states the System Bandwidth is 12.5GB/s
but when we drop the CPU multiplier down lets look at the numbers again . .
Example E8400 @ 2.4GHz (6x400)
CPU Native Multi = 9
CPU Set Multi = 6
Set FSB = 400
9/6= 1.5
1.5 x 400 = 600MHz NBCC
The CPU-z screenshot still is telling us we are running a 400MHz-FSB but behind the scenes the NBCC (the real FSB) is now running 50% faster! with the reported System Bandwidth now a much more healthier 18.75GB/s
Even though one is unlikey to use an E8400 at 2.4GHz as in the config above there are plenty of usable configs that will blast your northbridge into Oblivion!, the higher the chips native multi the easier it is to overclock the NBCC and to speed up the Data Transfer of the system . . .
The northbridge is totally just a few years old microprocessor and the NBCC frequency acts like one, the Intel® P45 Express reminds me of an older Intel® Pentium® III, running on average at 333MHz-400MHz and overclocks from 600MHz all the way up to 900MHz and beyond with big volts and good cooling!
The Last Of The Northbridges!
So the NBCC is the Frequency that shovels data backwards and forwards between the Nortbridge and the CPU and it's really not that hard to get it running @ 600MHz+ . . . so if we adjust your earlier calculation based on this new information look what happens . . . .
Happy to discuss, just here to learn . . . .
I'd like to try and get to the bottom of this thread and if possible come out of it where both of us are not going to offer conflicting advice on OcUK, I respect you as an old forum user and am aware you understand the mathmetics better than I, but this subject has been of particular interest to me for over a year and I have piously devoted a large chunk of my Geek Time-Allowance farting about trying to get my head around everything Intel® AGTL+ . . . therefore I am most perplexed why your calculator is at odds with my hands-on testing?
I think the subject matter is ridiculously complex when you drill down into it and I'm sure there is no one single person that understands all the in's and out's intimately! so I am hoping between the two of us we can get this straightened out as IMHO there is clearly some other factors at play which perhaps are not being considered . .
tRD & NBCC
I didn't know much myself about tRD (aka Performance Level) myself until I read this article
AnandTech ASUS ROG Rampage Formula Why we were wrong about the Intel X48
The Clock Crossing Procedure and MCH Read Delay (tRD)
Found that a very interesting read and it seemed to imply that running your memory asynchronously was preferable to synchronously [1:1] on an Intel® Core™2 platform . . . . so that's what kickstarted me to have nose into it and found once I got the memory running faster and with good timings/tRD the whole PC seemed to become more responsive/snappy compared to how it had been previously when running [1:1] sync . . .
So I'm sure that once you factor tRD in to the equation the argument for fast RAM becomes a bit more solid . . . . and then there is the second consideration which you have not yet mentioned and that's NBCC!
NBCC (North Bridge Core Clock) is the true frequency that is veiled behind what we know as FSB and its workings are not common knowledge . . . . having some basic understanding of how it works leads me to believe that your figures may be Skewed . . . In your calculations you are using the figure 400 when working out the FSB bandwidth for a 400MHz-FSB (1600MHz System Bus) right?
So basically the NBCC frequency is generated from the following equation . . . . which I don't know how to express mathmatically
You take the native CPU multiplier and you divide it by the set CPU multiplier and you get the NBCC Multiplier . . . . . You then multiply the NBCC multi with the set FSB and you get your NBCC MHz figure
Example E8400 @ 3.6GHz (9x400)
CPU Native Multi = 9
CPU Set Multi = 9
Set FSB = 400
9/9= 1
1 x 400 = 400MHz NBCC
Using a screenshot of CPU-z and Sisoft-Sandra to illustrate the above:
The CPU-z screenshot says we have a 400MHz-FSB and so does SiSoft Sandra which also states the System Bandwidth is 12.5GB/s
but when we drop the CPU multiplier down lets look at the numbers again . .
Example E8400 @ 2.4GHz (6x400)
CPU Native Multi = 9
CPU Set Multi = 6
Set FSB = 400
9/6= 1.5
1.5 x 400 = 600MHz NBCC
The CPU-z screenshot still is telling us we are running a 400MHz-FSB but behind the scenes the NBCC (the real FSB) is now running 50% faster! with the reported System Bandwidth now a much more healthier 18.75GB/s
Even though one is unlikey to use an E8400 at 2.4GHz as in the config above there are plenty of usable configs that will blast your northbridge into Oblivion!, the higher the chips native multi the easier it is to overclock the NBCC and to speed up the Data Transfer of the system . . .
The northbridge is totally just a few years old microprocessor and the NBCC frequency acts like one, the Intel® P45 Express reminds me of an older Intel® Pentium® III, running on average at 333MHz-400MHz and overclocks from 600MHz all the way up to 900MHz and beyond with big volts and good cooling!
The Last Of The Northbridges!
So the NBCC is the Frequency that shovels data backwards and forwards between the Nortbridge and the CPU and it's really not that hard to get it running @ 600MHz+ . . . so if we adjust your earlier calculation based on this new information look what happens . . . .
Happy to discuss, just here to learn . . . .
Wayne mate, that is simply one of the best and most informative posts Ive ever read on these forums.
Thanks for taking the time to pass on info like that buddy. Its helped clear up what was a huge grey area for me and I'll be tweaking my set-up based on your info.
/me hands Big.Wayne some Maryland Cookies
Thanks for taking the time to pass on info like that buddy. Its helped clear up what was a huge grey area for me and I'll be tweaking my set-up based on your info.
/me hands Big.Wayne some Maryland Cookies