FSXMark11

[Superpilotv2 2]

Hi, how´s the 560 doing? I´m on buying the same card, so I thought you could tell me something about.

[wil02081 0]

Hi, how´s the 560 doing? I´m on buying the same card, so I thought you could tell me something about.

I have only played fsx and need for speed on it so I can't say with other games. But it works great, no complaints. You can't use nHancer with it but you can use nVidia's inspector to change profile settings. There was a work around for getting nHancer settings but I don't remember off the top of my head - I just used the nVidia settings. My main issue is my cpu, I'm going to swap to a i5 2500k as I'm not too impressed with the AMD x6 compared to the intel bunch.

[Superpilotv2 2]

I have only played fsx and need for speed on it so I can't say with other games. But it works great, no complaints. You can't use nHancer with it but you can use nVidia's inspector to change profile settings. There was a work around for getting nHancer settings but I don't remember off the top of my head - I just used the nVidia settings. My main issue is my cpu, I'm going to swap to a i5 2500k as I'm not too impressed with the AMD x6 compared to the intel bunch.

All right. I´ve allready tweaked my grafic´s setting with nVidia inspector, so I´m kinda experienced to that. Leaning towards the EVGA SSuperClocked edition with 850MHz. With that and my i5 at 4.2 GHz or more it´ll be an real pleasure. Btw. you´ll be amazed how the enjoyment in FSX changes If you don´t have to look for your frames.

[SpiritFlyer 987]

Btw. you´ll be amazed how the enjoyment in FSX changes If you don´t have to look for your frames.

^^^^Exactly! This is the primary lesson to be learned about comparing results in FSX.^^^^Kind regards,

[Superpilotv2 2]

^^^^Exactly! This is the primary lesson to be learned about comparing results in FSX.^^^^Kind regards,

Yeah, the new card is working and everything is all right now. FPS now more or less stable a 30 whan I´m not at really big hubs.

[Chocorua 112]

So finally I took a fresh install of Win7-64 and FSX, applied all the service packs, tweaks, etc., tweaked Windows a bit and ran the tests over and over on this 1100T X6 @ 3.8. Here are the numbers:test 2: Min 14 Max 31 Avg 22.3test 3: Min 15 Max 31 Avg 22.2test 4: Min 14 Max 31 Avg 22.3Funny, these are the same results as wil02081's 3.0GHz X6. It says something. I just don't know what... :-)No Excel to upload these correctly, but maybe it will be of interest to some. Meanwhile, I can and do get very nice performance in most scenarios. I am pondering, however, ditching this platform and going to i5-2500K. All those 40's look so dam good!Cheers,

[dazz 69]

So finally I took a fresh install of Win7-64 and FSX, applied all the service packs, tweaks, etc., tweaked Windows a bit and ran the tests over and over on this 1100T X6 @ 3.8. Here are the numbers:test 2: Min 14 Max 31 Avg 22.3test 3: Min 15 Max 31 Avg 22.2test 4: Min 14 Max 31 Avg 22.3Funny, these are the same results as wil02081's 3.0GHz X6. It says something. I just don't know what... :-)No Excel to upload these correctly, but maybe it will be of interest to some. Meanwhile, I can and do get very nice performance in most scenarios. I am pondering, however, ditching this platform and going to i5-2500K. All those 40's look so dam good!Cheers,

Thanks for sharing Greg. Interesting results as you say when compared to wil's. Would you mind running it again at 3GHz or something please?

[Chocorua 112]

Thanks for sharing Greg. Interesting results as you say when compared to wil's. Would you mind running it again at 3GHz or something please?

Sure thing, I'll get at it later today and post the numbers, should be interesting indeed.

[Chocorua 112]

More strangeness: the MB will not allow me to under-clock the CPU as it freezes at "checking NVRAM". At this time I can only get it to boot at 200x19=3.8. I made some attempts at other multipliers but so far no joy. So for the moment, 3.0ghz test results not possible.

