Jump to content
Sign in to follow this  
tfm

Effect of RAM speed & timing on FSX performance

Recommended Posts

It is a pity that there are so few head-to-head tests for FSX. I am not well qualified to do this sort of thing, not least because I lack the patience for it. But I did some mind-numbingly dull tests myself last night. Basically I recorded a Bell helicopter flight over Manhattan with inadvisedly high settings and used "replay video" with the RAM at different speeds. I used FRAPS to record the FPS etc for the first 180 seconds. This is on an i7 975 with 6Gb of RAM rated for 2000 7-7-7-20. The GPU is a GTX 285 factory-clocked to 702MHz. The i7 was clocked at BCLK 133 with a 33 multiplier giving a CPU speed of 4410MHz. The exception is the 1987MHz RAM test, for which I had to make a change to BCLK 142 with a 31 multiplier giving a CPU speed of 4399MHz. I also tried testing at 2000MHz RAM, but FSX was unstable at the settings I used and I couldn't be bothered to try to fix it.Here we go:And here is a graph for those who like them: Several health warnings:(i) Apparently it is possible that FRAPS itself may interfere with the measuring process.(ii) This is on a clean Win 7 RC 64-bit and FSX installation, but I have not fine-tuned the OS so results might be distorted by (eg) background services. The second "run" for the 1600MHz RAM may have been dragged down by something like this.(iii) I used the "Fair weather" setting so there is the scope for distortions due to variations in cloud density and so on.(iv) No doubt there are a myriad of things that could push individual results artificially up or down: I have made no attempt to exclude them and obviously a test using just two runs is far from ideal.My observations: (i) First, draw your own conclusions.(ii) Secondly, my own view is that these numbers are consistent with better RAM delivering a subtle improvement in smoothness, which is what I have seen "subjectively" in the past. There seems to be a correlation between the speed / timing of the RAM and the number of frames that FSX renders during the same 180 second flight. I don’t know how you measure “smoothness” but I suspect that this is a close proxy. (iii) My hunch is that with the i7, the law of diminishing returns has begun to make itself felt pretty keenly. As I recall, the jump from DDR2-800 to DDR3-1600 which I made with last year's computers was more noticeable than the differences shown in this test. I speculate that the i7 CPU itself makes better use of available RAM bandwidth, so while i7 systems can still benefit from getting the right RAM speed and timings, they may be more forgiving than older generations of CPUs. This, I think, is pretty much what Nick and other technically-minded individuals have been saying for some time.(iv) Nobody has ever said that faster or better-tuned RAM is a silver bullet for FSX. We all know there aren't any of those. And I am not suggesting that it's good value for money: that's a different debate and not a productive one, in my view.(v) Despite the "blip" in the results for the 1600MHz RAM timing on its second run in my test, I concur with those who say that 1600MHz RAM at 6-6-6 or better is a sensible compromise. It would be interesting to see 2000MHz at 7-7-7, but I have read that the stability issues with this setting may be related to a widespread problem in getting the i7 stable when the Uncore Clockspeed exceeds 4000MHz (which it must do, for RAM of 2000MHz or more).Tim

Share this post


Link to post
Share on other sites

Interesting tests. I agree with you that there are too few tests around FSX. As long as the tests are well explained in terms of hardware and software, users can only benefit. Too many times, negative comments are received. So good work Tim,Pierre


