[BradleyW]

I've often wondered if faster RAM helps when you are playing a game that's clearly facing optimization issues on the CPU side of things. We've all been here. From the grass land of Crysis 3 to the Shanty Town in Tomb Raider, our fps has fallen, CPU levels raised and GPU's sit dormant. Time to see if the RAM contributes.

Test System:
CPU : 3930K 4.5GHz HT
GPU: AMD 290X CFX (2)
MB: Gigabyte X79-UP4 Rev 1.1
OS: Win 10 64 bit.

RAM speeds and timing information:

GEil Evo Potenza 2666MHz DDR3 (4x4GB):
1333MHz @ CL 8
1600MHz @ CL 8
1866MHz @ CL 9
2133MHz @ CL 10
2400MHz @ CL 11

All the results you are about to see are the minimum recorded fps.

Test One - RAM Speed:

Tomb Raider ; Ultimate Preset, 1080p, FXAA.
Location: Helicopter Pad overlooking Shanty Town during sunrise.

1333MHz = 62 FPS
1600MHz = 67 FPS
1866MHz = 69 FPS
2133MHz = 72 FPS
2400MHz = 74 FPS

Crysis 3 ; Very High, FXAA, 1080p
Location: Welcome To The Jungle, overlooking mine field within the park.

1333MHz = 78 FPS
1600MHz = 82 FPS
1866MHz = 83 FPS
2133MHz = 84 FPS
2400MHz = 86 FPS

Hitman Absolution ; Ultra Preset, FXAA, 1080p
Benchmark Min fps.

1333MHz = 66 FPS
1600MHz = 69 FPS
1866MHz = 72 FPS
2133MHz = 74 FPS
2400MHz = 76 FPS

Fallout 4 ; Ultra, Light Rays Low, 2560 x 1080
Location: Outside "Diamond City", overlooking the streets during daytime.

1333MHz = 27.3 FPS
1600MHz = 28.8 FPS
1866MHz = 29.1 FPS
2133MHz = 29.3 FPS
2400MHz = 29.9 FPS

Assassin's Creed Unity ; Ultra, HBAO+, PCSS, 2MSAA, 2560 x 1080
Location: Random location overlooking large group of people.

1333MHz = 54 FPS
1600MHz = 56 FPS
1866MHz = 58 FPS
2133MHz = 60 FPS
2400MHz = 62 FPS

Mad Max ; All Settings Max, 1080p
Location ; Rot N Rusties - Enemy Camp (South).

1333MHz = 70 FPS
1600MHz = 74 FPS
1866MHz = 74 FPS
2133MHz = 74 FPS
2400MHz = 74 FPS

Rise of the Tomb Raider, Very High, 1080p
Location ; Soviet Installation (Small house after the second zip line).

1333MHz = 45FPS
1600MHz = 47FPS
1866MHz = 48FPS
2133MHz = 49FPS
2400MHz = 50FPS

Well, the results are interesting and I'm pleased to be using 2400MHz.

_____________________________________________________________________

Test TWO - Memory Single Rank vs Dual Rank:

This next test will determine the difference between single rank and dual rank memory in synthetic tests and CPU bound locations within 3D applications:

Memory Information and timings:

GEIL Evo Potenza 2666MHz DDR3 (4x4GB) SINGLE RANK
2400MHz CL11
2133MHz CL10
1866MHz CL9

Hyper-X Beast 2400MHz DDR3 (4x8GB) DUAL RANK
2133MHz CL10
1866MHz CL9

AIDA64

1866MHz

GEIL Single Rank
Copy = 43536MB/s
Latency = 60.3ns
Read = 47247MB/s
Write = 49435MB/s

Beast Dual Rank
Copy = 48364MB/s
Latency = 64.3ns
Read = 51126MB/s
Write = 52219MB/s

2133MHz

GEIL Single Rank
Copy = 48243MB/s
Latency = 56.5ns
Read = 51999MB/s
Write = 54927MB/s

Beast Dual Rank
Copy = 53510MB/s
Latency = 61.5ns
Read = 55223MB/s
Write = 58.98MB/s

2400MHz

GEIL Single Rank
Copy = 52364MB/s
Latency = 55.6ns
Read = 55777MB/s
Write = 59966MB/s

3DMark API Feature Test

1866MHz

GEIL Single Rank
DX11 S = 1 405 276
DX11 M = 1 307 128
Mantle = 21 334 564
DX12 = 17 240 632

Beast Dual Rank
DX11 S = 1 437 121
DX11 M = 1 335 030
Mantle = 23 785 750
DX12 = 17 950 241

2133MHz

GEIL Single Rank
DX11 S = 1 484 028
DX11 M = 1 366 885
Mantle = 22 814 225
DX12 = 17 453 983