[dazz 69]

No worries Greg, thanks again

[kama2004 74]

Hi all,Spirit flyer pointed me to this thread to look at performance of GPU in a FSX setting as I was hesitating between a GTX 560 and GTX 580.I decided to apply some statistics analysis (multi regression) on the benchmark data for the Sandbridge processor only (I5 and I7) to find the relationship between test FPS results and system characteristics. I found the following relationship in the data reported.FPS (over Boston, vanilla clean FSX install) = (6.32 x Overclock speed) + (0.002454 * GPUpassmark) + (0.004729 x Ram speed)Where :Overclock speed = the speed you run the processor (I5 of I7 doesn’t matter)GPUpassmark = the video card benchmarking value found on http://www.videocard...h_end_gpus.htmlRam speed: as the name imply (1333, 1600, 1866, 2133)With this relation in hand, I was able to calculate the expected gain in FPS for a 580 card over a 560 with my system specs. I was also able to calculate the gain expected from increasing ram speed or overclocking the system even more. All results below are compared to base case and only one option at the time (no combination).base case GTX 560, OC = 4.50, GPUpassmark = 2744, RAM =1866, FPS formula = 44.0 Option GTX 580, OC = 4.50, GPU = 3909, RAM = 1866, FPS formula = 46.9, Gain over base = 6.5%Option Ram at 2133, OC = 4.50, GPU = 2744, RAM = 2133, FPS formula = 45.3, Gain over base = 2.9%Overclock + 0.1 Mhz, OC = 4.60, GPU = 2744, RAM = 1866, FPS formula = 44.6, Gain over base = 1.4%So I can expect a gain in speed of 6.5% from a switch from GTX 560 to GTX 580, 2.9% from fitting a ram at 2133 Mhz vs 1866 Mhz and about 1.4% gain for each 0.1 Mhz overclock frequency added.Could be interesting to add price increase for each change witch could help people to determine what is the next most effective move to increase a system performance when budget is tight.This only is valid for the Boston testing and may be not indicative of others situation.The results are just considering FPS, not others improvement in quality of image (less shattering, blurring etc..)Conclusion: I agree taht the GPU is a significant factor in the performance of any new rig.Bottom line, I will return the two 560 card I had bought and will order two 580…… just to be on the safe side ;-)For the statistician in you , some more details on how the formula was derived===============================================================================I run a multiple regression using initially the maximum number of parameters to find the one that affect performance. I included in this first run the processor type, overclock speed, GPU benchmark (as a proxy to their performance in fsx), the RAM speed and latency and the HDD/SDD average data transfer rate.This analysis (t and P-factor) revealed that the processor type is not statistically significant, neither the ram latency. The SSD also didn’t show a significant improvement statistically (which seems impossible for me as the SSD is such an improvement in game loading speed, but maybe it have less effect in flight). The SSD effect may also not shown due to the fact that there is a relationship in the data (observed by inspection) between overclocking and SSD purchase. The people that tend to overclock a lot also purchase SSD so a statistical analysis cannot differentiate the two causes and the effect of SSD is somewhat blended in the OC multiplication factor in the equation.This first analysis confirmed that 3 factors explains the variation in FPS : OC speed (probably blended with SSD), GPU benchmark and RAM speed.I thus produced a first version of a formula that just considered those 3 factors.Then I applied the formula to the 28 Sandbrige test result and find a good agreement with 26 of them. Two tests are clearly outliers. These are Ryanbackum test at 29.5 fps (a result 8 FPS below the statistical formula) and the Flying Fred test at 48.2 fps (5 FPS over the statistical formula). All the others one were within 1 or 2 FPS of the formula.I eliminated those the two outliers test as being suspects (wrong procedure or something that makes their system behave differently from the rest).Then I issued a final formula with the 23 remaining tests. Agreement between the formula and the real test is very good with 16 of them showing less than 1 FPS between the formula and real results while 7 of them are below 2 FPS of differences. Other statistical indicators shows a robust relationship between the FPS and the 3 factors.Hope this could help to analyze the benchmark results.Pierre

[Superpilotv2 2]

Thanks Pierre for that formula!Sound like lots of testing till that worked.Used the formula and got 40,83 FPS withg my new GTX 560 SC instead of 36,38 FPS with my old GTX 260. No big difference, only 4 frames or 12% gain.

[SpiritFlyer 987]

Hi Pierre,Excellent analysis. I will be pondering and relooking at the available evidence in light of your findings. It seems more or less what I thought was the case. Thank you for the clarity.Kind regards,

[cmeeks 8]