Pierre

Share this post


Link to post
Share on other sites

Excellent post, thanks for taking the time to test this.Since minimum latency in DDR RAM is CAS + TRCD+CR, we can determine the best case scenario access time at each clock speed & timing combination you have posted.@ 1066 5-5-5 you have an 11 cycle minimum access time giving a real-time latency of 10.31ns or 10.31 billionths of a second@ 1333 5-5-5 you have a 11 cycle minimum access time giving a real-time latency of 8.25ns or 8.25 billionths of a second@ 1600 6-6-6 you have a 13 cycle minimum access time giving a real-time latency of 8.12ns or 8.12 billionths of a second@ 1866 7-7-6 you have a 15 cycle minimum access time giving a real-time latency of 8.03ns or 8.03 billionths of a second@ 1987 7-7-7 you have a 16 cycle minimum access time giving a real-time latency of 8.05ns or 8.05 billionths of a secondAs you can see, there is marginal latency differential between the various settings, except the initial 1066MHz speed which is simply too low. Since CPU clock remains nearly identical between the different settings, clearly it is bandwidth that makes the difference in FSX RAM performance, not latency. Note: if we were to repeat the test with different latency settings for each RAM speed we would see a difference in performance yet again, but it would be even less than the one we already observe.

Share this post


Link to post
Share on other sites
Guest Nick_N

Great post Tim as usual and some people may miss the point so I wish to point out 4 very important items that may appear minor on the surface as reported however you have hit the nail in the head in your observations. Key word with FSX and testing using a benchmark is in fact: OBSERVATIONFirst, the flight path itself can alter the test result. With FSX we not only want an urban environment but a balanced menu of everything FSX demands of the system to include urban, rural, airport, air traffic, ground/water traffic, topography with respect to mountains and autogen which will allow both vegetation and buildings to be placed in the loop. Garys original FSmark benchmark flight was specifically designed for that purpose and as I recall he discussed the flight path with Phil Taylor as well as other issues such as the FSX priority system. Although one can use the test to make observations in different scenarios, do be aware that bench or load testing FSX correctly to the system is about much more than just urban and high impact cities. As a matter of fact a large pine forest can bring FS to its knees faster than a large city. As I recall Garys flight also applied a specific weather to the mix which added more clouds than usual in the fair weather theme.. I would need to check that since its been a while but I am fairly certain the weather was set up specifically as well for his flight. Also, and very important is the factor of SPEED.. the Bell is not going to be traveling at 250Kts so there will be changes with terrain in a jet ie; did we OBSERVE sharp and clear ground textures during the test with sharp distance visuals and good autogen fill in the distance without being extremely slow to appear on the screen @ 250Kts?Second, your statement: (i) Apparently it is possible that FRAPS itself may interfere with the measuring process.You are correct. Not only did I discover that FRAPS has a defined impact on the result, I found that if the default FSX traffic is not used it will further skew the benchmark result. I discovered when I first tested FSX with FSMark that with WOAI or other addon traffic installed the average benchmark FPS was skewed. There was no consistent result possible (at that time) which could be examined to draw a reasonable conclusion. Once default air traffic was restored the results became stable and consistent. FRAPS itself can alter the test result based on the dynamics of FSX and how the priority system of FSX is addressing the flight. Third, (iii) My hunch is that with the i7, the law of diminishing returns has begun to make itself felt pretty keenly.Excellent deduction Tim but if I may, please allow me to alter that statement slightly: With i7, the law of the diminishing NORTHBRIDGE has made itself felt pretty keenly.Your deduction is quite correct Tim however the reason for it is the NB memory controller is no longer in the mix and is now located inside the CPU. Our bandwidth efficiency has gone from 60% max (WITH a proper clock in FSB/STRAP) to nearly 90% without any formula for lowering Northbridge memory controller latency required, therefore the result will become

Share this post


Link to post
Share on other sites

At the risk of boring everyone to death, I managed to get my PC stable (enough) for two runs at 2000MHz. Perhaps I'll try to keep it there, to get something useful out of this little foray into uber geek-ism.Here is an updated table:And here is an updated graph:For this the settings were BCLK 143 and multiplier 31 = 4432 MHz CPU.By the way the trick to getting the RAM to 2000+MHz seems to be to boost the vQPI: in my case, to 1.425v. Whether this is safe and/or good enough for long-term stability - who knows?I've taken screenshots of the Everest cache & memory benchmarks for each of these RAM settings, but I won't burden this thread with them unless there's enough interest.And in answer to Nick's percipient question: having used the good stuff, personally I wouldn't happily settle for slower RAM. For the way I like to use the sim, in the heavies under demanding graphics conditions, every little helps; and the RAM makes a difference: not much perhaps, but some; and enough IMHO to make it worthwhile, albeit in my own possibly idiosyncratic scale of values.Tim

