Sign in to follow this  
Michael Moe

Interesting RAM observations (Performance increased)

Recommended Posts

First,

 

I must have been living under a rock for quite a while not having checked my RAM . I must admit i have a really bad knowledge here and thats properly why i first noticed this now. :fool:

 

I have Corsair Dominator DDR3-1600MHZ 4 x 2GB modules with different latency. 2 pieces at CL7 and 2 at CL8. (BUT running at 1333MHZ !!)

 

Now i thought my increased 10-15% performance (FPS) at EKCH FT Gate A20 was about the new installers for V3 BUT its not.

 

I had my RAM speed set at 1333MHZ normally , and just for a test earlier i had increase RAM to 1800mhz (1,65v as thats what they need and stamped on them)  before installing the new EKCH v3 installers and the redwine had me forget all about it :Tounge:  

 

Interesting , i can reproduce a 10-15% increase in FPS by doing so. So what does that tell me ? AN unbalanced system with potientiale for even more FPS i guess.

 

Thats why i need your help here.

 

The motherboard is a Asus Z87-Pro and as far as i can see DDR is supported all the way up to 3200MHZ . Would i gaine even 10-15% further by switching my RAM ? who knows ?

 

Anyone with knowledge for a pair of 2x8GB to go for my system? 

 

Thanks (what a cheap improvement :wink: )

 

 

I7-4770K@4,6 HT OFF 1,25V

2x GTX970SC

SSD128 W10-64bit

Seperate drive for P3Dv3

 

Michael Moe 

Share this post


Link to post
Help AVSIM continue to serve you!
Please donate today!

Hi Michael,

 

From my testing going from 1600mhz to 2133mhz you will gain a solid 3 fps. When scenery is dialed in about 30 fps. So yeah thats 10%.

 

I just recently bought some 2800 ram and noticed another 3 fps increase. 

 

To further prove this I dropped the ram back to 1600 speed and lost 6 fps. So it definitely makes a difference. But thats up to you if buying faster ram is worth the few fps increase.

Share this post


Link to post

Thanks for reporting. Well its a rather cheap 15-20% gain for rather few bucks. What i am a little nerveous about is bying RAM thats not supported. Stated in the manual it says o.c as in overclocked ram but i am not sure. There is some 3000mhz RAM modules for about 150 euro 16GB. How can i tell if the work?

 

Thanks Michael Moe

Share this post


Link to post

Michael,

 

You should go to the ASUS support site and get hold of the latest DRAM Qualified Vendors List (QVL) file for your specific board, they regularly update the list when new memory products become available from the various manufacturers. If a product is on the latest version of the list it will work.

Share this post


Link to post

Michael,

 

You should go to the ASUS support site and get hold of the latest DRAM Qualified Vendors List (QVL) file for your specific board, they regularly update the list when new memory products become available from the various manufacturers. If a product is on the latest version of the list it will work.

 

 

Thanks doing that right now and wow what a jungle  :wink:

 

Michael moe

Share this post


Link to post

Good news, as I've just discovered my mobo is faulty and runs my 1600mhz RAM in single channel only. Thus running at only 800-ish mhz.

 

New system in the pipeline with dual channel 2133mhz, can't wait to witness the fitness  B)

Share this post


Link to post

Good news, as I've just discovered my mobo is faulty and runs my 1600mhz RAM in single channel only. Thus running at only 800-ish mhz.

 

New system in the pipeline with dual channel 2133mhz, can't wait to witness the fitness B)

How did you checked that?

 

Why i am askin is because cpu-z shows 899 mhz under memory section 1.65v on my system . Is that even correct?

 

Michael Moe

Share this post


Link to post

It looks lige the QVL is not up to date over at Asus or maybe it just meens my Motherboard doesnt supports newer DDR3 ram ?

 

Many of these Ram modules is now not in production any more . Any suggeestion?

 

for instance from the list CORSAIR CMZ16GX3M2A2400C10 (Ver4.21) 2400MHZ 

 

http://dlcdnet.asus.com/pub/ASUS/mb/LGA1150/Z87-PRO/Z87-PRO_DRAM_QVL_1029.pdf

http://dlcdnet.asus.com/pub/ASUS/mb/LGA1150/Z87-PRO/Z87-PRO_DRAM_QVL_0820.pdf

 

Thanks

Share this post


Link to post

How did you checked that?

 

Why i am askin is because cpu-z shows 899 mhz under memory section 1.65v on my system . Is that even correct?

 

Michael Moe

 

CPU-Z also shows if you're running single or dual channel in the memory section. Now that I've upgraded my system it shows "Channel #: Dual" and "DRAM Frequency: 1067 MHz" which should add up to 2133 MHz if I understand things correctly. 

