The impact of different RAM speeds and timings in FSX

[SAAB340 28]

With the FSXmarkCPU comparison benchmark I’ve found a completely CPU limited scenario in FSX. That’s also the perfect way to test the impact of RAM speeds and timings. Your RAM will affect performance, but only when you are CPU limited. The RAM and the CPU work very close together. Slow RAM will keep the CPU waiting longer for data, wasting precious CPU cycles. In most games this is not a problem as it’s normally the graphics card that is the main bottleneck and the CPU is sitting waiting for the GPU most of the time.

 

FSX is, as we all know, a bit different to most other games. It’s very often fully CPU limited so it is an application where faster RAM has the potential to make a difference. How much? That’s what I’ve had a look at. I’ve tested a variety of different timings and frequencies using the FSXmarkCPU benchmark and here are the results:

 

The same results are shown 3 times in different ways.

 

At the top you can see the FPS achieved at the different speeds/timings. It can be a bit hard to see the impact by just looking at the raw FPS numbers.

 

In the middle you can see the same results but shown in % Vs the standard 1600MHz 9-9-9-28 1T benchmark settings. I find this a much easier way to see the impact of different speeds and timings.

 

At the bottom it tells you the equivalent CPU speed you would have to use in the FSXmarkCPU benchmark to achieve the same FPS at the different timing/frequency settings.

 

As you can see, both RAM speeds and timings make a difference in FSX. The difference might be small for each small step but added together you can get some tangible performance increases. We now have a wide range of speeds of DDR3 RAM to choose from. Using reasonably priced 2133Mhz RAM (or higher) instead of bargain bin 1333 9-9-9 RAM improves your FPS in the same way as changing your CPU speed from 3.8Ghz to 4.1Ghz or more. That’s quite an improvement for barely any extra money. At the moment, when passing 2400MHz the prices of the RAM seems to increase sharply but before than the difference is only a few bucks.

 

The tests have been carried out on a SandyBridgeE system. SandyBridgeE supports 2400Mhz RAM, but I’ve been unable to test at those speeds. It might be my RAM (it’s rated 2133) or it might be my CPU not being capable of those higher speeds. IvyBridge and Haswell all support even higher RAM speeds and I would expect these performance improvements to continue. Bear in mind thou that going from 1333 to 1600 is a large 20% increase in RAM speed, but going from for example 2800 to 3000 is a lot smaller with only 7% increase in RAM speed. If you go for absolute highest performance you’ll end up paying a small fortune for very little improvement just as usual.

 

You might think that this comparison is not valid for the mainstream platform as it has been done using quad channel. But as you can see, I tested with tri- and dual-channel as well and there’s almost no performance impact at all. So FSX makes good use of extra bandwidth from increasing the RAM speed, but can’t really take advantage of the extra bandwidth given by tri- and quad-channel. (I don’t recommend using just dual channel on x79 thou as the BIOS is most likely optimised for quad channel operation on x79).

 

[westman 253]

Thanks.

Great work , i hope that this gone be the proof for they beliving mems have

No impact in fsx.

 

Hassse

[Illegitimi non carborundum 4]

Saab

Can you explain the per cent figure where is the 100% figure in other words in the first column what is 95.5% relative to?

 

I analysed those figures (ANOVA) as presented but could not find any significant difference between them, but that may be because there was not enough data to compare?

 

You didn't state the BCLK (FSB) value used - was that changed from the stock value of 100Hz? If it wasn't then the RAM speed or latency is not likely to have much effect as the transfer rate to and from the cpu is static @ 100Hz. As I understand it faster RAM only comes into its own if you are able to overclock the FSB and not the cpu multiplier - but I could be wrong!!

 

By all means use faster RAM it may well be more efficient but its unlikely that you will see significant increase in performance that you might see with a faster cpu.

 

Regards

pH

[SAAB340 28]

Saab

Can you explain the per cent figure where is the 100% figure in other words in the first column what is 95.5% relative to?

Regards

pH

RAM running at 1600MHz 9-9-9-28 1T it the 100% reference.

 

If you just change the RAM to 1333MHz 9-9-9-28 1T you only get 95.5% of the FPS you get with the RAM @ 1600 9-9-9-28 1T.

 

I don't touch the BCLK here. Its always the same. It's only the memory speeds and timings that I change. Nothing else. The CPU remains static at 4.00GHz at all the tests but as you can see, the different RAM settings alone are able to change the average FPS as much as varying the CPU speed between 3.8 and 4.2

[HLJames 2,568]

Its official memory speed/latency matters.

 

HLJAMES

[westman 253]

Sticky?

[McLaren 21]

Just in time, I am in the process of OCing some new RAM. Thanks for the info! 

[fghdgdfdfgfgf 111]

Yeah, last week I upgraded the ram on my 2500k (still running strong at 5 GHZ) from 1600 8-8-8-24 to 2133 9-11-10-27.  I'm happy to report that Im enjoying a nice lil performance bump in the virtual cockpit of my PMDG 737. The $125 was a solid investment,

 

....In other news, I did buy a ASUS Z87 Pro and a 4670K today but .....it's for my wife. :-) 

[SAAB340 28]

....In other news, I did buy a ASUS Z87 Pro and a 4670K today but .....it's for my wife. :-)

Any chance you could put FSX + Acceleration or SP2 on it temporarly and run FSXmarkCPU? That way we could add the i5 Haswell to the comparison

[fghdgdfdfgfgf 111]

Sure, if it's possible to install the service pack without having to valdidate the installation with Microsoft -I don't want to unvalidate my Sandy box.  ....I'll put this thing together as soon as I decide whether or not I want to get her a new case ....and I'm hoping her old 550w p.s. wil be enough ( the box is from vintage e6600 days). :-)

[Bjoern 862]

I very much agree that RAM timings and frequency are very important for FSX as CPU clock itself.

A factory equipped with the newest and fastest machines will never run up to spec when raw materials and goods are pulled from and put into the storage at a snail's pace.

 

 

 

Tidbit of history:

 

Managed to grab a pair (2*2GB) of G.Skill Ecos* back in 2011 and running them at 1752, 6-8-6-24, tRD 7 and ~1.6V. Best memory chips in history and the difference between stock and OC is notable.

 

But as this is all on a 2008-ish (Q9450, X48...) board, so the comparison for a FSX database is moot.

 

 

*Stock: 1600, 7-8-7-24, 1.35V