Share this post


Link to post
Share on other sites
Guest Nick_N
By the way the trick to getting the RAM to 2000+MHz seems to be to boost the vQPI: in my case, to 1.425v. Whether this is safe and/or good enough for long-term stability - who knows?
According to my contact a Intel.. 1.42v would be fine assuming the airflow in the tower is adequate and that setting will impose no long term detrimental effect, however, his advice was not to exceed that for long periods of time and he too questioned the practice of the memory companies advising to run 1.60-1.65Which is why when I post QPI/DRAM (QPI/Vtt) I specify a max of 1.42 (1.425 is fine Tim)Your results appear in line .. as the memory speed increases with the correct CAS, the perf increases and as I posted in other threads.. however observation is the key to placing value into the numbers... by the engineering the next step UP in perf which may have value over DDR3 1600 6-6-6 is DDR3 2000 7-7-7The question is.. will the system run it stable within safe QPI and other votage values

Share this post


Link to post
Share on other sites

Very interesting post, a little above my head with the technical speak....Can anyone answer this question in basic terms?The faster DDR3 memory (DDR3 xxxxx high number) you get the better FSX performs with regards to texture loading/smooth flight???


| FAA ZMP |
| PPL ASEL |
| Windows 11 | MSI Z690 Tomahawk | 12700K 4.7GHz | MSI RTX 4080 | 32GB 5600 MHz DDR5 | 500GB Samsung 860 Evo SSD | 2x 2TB Samsung 970 Evo M.2 | EVGA 850W Gold | Corsair 5000X | HP G2 (VR) / LG 27" 1440p |

 

 

Share this post


Link to post
Share on other sites
Very interesting post, a little above my head with the technical speak....Can anyone answer this question in basic terms?The faster DDR3 memory (DDR3 xxxxx high number) you get the better FSX performs with regards to texture loading/smooth flight???
Theoretically, although this can't actually be quantified, no such thing as a texture loading benchmark for FSX.

Share this post


Link to post
Share on other sites

Thanks, and the lower the latency the better (in theory) the texture loading could be?


| FAA ZMP |
| PPL ASEL |
| Windows 11 | MSI Z690 Tomahawk | 12700K 4.7GHz | MSI RTX 4080 | 32GB 5600 MHz DDR5 | 500GB Samsung 860 Evo SSD | 2x 2TB Samsung 970 Evo M.2 | EVGA 850W Gold | Corsair 5000X | HP G2 (VR) / LG 27" 1440p |

 

 

Share this post


Link to post
Share on other sites
Thanks, and the lower the latency the better (in theory) the texture loading could be?
What this thread has shown us is that absolute bandwidth is more important than an the extra couple tenths of a billionth of a second lower timings will buy you. IOW: DDR3 2000 @ 7-7-7 is better than DDR3 1600 @ 6-6-6. Now if there were CAS5 DDR3 1600 I think it would be a different story, as then we'd be talking about a difference of 11/12 cycle access time vs. 15/16 cycle access time with not enough of a difference in clockspeed to make those individual cycles short enough to make up the difference.

Share this post


Link to post
Share on other sites

Ok but I thought I read somewhere that the DDR3 2000 sticks were having issues with a lot of the motherboards out there today?


| FAA ZMP |
| PPL ASEL |
| Windows 11 | MSI Z690 Tomahawk | 12700K 4.7GHz | MSI RTX 4080 | 32GB 5600 MHz DDR5 | 500GB Samsung 860 Evo SSD | 2x 2TB Samsung 970 Evo M.2 | EVGA 850W Gold | Corsair 5000X | HP G2 (VR) / LG 27" 1440p |

 

 