Share this post


Link to post

Just an update to those who are interested. I just upgraded my 1600MHZ Dominator DDR3 1.65V ram to 2400 MHZ Corsair . I saw an increase in FPS about 6-7% again going from o/c 1800 Mhz to 2400MHZ.

 

In my case i totally have gone from 46fps (1333MHZ) at gate A20 EKCH FT with the NGX VC with pretty heavy sliders and 2048 textures to 54fps.

 

i sold the old ones and spent 45 dollars. What a early christmas gift to my self  :wink:  Not even Skylake (or maybe just)would have given me this i guess in terms of FPS. 

 

As Rob says its all about having a balanced system. Maybe 3000MHZ would increase 10% again but to exspensive right now.

 

 

 

Michael Moe

Share this post


Link to post

Haswell DDR3 memspeed the most effency you get with 2800 cl10 24/7.

Not all cpu.s can make 3000mhz memspeed it depending how lucky you ar with the lotteria.

Yes you have 2 , first how much it overclock in ghz then how strong memcontroller you get, if lucky you get a 3200 mhz stable one.

For Haswell i recomend 2666-2800 8gb for fligtsim , 16gb is harder on memcontroller 2666mhz recomended not all CPU.s can make 2800mhz stable.

Share this post


Link to post

Yes, it's all about balance ... word on caution on RAM, I've found that good RAM doesn't need to cost a fortune.  Things to look for are who manufactures the RAM chips - I don't mean brand names like Corsair, GSkill, but who actually makes the memory chips like Hynix, Micron.  Also be careful with CL ... I see 3400Mhz RAM but at very high CL.

 

I use several different products to test my RAMs performance, but I like SiSoftware Sandra the most for RAM testing.  On my main system I'm operating at 63GB/s with CL15 3000Mhz DDR4 quad channel with very low latency.

 

I was just testing out DD New York X V2 last night and was locked at a solid (stutter free) 30 FPS (30Hz Vsync) with ASN weather using my high graphics settings (also amazed I was only at 2.8GB VAS usage) using higher quality textures from DD.  VRAM was high 4.5GB used ... but this is a good thing (will post video later) ... yes even the PMDG 737 and 777 can fly DD New York V2 without issue in P3D V3.

 

But yes, it's all about balance and one's graphics settings will be determine by that balance.

 

Cheers, Rob.

Share this post


Link to post

For Haswell i recomend 2666-2800 8gb for fligtsim , 16gb is harder on memcontroller 2666mhz recomended not all CPU.s can make 2800mhz stable.

 

westman, what about Skylake, 8GB or 16GB?

 

gb.

Share this post


Link to post

It's really incredible how many false information is visible in such a short thread, that's why I would like to clarify it a little bit:

 

1. Dual Channel and Double Data Rate (DDR) is NOT the same. While all current RAM modules use DDR, Dual Channel (or Quad Channel) is depending on the memory controller inside your CPU. While most i5/i7 are capable of using Dual Channel, only the -E processors like 5820K on the 2011 (and following) platforms are capable of Quad Channel.

 

2. Some software only reads out the frequency without calculating the "double" of "DDR". That's why you see 1600MHz as 800MHz only or 2133MHz as 1066MHz only in those softwares. Nothing to bother about.

 

3. Dual (or Quad) Channel basically doubles your data transfer rate (bandwith). For example, with a single 1600MHz RAM module you get 12.8 GB/s, for a single 2133MHz RAM module it is already 17.0 GB/s and so on. You can calculate that rate easily like this: 1600MHz -> 1600 x 8 =12800 MB/s. Now, if you use Dual Channel, you have 25.6 GB/s with a standard 1600MHz module and even 34 GB/s with a 2133MHz module (and so on). I realized, that going higher and higher results in a plateau, as most people using a Quad Channel system already providing 51.2 GB/s with 1600MHz modules, do not profit anymore by upgrading to 2133MHz modules (providing basically 68 GB/s).

 

4. Directly linked to the frequency is the latency (CAS Latency or simply CL). The lower those numbers, the faster the memory can be addressed. To calculate THIS, there is another simplified calculation, for example for 1600MHz CL9 modules it is: 1000/800 x 9 -> 11.25ns latency. That means, the ram can be addressed freshly only every 11.25ns (not scientifically correct, but it is kinda like that). Now, if you have a 2133MHz module with CL12, the same calculation (1000/1066 x 12) gives you again 11.25ns, meaning that those 2133MHz CL12 modules only offer the advantage of higher bandwith, but regarding addressing speed, they are the same. That is why you should also consider the CL numbers. I for example run 2133MHz with CL10, this offers not only the higher bandwith but those modules are also "faster" (9.4ns) compared to basic 1600MHz CL9 modules.

