I've been gone all day - but I must say that I really enjoyed everything that's been posted here today, this is excellent reading. Great work Dazz! I think you're on to something. As you have already showed, memory has almost no effect on min, max, and avg frame rates. However, the difference looks to be rather significant in your other test. It appears that singling out the stutters reveals the difference between memory speeds. Now I wish I had some 2133 CL8 to play with!Dazz, thanks for sharing the raw data - just saw you had already posted the link! Must fulfill my need to create graphs!!!For anybody looking for some high speed / low latency memory - here it is! Even "designed for P67"!

Thanks Corey for a thorough reply.My question remains though: is it ok to use 1600MHz and 1.65v while overclocking an i5-2500k from stock 3.30 to -say- 4,8GHz? Does CPU overclocking affects (raises) RAM's voltage? IF this is a YES, then I guess OC is out of the question since 1.65v is already marginal for the i5-2500k/P67 combo. Right?Given so, next available option is an 1.5v RAM. How does this voltage affects (restrains) the maximum OC result frequency I can get to my CPU?

CPU and RAM are not in any way tied together on the LGA1155 platform as they were on previous platforms. Unlike 1366, you can overclock the CPU completely independent of your RAM and you can overclock your RAM completely independent of your CPU. With 1155, the BCLK is set to 100MHz - and you really shouldn't change it. The CPU speed is simply the CPU multiplier x the BCLK and the memory speed is simply the memory multiplier x the BCLK. One has no effect on the other. Likewise, voltage to the CPU has no effect on voltage to the RAM. My memory, by the way, is rated at 1.65v. But I run it at 1.56v because I can. Just don't populate all four memory slots or you may have to apply more voltage. But again, 1.65v is perfectly fine for Sandy Bridge.

Thanks Corey. Truth is there seems to be something there showing faster RAM helps more than what plain max/min/avg figures do. I posted the dropbox link already hehe. Here it is once again ^_^http://dl.dropbox.co...76/stutters.rarTests at 2133 7-10-8-27 Frames, Time (ms), Min, Max, Avg 10878, 300000, 24, 50, 36.260<30 1354<20 28<10 4Give me a minute and I'll try tightening those timings a bit moreNo luck. I tried 7-10-10-30 instead:Frames, Time (ms), Min, Max, Avg 10785, 300000, 24, 50, 35.950 <30 1364<20 29<10 4

I built another graphic. It shows the instant variation in FPS from frame to frame:As you can see, it's very simetric: every time there's a stutter, it bounces back and conversely when there's a sudden instant FPS increase (two frames very close one from the other) a stutter will follow. It suggests I should try limiting FPS and see if it changes anythingEDIT: It took me a couple of minutes to realise how stupid this ^^ was. Disregard this post please

Edited March 8, 201115 yr by dazz

Dazz,Thanks for allowing me to geek out with all the data you created. I tried my hardest to get it to say something - so here it is. For each frame, I rounded the fps to the nearest multiple of 5, then plotted the percentage of total frames at 5 fps, at 10 fps, at 15, and so on. Here's the link to the excel file.EDIT: link updated to 97-2003 .xls format

Ok, so for example 90% of the frames are in these ranges:1600 CL7: 26.4 - 46.12133 CL7: 27.9 - 47.6Nice, thanks to you for sharing your graphs. That looks very much like a Gauss bellCan I ask you a favour please? My ancient 2002 version of office can't open xslx spreadsheets :P could we use xls from now on please?

I had some errors in the results I posted previously. These are accurate (I think, LOL)The updated sheetsAs you can see I added another row: Number of frames below the minimum for each test. There's not so much difference there when we compare results that take in consideration the average frame rate or the minimum for each config.I really couldn't tell one test from another visually, and if you take a look at the graphs, the profile is pretty much the same for all of them (interestingly the stutters appear to happen almost always at the same times in all runs)A 2% of difference in the minimums between 2133 CL 7 and 1600 Cl7 is not much, but considering that we're talking about 10K+ total frames per test, it's not hard to believe that a 2% lower frame rate can translate into an increase from 1354 to 1647 frames below 30FPS.I did a simple correction to the 1600 CL7 data set and increased each frame by a 2%. Here's the result:max 49.935 min 24.305 avg 36.547 <30 1418 <23 102 <20 32 <10 4 <min 189That's virtually the same as 2133 CL7 To put it in another way we could ask ourselves, what would be the frame rate below which we have exactly 1354 ocurrences in the 1600 CL7 test?29.27 vs 30 for the 2133 CL7 run, and that's again a 2.5%. So it looks like the only reason why the figures are better in the 2133 CL7 test is the overall FPS increase. We could repeat the same procedure for the rest of the rows with a similar outcomeSo I really don't think there's a correlation between RAM speed and less stutters anymore, at least based on these tests. We should do the video comparison but I'm willing to bet no one will notice the difference, even with the 4% overall average FPS gain EDIT: Here's a section of the graph I was talking about:Notice that the same dips happen almost at the same times and with the same amplitude. The 2133 CL7 graph below shows how because ot the slightly higher frame rate there's an offset in the position of the stutters (horizontal axis is not time, but # of the frame) but the pattern is pretty much the same. The complete graphs are available in the updated spreadsheets for you guys to check them out if you want to and you will see the same thing goes for all of them from begining to end

Nice stutter charts (worthy of a Kings Speech!), but charts are difficult to compare, so what you really want is to reduce the charts to a single number: Standard Deviation! (Available in Excel, of course!)Perhaps this number should be part of the standard test for FSX?HTH