Share this post


Link to post
Share on other sites
Guest Nick_N
What this thread has shown us is that absolute bandwidth is more important than an the extra couple tenths of a billionth of a second lower timings will buy you. IOW: DDR3 2000 @ 7-7-7 is better than DDR3 1600 @ 6-6-6. Now if there were CAS5 DDR3 1600 I think it would be a different story, as then we'd be talking about a difference of 11/12 cycle access time vs. 15/16 cycle access time with not enough of a difference in clockspeed to make those individual cycles short enough to make up the difference.
Tims results are consistent with the engineeringas the memory speed increases with the correct CAS, the perf increases and as I posted in other threads.. however observation is the key to placing value into the numbers...As I posted above to Tim there has been a defined change with i7 which is the northbridge is no longer a factor in results from the past, therefore the increase will not be as defined but it will still present itselfThere are many posts on the net in threads early on when i7 was initially released where the FSX user was running DDR3 1333/1600 9-9-9 memory with poor results and upgrading to 7-7-7 or 6-6-6 corrected those perf issuesTiming is indeed a key factor..

Share this post


Link to post
Share on other sites
Ok but I thought I read somewhere that the DDR3 2000 sticks were having issues with a lot of the motherboards out there today?
RyanI think there are two separate reasons why one might buy the 2000+MHz RAM.First, it gives you greater flexibility by giving you a larger margin, or overhead if you like. You seldom know exactly where your overclock will end up. You might find that your CPU prefers a particular BCLK which leaves you with uncomfortable multipliers for the RAM if you've got limited options. You might get stuck with the choice between a high CPU speed and a low-ish RAM speed, or the other way around. The higher the RAM speed, the less likely you will have to make that choice and the more likely you can have your cake and eat it.Secondly, if you CAN get the 2000+MHz stuff stable at its rated speeds, there is a slight performance advantage provided the timings are right. As I have just discovered, it does seem to be possible to get the 2000+MHz stuff stable but you (or at least, I) have to push one of the voltages further than might be expected - and beyond Intel's specification.Tim

Share this post


Link to post
Share on other sites

ahhhh so with the higher speed you could actually underclock the RAM, well, lower the speed if you need a certain multiplier when overclocking?


| FAA ZMP |
| PPL ASEL |
| Windows 11 | MSI Z690 Tomahawk | 12700K 4.7GHz | MSI RTX 4080 | 32GB 5600 MHz DDR5 | 500GB Samsung 860 Evo SSD | 2x 2TB Samsung 970 Evo M.2 | EVGA 850W Gold | Corsair 5000X | HP G2 (VR) / LG 27" 1440p |

 

 

Share this post


Link to post
Share on other sites
Guest Nick_N
ahhhh so with the higher speed you could actually underclock the RAM, well, lower the speed if you need a certain multiplier when overclocking?
careful with this one.. There is another factor involved with memory design. Under-clocking memory rated for higher speed can present a different result than posted however that is typically observed with pre- i7 systems..I went through this with another member on this board. He had 1st generation DDR3 1800 memory downclocked to 1600 on the same timing and having all sorts of problems with FSX. On advice he obtained correctly spec'd DDR3 1600 memory and many of his issues ceased. The rest was in correct tuning of FSX. That result followed issues that have been defined with 1800 and above memory product and the northbridge chipsets of the past. Not enough information at this time to make a good call on that with i7 but it would be a good test to apply and check It would require the test system switch memory product to make a correct assessment on those lines with i7 but I suspect since the memory controller is no longer in the northbridge presenting stability/perf issues it may not apply. Quickly looking at the numbers here I would say downclocking should not present a degenerated result with i7, however, we all know there is theory and then there is field application

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  

  • Tom Allensworth,
    Founder of AVSIM Online


  • Flight Simulation's Premier Resource!

    AVSIM is a free service to the flight simulation community. AVSIM is staffed completely by volunteers and all funds donated to AVSIM go directly back to supporting the community. Your donation here helps to pay our bandwidth costs, emergency funding, and other general costs that crop up from time to time. Thank you for your support!

    Click here for more information and to see all donations year to date.
×
×
  • Create New...