 

5. Todays memory controllers are inside your CPU. This means that overclocking your memory results in overclocking your memory controller inside your CPU. In my case, my Ivybride memory controller is officially only ready for 1600MHz. Running my RAM at 2133MHz, I accept that my IMC is overclocked. I tried once 2400MHz (my RAM is actually specified for this) and it did not work, simply because the IMC of my 3770K can not stand 2400MHz in combination with the core clock of 4.5GHz. So, be careful when you overclock RAM, as not only the RAM module is overclocked, but also the memory controller inside your CPU...

 

Taking all this into consideration, you can better compare the RAM modules. DDR4 did not change those calculations, meaning that those basic DDR4 2133MHz CL15 are offering you a bandwith of 17 GB/s (single channel) and a latency of 14ns, so they are pretty bad. If you only think about latency, you will see that plenty of DDR4 modules offer CL15. Now, taking above calculation into account, you can easily calculate the frequency needed for those modules to be as "fast" as basic DDR3 1600MHz CL9 RAM: it is CL12 for 2133MHz, CL15 for 2666MHz and CL17 for 3000MHz.

Share this post


Link to post

 

 I realized, that going higher and higher results in a plateau, as most people using a Quad Channel system already providing 51.2 GB/s with 1600MHz modules, do not profit anymore by upgrading to 2133MHz modules (providing basically 68 GB/s).

 

 

SAAB340 on page 50 of the FSXMark11 thread shows a 5FPS gain in going from 2133 to 4133.

 

http://www.avsim.com/topic/329116-fsxmark11/page-50

 

And 4133 is even further from 1600, so there must be somehow more room for improvement than theory predicts?

 

gb.

Share this post


Link to post

Did you actually understand, what I was writing? I wrote, that a socket 2011 (v3) System does not profit anymore as due to the Quad Channel, the bandwith is already at 51.2 GB/s with 1600MHz Modules. Now you are telling me, that a 4790K running on a 1150 profits from higher clocked RAM? Do you see the difference? YES, as I have written, with a Dual Channel CPU such as the 4790K, you will profit from higher clocks. But now look more closely to the results from SAAB340:

 

1. Running @stock clocks, he varies the RAM clocks from 2133MHz up to 3866MHz and sees a increase in FPS of 7% (clock goes up 81%!!!).

 

2. He then increases the RAM clocks even further to 4133MHz, but also increases the CPU clock to 4.3GHz, which provides him another 9% increase in FPS, which is basically 1:1 the increase in CPU clock (+8%).

 

3. He then compares FPS with a CPU clock of 4.7GHz: increasing the RAM clocks with this processor speed from 2133MHz to 4133MHz provides also 9% increase in FPS in total. BUT: you can clearly see, that the increase of FPS is biggest when he increases the RAM clocks from 2133MHz to 3000MHz (+7%) while the further increase from 3000MHz to 4133MHz only provides an additional 2%, clearly indicating a plateau.

 

Now, if we calculate the bandwith behind those numbers, you will see how right I was: 2133MHz provides you with 34 GB/s in Dual Channel (as I wrote...), 3000MHz provides you 48 GB/s and 4133MHz finally provides you 66 GB/s. That's a wonderful example, why you do not profit that much when you increase your RAM speeds on a Quad Channel system where 1600MHz already provides you 51.2 GB/s. This and nothing more I said.

 

PS: if we take a even deeper look into the results from SAAB340, we see that he does not keep the CL (latency) constant. This further modifies the result and you should be aware of. For example: he uses his RAM with a CL15 (14ns) at 2133MHz but a CL18 (9ns) at 4133MHz. So, besides the higher bandwith, his RAM running at 4133MHz are also 5ns faster (55%!) than running at 2133MHz. In contrast, he uses the same CL numbers for both 2133MHz and 3000MHz, and there you can really compare them. In the end, running at 4133MHz with the same latency (would equal CL28!), would not provide any relevant increase in FPS anymore, not even those statistically irrelevant 2% he observes (due to the lowered latency).

  • Upvote 1

Share this post


Link to post

Did you actually understand, what I was writing? I wrote, that a socket 2011 (v3) System does not profit anymore as due to the Quad Channel, the bandwith is already at 51.2 GB/s with 1600MHz Modules.

 

Yes sorry about that.

My mistake was thinking that Skylake (that SAAB240 tested) has a quad memory system.

I now see that it is only dual.

 

gb.

  • Upvote 1

Share this post


Link to post

 

Thanks for your informative posts relating to memory timings and freq. It was previously a bit mystical to me, but now I can see how it relates to balancing a system for best performance :) Cheers K
  • Upvote 1

Share this post


Link to post

I use basic the same calculation for unload latency CL/mem frq x2000.

