Discussion in 'General Software' started by Asylum, Dec 8, 2009.
jlewis02|HD5850 x2|980/1125|30404|i7 firstname.lastname@example.orgGhz
Go back to your bios and set your FSB to 200 and your multiplier to 20. Set your memory to it's proper ratio. Then Lock your Northbridge and QPI to a 1:1 ratio. 3609/ 7218 qpi. Then run the benchmark again. You will see a fair amount of gain on your cpu score but mostly your gpu score will go up QPI clock is what feeds your Graphics cards pci-e bus and enhances throughput. GOOD LUCK! POST BACK RESULTS!
I will try that thanx.
Post a shot of cpu-z's main page and memory page. Make sure the frequencies are the same. Also, I take it you have Hyperthreading disabled? Since your Vcore is so low. I'm running that 200x20 multi @ 4009Mhz, 1:1 ratio QPI/ Memory Frequencies with Hyperthreading enabled and vdroop disabled @ 1.4V Vcore, 1.4V QPI @ 70-75C MAX on my Core i7 930 lapped with lapped Cocage true spirit and IC Diamond paste. I'm putting together my New Lian Li PC V354 build and still perfecting the cooling design. I'm going to be using Arctic Silver Ceramique. Hopefully the temps will still be acceptable!
HT is on and voltage is at 1.31v now.
Wow, those are really low voltages for 4ghz. I have an ASUS Rampage II Gene. You should run it through 2 runs of intel burn test at Maximum stress level and 32 threads. See if it's stable, i doubt that it will pass at that voltage. With my bios set to 1.4V core voltage and 1.4V QPI. Vdroop enabled results in added voltage in windows, disabled results in reduced voltage in windows. Vdroop disabled makes your temps lower while still being stable. The vcore in windows, sits at around 1.36-1.38V. I can run it through intel burn test and pass. It will heat your cpu to it's absolute maximum temperature, so make sure your running cool. Post a confirmation of if you pass and your core temps while it's running with everest or equivalent. The cool thing about intel burn test is you can know if your overclock is stable in just a couple minutes! I rely on this more than Prime95 since i've witnessed running Prime95 in blend or any other mode for 12 hours with 0 errors. Then run it through 10-15 passes of intel burn test and FAIL! that's like only 2 hours. It pushes the CPU so far into it's maximum performance output! It's crazy how insanely powerful these Core i7s are! They have tons of peformance available past 100% utilization. If you know what i mean? Like there are intensity levels at given utilization levels. I'm curious for your results!!
MetalRacer|2 X HD5970's|980/1250|47179|i7 980X @ 4.61GHz
Lock down your QPI Frequency with your Memory Controller frequency. Added memory bandwidth, and more performance. Locking down the Memory Controller with your QPI Quick path will allow for a more linear data path for your memory and qpi to talk to eachother. Your score should go up by doing this! Give it a shot and post back!!!!
QPI Data Rate @ 7218Mhz /2 = 3609Mhz
Memory Controller Data Rate @ 3609Mhz
QPI Rate Divided By Memory controller data rate = 1:1 Ratio.
I may be getting ahead of myself. I'll do some further experimentation to find the ideal ratio for best performance. 1:1 ratio is ideal for any configuration for compatibility and linearity. So far I have come to a post-conclusion of a non-linear QPI/Northbridge ie memory controller data rate may perform better by a tiny amount. This is only a little research. When i get home from work I'm going to do some hands on testing and post results. With what tests I have done personal in regards to Everest/ AIDA64s memory/ cpu cache benchmark. Linear QPI/ Northbridge ratios equal better performance across the board compared to 3207 northbridge frequency and 7218 qpi. More is better. If you can boot at 8821 qpi and 4411 northbridge frequency that will give your memory an additional 10GB/s of memory bandwidth and approximately 20+ Additional CPU Cache GB/s.
Krony|eVGA GTX580SC|910/1820/1025|33584|i7 email@example.comGhz
yogurt_21|MSI GTX480|850/1700/990|26744|i7 firstname.lastname@example.orgGhz
HammerON|EVGA GTX 470 SLI|860/1025|43272|i7 email@example.com
I would like to see what conclusion you come to. It was to my understanding that QPI and IMC freq were supposed to be different to prevent possible errors. what is the opinion from others?
Well pulling a 1:1 ratio with the qpi/ memory controller yields memory bandwidth gains but isn't 1:1 with the memory itself. It depends on what speed your memory runs at. More memory controller speed equals more memory throughput. More QPI speeds equals more CPU L(x) cache throughput. From what i've read, people who run a 3207 memory controller data rate who have ddr3-1600 yields a 1:1 ratio with the memory modules and the memory controller. That does translate into more performance. Although memory controller data rates did increase memory bandwidth equally as you go up. I know that i've ran a QPI 8821/ 4410 memory controller and had memory data rates of 28GB/s! compared to 19-20GB/s on the 3609 memory controller data rate. The QPI rate is supposed to control L(1,2,3) cache and also feeds your bus's. Including the PCIe bus. I can't boot at 8821 QPI now, I was only able to boot at that once. Pulled some crazy scores on some tests though! Exceeding that of much higher end cpu's with that 8821 QPI. At the 7218 it doesn't really stand out so much. Again QPI rate generates a huge amount of heat. The best way to test this theory is by turning your QPI to "slow mode" in the bios and test the memory controller at 3207 and 3609 and see where the results are with the GPU score and CPU score. I'll run some tests on Vantage once i get some time and a day off.
Edited post new score.
3DMark Vantage Compilation
Wow Hammer thats insane....That beat my 480's score.
Did the new drivers give a performance boost?
Not sure becuase I never tried at the same settings with the old drivers. May have though...
Strange as I couldn't get the core higher on the gpu's. Used to be able to get them up to 880, but I was able to get the memory higher with the new drivers
Gas2100|MSI GTX480|701/924/1401|P22797|i7 firstname.lastname@example.orgGhz
better then my Q8300 + 460 combo
yogurt_21|2x MSI GTX480|823/1646/992|40574|i7 email@example.comGhz
aha it was faster with newer drivers
sorry for the pics, I ran the trial version
yogurt_21|2x MSI GTX480|882/1764/1052|43977|i7 firstname.lastname@example.orgGhz
Edit: updated, gpu's seem to be craving more volts, will mess more around with that later along with finding the cpu max. but for now it's a decent score with only 2 nvidia scores higher. 2nd bench tonight I'm a handful away from the next thousand marker.
Nice run yogurt
I don't think I can beat that score
Can't wait for the GTX 590 (and a nice waterblock)
a 590 would definetly pair with your cpu well, seems like your rig has more to show us but the 470's are tapped. you notice how vantage seems to crash at lower clocks than 3d11? wonder what that's about?
improved my score a little.
Mydog|MSI HD5970|1000/1300|+2x MSI HD5870|1020/1330|P49223|i7-980x @4716
Nice score Mydog
Now break the 50K barrier already
second this. you're soooooo close.
Separate names with a comma.