Nice stutter charts (worthy of a Kings Speech!), but charts are difficult to compare, so what you really want is to reduce the charts to a single number: Standard Deviation! (Available in Excel, of course!)Perhaps this number should be part of the standard test for FSX?HTH

Good point man. that came to mind when Corey posted those graphs that looked so much like a Gauss bell, but completely forgot about it. I'll do it right now. Thanks for the suggestionEDIT: standard deviation: Again, the slightly higher FPS in the higher clocked memory tests result in a slightly magnified std dev. Anyway, with so many frames involved (10000 - 11000) and such few stutters (5 - 50 depending on the criteria) there would have to be a huge difference in the number and intensity of the stutters for the std dev to reflect the fact and provide valuable data IMO

Edited March 8, 201115 yr by dazz

Thanks Corey. Truth is there seems to be something there showing faster RAM helps more than what plain max/min/avg figures do. I posted the dropbox link already hehe. Here it is once again ^_^http://dl.dropbox.co...76/stutters.rarTests at 2133 7-10-8-27 Frames, Time (ms), Min, Max, Avg10878, 300000, 24, 50, 36.260<30 1354<20 28<10 4Give me a minute and I'll try tightening those timings a bit moreNo luck. I tried 7-10-10-30 instead:Frames, Time (ms), Min, Max, Avg10785, 300000, 24, 50, 35.950<30 1364<20 29<10 4

Hey! Dazz, thank you for trying, tighter timing will need more QPI/vtt voltage and probably higher dram voltage also to stay stable.Did you see a difference in smoothness visually when flying with high MHz and the tightest timing? Also are you running at CR at 1T?

Vtt (Vccio in LGA1155) plays minimal or no part in OC in SB, I guess because the BCLK sits at 100MHz and it wouldn't even boot at 7-9-7-24. At 7-10-7-25 it BSODed as soon as I fired up FSX. Anyway provided that 7-10-10-30 remained within a 1% of 7-10-8-27 in all areas while 8-10-8-27 was a 2.4% slower it seems that it doesn't matter all that much.If you want to test your current timings vs slower settings, feel free to run your own tests and send me the fraps files. I'll be glad to set the spreadsheets for you if needed be. In all honestly I couldn't tell one test from another, even at the slowest RAM settings, but it's just one test and it doesn't prove anything. There may be other scenarios where RAM speed shines. Bad weather maybe? feel free to suggest anything

Vtt (Vccio in LGA1155) plays minimal or no part in OC in SB, I guess because the BCLK sits at 100MHz and it wouldn't even boot at 7-9-7-24. At 7-10-7-25 it BSODed as soon as I fired up FSX. Anyway provided that 7-10-10-30 remained within a 1% of 7-10-8-27 in all areas while 8-10-8-27 was a 2.4% slower it seems that it doesn't matter all that much.If you want to test your current timings vs slower settings, feel free to run your own tests and send me the fraps files. I'll be glad to set the spreadsheets for you if needed be. In all honestly I couldn't tell one test from another, even at the slowest RAM settings, but it's just one test and it doesn't prove anything. There may be other scenarios where RAM speed shines. Bad weather maybe? feel free to suggest anything

Hey Dazz, I did some test long time ago that's why my conclusion at the time was high MHz + tight timing = less stutterings but not increasing the FPS by much. I run my FSX at unlimited all the time cause if I lock the FPS I'm gettings a lot of stutterings (beats me) and with the latest driver it seems better also, one thing I've notice is the FPS fluctuations from let say 55 down to 20 up to 50 down to 25 again does nor seem to = more stutterings but the new driver produce more shimmerings.Anyway thank you for your testing, much appreciated.

Another instresting stat would be correlation of FPS (max, avg, min, stutt_stdev) vs. benchmarked RAM throughput in GBy/Sec for each RAM/Timing combo (in ascending throughput order).Cheers,- jahman.

Well, the correlation is so evident between all the relevant variables that I don't think it's necessary to calculate a coeficient How could we determine the memory throughput for a given test?

What a road...! Nice work Dazz.Don't be disappointed. Two things were made clear through this thread:1) RAM frequency and latency don't increase Frames Per Second.which translates to-->Don't buy Faster Ram just to Improve your FS FPS. You won't go any higher.2) RAM frequency and latency don't even affect Stuttering, in any obvious/visible/tangible/objective and non-idiosyncratic, worth mentioning way.which translates also to --> Don't buy Faster (or/and "tighter") Ram just to Improve FS' motion fluidity (stuttering). You won't.To sum it up in a more realistic manner, only Size Counts... To put it even more bluntly, when it comes to FS builds, moving from 1333Mhz to 2000+Mhz is a complete rip-off ($$$ / performance gain)... RAM marketing hype and speculation, FS-wise, shone bright once again. That's my reading.Thanks for the hard work, looking forward to finally assemble my new rig and run Corey's benchmark.

I wouldn't conclude that there is no performance gain. If you're truly looking for every possible FPS, 2133 CL7 appears to provide about 1.5 fps above 1600 CL7. It's obviously a small advantage, but hey, it's something. Just don't expect it to improve stutters whatsoever. Another point worth noting is that the stutters appear to happen at a regular time interval - just look at the dips in the plots below. Each dot represents a frame. I see four prominent stutters occuring at 58s, 109s, 152s, and 195s - so approximately every 45 seconds. Perhaps we should investigate the cause of that.