 

cl15 2133 14.0ns
cl15 3000 10.0ns
cl18 3886  9.3ns

 

that give approx 0.1 fps 1% faster unload latency, with 56fps at 2133mhz we get 3000mhz 60fps and 3866mhz 61fps.

you gain on both memspeed and CL latency thats clear, i dont see any plateau if locking at unload latency.

Its a lot more complicated then do a basic calulation on cas latency, you have second and third timings, componemt latency and more.

 

To get relivant info on your systems reaction to memspeed with different settings load etc, you need to run tests with for example Intel MLC.

 

balancing a system for best performance or as i like to say, get the best efficiency out of it is not easy.  

Be careful before write things like this ..It's really incredible how many false information is visible in such a short thread.

Its what i like to say, happy flying  with your fast or slow mems...

Share this post


Link to post

Kevin,

 

To add to this ... there is also Cache performance, Cache size, and the ability to set Cache frequency, and HT is ON of OFF.

 

It can be pretty exhausting getting the best efficiency.  My advice, find someone that has done all the leg work with verifiable proof it's working well with the FS platform(s) you use and get the exact same setup to the letter and use the same settings ... and "hope" it works.

 

Or just fly with a few less graphics options checked ... I'm very happy with my main FS PC (per all my videos) at 4.6Ghz HT OFF - CL15 DDR4 3000MHz quad channel (about 61-63GB/s pending which app is used to test) -- in terms of actual FPS in P3D V3.x, I'm about 40% higher FPS than if I used a standard CPU base clock with 2133MHz but most important is smoothness, temps, and reliability.

 

Over the decades of testing of RAM performance there can be gains, but there is a point of diminishing returns.

 

Cheers, Rob.

Share this post


Link to post

Kevin,

 My advice, find someone that has done all the leg work with verifiable proof it's working well with the FS platform(s) you use and get the exact same setup to the letter and use the same settings ... and "hope" it works.

LOL I cant afford your setup Rob ;) Besides, I know I'll be better off understanding exactly why a particular hardware set works best in a particular way rather than just cribbing from someone else!

Cheers K

Share this post


Link to post

LOL I cant afford your setup Rob ;) Besides, I know I'll be better off understanding exactly why a particular hardware set works best in a particular way rather than just cribbing from someone else!

 

My setup is NOT a good example of best performance to $$$ ratio ... absolutely no need for Titan X when a 980Ti will get it done and you can probably get very similar performance with a CPU that costs 1/2 as much.

 

I've been building PC's for 30+ years and I'm always learning something new as technology progresses ... but I honestly wouldn't wish that on anyone, however I applaud your willingness to understand it "all" ... but be warned it's constantly evolving, keeping up with technology is time consuming -- there are some EFI/BIOS settings I have no clue what they do and why I would want to adjust them ... every new motherboard that comes out seems to have another slew of completely new adjustments ... and on and on and on.

 

Out of all this, my most important advice is to NOT go cheap on the motherboard, it's the single most important part of any system build, especially if you plan to keep that system around for more than a year and grow it some.

 

Cheers, Rob.

Share this post


Link to post

My setup is NOT a good example of best performance to $$$ ratio ... absolutely no need for Titan X when a 980Ti will get it done and you can probably get very similar performance with a CPU that costs 1/2 as much.

Understood, I am aiming for an upgrade to a 980ti just looking for the right deal. CPU-wise my old i5 2500k isnt too bad at 4.7Ghz so I may just keep for a while since the cost of new mb, cpu, cooler, mem etc doesnt really look like a good investment of ££/performance (even to me so lord knows what my wife would make of it ;) )

 

I applaud your willingness to understand it "all"

I think its too much to understand it ALL in the time available, I'll settle for 80/90% and keep a decent family/hobby balance :)

 

Cheers K

Share this post


Link to post
On 11/9/2015 at 0:45 PM, turboken said:

Hi Michael,

 

From my testing going from 1600mhz to 2133mhz you will gain a solid 3 fps. When scenery is dialed in about 30 fps. So yeah thats 10%.

 

I just recently bought some 2800 ram and noticed another 3 fps increase. 

 

To further prove this I dropped the ram back to 1600 speed and lost 6 fps. So it definitely makes a difference. But thats up to you if buying faster ram is worth the few fps increase.

Can you send me a link to the 2800 ram u got? U fly in orbx regions?

Share this post


Link to post

considering this thread is over 1 1/2 years old I would seriously doubt that specific memory would be available anywhere today.

 

Vic

Share this post


Link to post

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