dasa

i have run some new tests on ram timings but there is obviously some inconsistency with the tests since i only bothered to run them all once with 1600 8-8-8 performing the same as 9-9-9 in one test and 9-9-9 performing the same as 10-10-10 in another this is just a hickup one 7970 is now dead so im back to a single gpu for these tests minimum fps in altis was to inconsistent to bother including current ram prices in au 2x4g 1333 9-9-9 $93 1600 11-11-11 $89 1600 9-9-9 $103 1600 8-8-8 $109 1866 9-10-9 $105 2133 11-11-11 $102 2133 9-11-10 $115 2400 10-12-12 $125 2x8g 1333 9-9-9 $152 1600 11-11-11 $175 1600 9-9-9 $203 1866 10-11-10 $199 1866 9-10-9 $205 2133 11-11-11 $205 2133 9-11-11 $225 2400 10-12-12 $229

im surprised your even seeing 4% gains on quad channel ram interesting to see it is there though thanks for the reply MattLDempsey 2133 9-11-10 and 2400 10-12-12 often cost about the same amount its just my old cpu cant run 2400 so i cant test it but as you go faster the gains do drop of as can be seen by the big jump Maxon got from 1333-1600 then less from there on the rate at which it drops off depends on just how starved for bandwidth the cpu is and it becomes more starved the higher you oc

i look forward to your results what is your hardware? i cant even find 2400-10-11-10-30 ram if you wouldnt mind i would like to see benchmarks with old ram then the new ram at the same speed as the old ram and then its rated speed thanks

'you left out that nobody will notice the difference between 170fps and 190fps like i said they are testing console ports which are normally totally gpu bottlenecked and the way they make it more of a cpu limit is to drop the res but with a single gpu even dropping the res isnt enough to move the bottleneck away from the gpu with those crap games while it in no way shows any real world usage benefits it is still a good ram benchmark that i believe shows close to the correct % differences you would see in a game like arma that is cpu limited even at higher res from tighter timings my ram is rated 1600 11-11-11 1.25v made by samsung but the same chips were used in ram sold by other brands at 2400 10-12-12 1.65v i run it at 2133 10-10-10 1.425v so overclocking ram can be fine but it does depend on what chips it uses

read it again they also run some tests with a 6950 and some more with 5970 & 5870 for tri-fire which sees up to 13% gain moving from 1333c9 to 2133c9\2400c10 but the reason i linked it wasnt for the performance gains but due to the lengths they went to when testing timings vs frequency it shows that 1600c8=2133c11 in performance so you need to hit a balance between the two going all for one may not yield great results or overclock it but not all ram overclocks well and you want to make sure its stable with something like memtest86+ usb before booting into windows as unstable ram can corrupt your os install ram sub timings still confuse the hell out of me after years of overclocking so for a noob it could be a little daunting

here is some tests on haswell that show the importance of getting a balance with timings\frequency http://www.anandtech.com/show/7364/memory-scaling-on-haswell/10 its just console ports unfortunately but still useful and here is something else to look out for with newer high performance ram that can make higher speed sticks slower http://hwbot.org/newsflash/2125_the_truth_about_hynix_mfr_based_memory_kits___overclockers.coms_g.skill_tridentx_8gb_ddr3_2933_memory_kit_review Roni does your mb have settings in the bios for taking the ram higher than 1600mhz? JumpingHubert it might be worth overclocking the ram try leaving at 2666 but tightening timings to 10-12-12 or as low as you can get and do you have any pics of your rig? i used a toyota camry heatercore in my rig several years ago :) the old heatercore before upgrading to the thermochil 4x120 rad

what would you like done different?

Started out as comparing cpu threads performance for interest sake but it turned into something more (Edit: Just tested 4.7GHz DDR1333 7-7-7-24-1T 38.9FPS) Test System 2600k ASUS Z77 MAXIMUS V GENE 2X4G SAMSUNG 2133 (9-10-10-24-1T used for all tests) Two 7970 (crossfire disabled) 1185\7000MHz 2TB 7200RPM with OCZ Synapse 128G Cache Seasonic 1000W Platinum win7 64 cat 13.3 beta3 Tests were run at maximum detail 720x480 aa disabled but fxxaa on Some may think running such a low res is stupid but it reduces the effect the gpu has on the tests so that the results are not just margin of error and if i had a gpu capable of running 1920x1080 close to 60fps i suspect it would give these same cpu bottelnecked numbers anyway maybe if we get some drivers that make crossfire work better i will test again at a higher res i used this benchmark to do the tests http://forums.bistudio.com/showthread.php?151794-ArmA3Mark-Benchmark-your-ArmA-3 performance gain from 1333-2133 ram when at the same timings goes from ~11% at 3.8ghz to ~14% at 4.7ghz but 1600 to 2133 is only ~8% running different timing may have been interesting aswell but that shall do for now here is the other test with core scaling the last three tests are just margin of error difference