I love this forum! Pierre, that's an AWESOME use of the results! If you don't mind me asking, what was your R^2 value for the final formula? Seems like the constant for the CPU multiplier should vary based on the CPU socket due to differences in IPC. Perhaps if we used a CPU benchmark score instead of the raw CPU speed, we could get a more accurate formula. After all, we know that an i7-920 at 4.0GHz < i7-2500k at 4.0GHz, for example.It would be really cool if we could find independent hardware benchmarks that relate well to FSX performance to run a multiple regression analysis on. Something like:A*(IBT Score) + B*(GPU Passmark Score) + C*(MaxxMEM score) = FSX performanceThat way people wouldn't have to jack around with doing a fresh FSX install and making sure to run the FSX benchmark properly. We would still need some people to contribute results for the sake of regression.The question is, how do you determine what variables even affect FSX performance? Passmark performance, for example, could be completely irrelevant or just a bad indicator of FSX performance. I think it has something to do with p-values and statistical significance or something like that... I was terrible at statistics.Finally, it seems like the model representing FSX performance wouldn't be linear... but I wouldn't know how to generate any other model in the first place. It would need to be able to indicate a bottleneck in the system caused by a terrible GPU matched with a really high end GPU, for example.

[kama2004 74]

What was your R^2 value for the final formula?Seems like the constant for the CPU multiplier should vary based on the CPU socket due to differences in IPC. Perhaps if we used a CPU benchmark score instead of the raw CPU speed, we could get a more accurate formula. After all, we know that an i7-920 at 4.0GHz < i7-2500k at 4.0GHz, for example.It would be really cool if we could find independent hardware benchmarks that relate well to FSX performance to run a multiple regression analysis on. Something like:A*(IBT Score) + B*(GPU Passmark Score) + C*(MaxxMEM score) = FSX performanceThat way people wouldn't have to jack around with doing a fresh FSX install and making sure to run the FSX benchmark properly. We would still need some people to contribute results for the sake of regression.I think it has something to do with p-values and statistical significance or something like that.Finally, it seems like the model representing FSX performance wouldn't be linear... but I wouldn't know how to generate any other model in the first place. It would need to be able to indicate a bottleneck in the system caused by a terrible GPU matched with a really high end GPU, for example.

@Cmeeks,I will answer the easy ones.... and ask questions for the more difficults....R square is 0.931257 which is excellent. In statistician wording, it means that 93% of the variation in FPS in the benchmark test data are explained by the formula. Generally we consider that 75% or more is an indication of good fit for data.I didn't include other processor in my analysis than the I5 and I7. So not sure the formula will stay valid outside of this realm. I was more interested in the I7 but as many people are of opinion that there is not a huge advantage of selecting the I7 over an I5 (at least for FSX). So I dropped the I5 also in the analysis. Could be relatively easy to include other processors in the analysis.In one of the first pass, I assigned the CPUpassmark value of 7448 for the I5 and 10035 for the I7. I looked at two statistical indicators that are the t statistic (which should be greater than 2). the t statistic was at 0.54 showing a very weak relationship with the CPU passmark value. the other statistical indicator was indeed the P-factor (your memory was good). I got 0.59 as the P-factor and we normally consider that value below 0.05 indicates a strong relationship. So here the CPU failed both test. The selection of the CPUpassmark could be questionnable as an appropriate proxy for CPU performance in FSX but that was the best I had in hand and that could be related to something physical and not pure assumption.t and P-factor for the 3 retained factor are:OC : t = 13.1, P-factor = 5.42 E-11 (the strongest)GPU : t = 4.37, P-factor = 0.000032 (very strong also but less than OC)Ram : t- 3.17, P-factor = 0.0049 (the weakest of the 3 but well over the threshold for statistical significance.... )Additional relationship development.... I will need to understand first what are the IBT score and MaxMemm score first (just recently got interested in FSX performance measurement......!!).By the way, if anyone is interested in receiving the Excel spreadsheet I used to perform additional testing. I can provide this for you.... We can achieve additional development faster probably. Warning; its not well documented...Non linearity: Could be non linear. But R2 is pretty strong so at first glance it is not needed to improve it by changing the type of relationship beyond linear. Also, could be dangerous on small sample to select non linear models just because they seems to fit the data better. Could be coincidential and we may end up selecting a formula that gives a better fit but that have no connection with reality.The big problem to develop a formula that is foolproof for bottleneck identification is that we don't have to much of them in the base test performed to date. So not sure the formula could be robust enough in combination of hardware that were not represented at all in the set of data used to create the formula. For the moment the set of data seems to have a point in common; people select the best combination of hardware for a given budget. But if the quantity of data increase, then it may be possible to identify bottleneck.@SuperpilotV2; This is an extrapolation outside of the available data as there was no GTX 260 cards in the benchmarked machines having I5 and I7. So it could be interesting to test the formula to see the results but there is no guarantee that the formula is valid for others cards than 580, 570, 560, 480, 460, 460 SE and HD 6990, Formula can never be better than the base data used to create it.Regards,Pierre