your right it made no difference to average fps same benchmark but used fraps to record from start to finish i added another 2x4g kit for 16g total made a 8g ramdrive and put the entire arma install on it this is at 4.7GHz DDR1600 blue line is hdd+ssd cache red lines is the ramdisk

while a ramdrive may help in some situations i wouldnt be surprised if it also lowered fps a bit by hogging some of the memory bandwidth... could be fun to test :)

the faster the cpu the more starved for memory bandwidth it becomes i suspect a 3#70k would see slightly higher benefits than my cpu if at the same clock speed upping the fsb use to bring decent gains back in the day it will be interesting to see if it makes any difference on haswell when it arrives around june as early info suggests it will have a greater range of bclk adjustment than sandy or ivy

im not sure that its worth getting so worked up about is it disappointing that its the same outdated engine with a few tweaks? yes but then there is so many other games out there doing the same thing and some even have the cheek to charge $90 while arma 3 is currently only $33 from what i have seen so far the changes and improvements being made will be worth the asking price im just glad they are still making it for pc and not trying to turn it into something like operation flashpoint 2 clearly they have done some things right to have people so passionate about this game and how it runs :)

just ran a quick test with tighter timmings for comparison 4.7GHz DDR1333 7-7-7-24-1T 38.9FPS its still slower that 1600 9-10-10 but not by much

no but i tried it then and it made no difference at 4.7ghz with ht enabled it was still getting around 42fps that is what i was thinking it would take a lot of time and money to do such a massive engine overhall and there is no way it will be done so late in the arma 3 development the fact is they probably cant afford it i wouldnt be surprised if none of the current devs even had the skills required to do it as from what i can gather it is no easy task

played with these in launch options at first -cpuCount=4 -exThreads=7 but in the end the only command i used was -noSplash for the tests i linked above and just used task manager to limit the threads arma could use or is there a .ini\cfg file i can check to see what the game engine has detected?

tried those commands they didnt seem to make any improvement although they could be used to limit the number of cores used arma was spreading the load across the cores but the total load was only ever around ~90% for two cores or ~40% over 4 cores like most game engines its threading beyond two cores just isnt that good

what do you know crossfire can work a bit big boost in the first test but not so much in the rest 2600k@4.7ghz using this benchmark http://forums.bistudio.com/showthread.php?151794-ArmA3Mark-Benchmark-your-ArmA-3 3840x2160 8xaa all details maxed crossfire disabled nTest One - 24.1841\nTest Two - 25.0545\nTest Three - 18.6695\nTest Four - 19.4301\nTest Five - 20.6165 OFPMark is 2159.09 crossfire enabled nTest One - 38.0181\nTest Two - 25.1354\nTest Three - 22.5264\nTest Four - 21.6724\nTest Five - 21.0793 OFPMark is 2568.63 at a lower res and with aa off those last few tests score a fair bit higher so there not completely limited by the cpu yet crossfire isnt scaling real well in them --------------------nTest One - 55.6129\nTest Two - 40.4999\nTest Three - 35.862\nTest Four - 39.6957\nTest Five - 40.1681\OFPMark is 4236.77 so your right it is working just not real well yet which is all you can expect for a game thats still in alpha i guess

what drivers? i see no performance difference between in fps with cf enabled or disabled with 7970 cf 13.3 beta3

i would have run the ultra preset but it leaves texture detail below ultra so here is my preferred settings at the moment hmm first post cant link the pic just yet 2600k@4.7ghz 1.38v 2133 9-10-10-1t 7970@1150\7000 1.16v 2560x1440 cf\single no difference http://i3.photobucket.com/albums/y83/dasa09/arma3alphabench_zpsf423ec48.png (270 kB) been playing with settings some more and for some reason the ultra texture setting doesnt seem to make any difference over very high at least not on the landscape i was looking at maybe it affects different items but hopefully the textures just arnt there yet in the alpha the other thing is fxaa seems to sharpen things which is the opposite to how it works in most other games where it blurs things smaa which is also usually sharper than fxaa is blurry just like with fxaa disabled its a shame we dont have more control over the post process effects from the menu as some things it does look decent but other parts of it look horid