Beast Dual Rank
DX11 S = 1 497 144
DX11 M = 1 355 706
Mantle = 24 153 724
DX12 = 18 065 698

2400MHz

GEIL Single Rank
DX11 S = 1 479 394
DX11 M = 1 391 146
Mantle = 23 081 269
DX12 = 17 657 303

3DMark Fire Strike

1866MHz

GIEL Single Rank
Physics = 15 915
Combined = 6 537

Beast Dual Rank
Physics = 15 981
Combined = 6 506

2133MHz

GIEL Single Rank
Physics = 16 038
Combined = 6 506

Beast Dual Rank
Physics = 15 983
Combined = 6 546

2400MHz

GIEL Single Rank
Physics = 16 046
Combined = 6 446

Cinebench CPU Test

1866MHz

GIEL Single Rank = 1151 cb
Beast Dual Rank = 1138 cb

2133MHz

GIEL Single Rank = 1148 cb
Beast Dual Rank = 1150 cb

2400MHz

GIEL Single Rank = 1155 cb

Crysis 3

Settings ; 1440p - Very High - FXAA.
Location ; Welcome To The Jungle - Play Yard Minefield.

1866MHz

GIEL Single Rank = 83 FPS
Beast Dual Rank = 84 FPS

2133MHz

GIEL Single Rank = 84 FPS
Beast Dual Rank = 84 FPS

2400MHz

GIEL Single Rank = 86 FPS

Hitman Absolution

Settings ; 1440p - Ultra - FXAA
Location ; Benchmark

1866MHz

GIEL Single Rank = 66 FPS
Beast Dual Rank = 66 FPS

2133MHz

GIEL Single Rank = 68 FPS
Beast Dual Rank = 68 FPS

2400MHz

GIEL Single Rank = 70 FPS

Mad Max

Settings ; 1440p - Highest Settings.
Location ; Rot N Rusties - Enemy Camp (South).

1866MHz

GIEL Single Rank = 74 AND 72
Beast Dual Rank = 74 AND 72

2133MHz

GIEL Single Rank = 74 AND 72
Beast Dual Rank = 74 AND 72

2400MHz

GIEL Single Rank = 74 AND 72

Tomb Raider

Settings ; 1440p - Ultimate - FXAA - All Settings ON - NO SMAA.
Location ; Shanty Town Overlooking From Heli Pad.

1866MHz

GIEL Single Rank = 69 FPS
Beast Dual Rank = 71 FPS

2133MHz

GIEL Single Rank = 73 FPS
Beast Dual Rank = 73 FPS

2400MHz

GIEL Single Rank = 75 FPS

To conclude, the IMC was not strong enough to run the memory at 2400MHz on the dual rank DIMM's due to the increased density. Dual rank is quicker in synthetic memory benchmarls. As for gaming, the single rank DIMMs pretty much match the speed of the dual rank DIMMs. In my case, the single rank 2400MHz kit is the obvious choice as it brings the highest minimum fps in games whilst remaining stable. I'm sure if my IMC could have handled the dual rank DIMM's, we would have seen them match or slightly surpass the single rank DIMM's in games at 2400MHz. However, the dual rank DIMM's are at least 1.5x the price of the single rank DIMM's. The Kingston Hyper-X kit was £189.99. The price does not justify the results when compared to the very quick and solid GEIL kit.

Evidence Content (Screenshot Form)

GIEL EVO POTENZA MEMORY

Spoiler: Warning: Spoiler!

SPEED - 1866MHz














SPEED = 2133MHz














SPEED = 2400MHz

KINGSTON HYPER-X BEAST MEMORY

Spoiler: Warning: Spoiler!

SPEED = 1866MHz














SPEED = 2133MHz

GAME SETTINGS:

Spoiler: Warning: Spoiler!

TEST EVIDENCE IN DOCUMENTATION FORM - DOWNLOAD LINK BELOW:

https://mega.nz/#!MZkHURKY!rPuHceOHp3rfCNxrJQN7HvkkJ7JawG6QnyHKq2skJdE

 

[nleksan]

Pretty consistent with the results I obtained. More interesting is the minimum framerate variations.

 

[BradleyW]

Quote:

Originally Posted by nleksan

Pretty consistent with the results I obtained. More interesting is the minimum framerate variations.

I also tested using HT and found that HT increased the fps more, when the RAM was slower. But HT did not increase fps when the RAM was running at the higher speeds such as 2133 and 2400MHz.

 

[BradleyW]

OP updated.

 

[mrawesome421]

Interesting. And good to know. +Rep and thank you.

 

[Anti!!]

Sweet to know!

 

[iRUSH]

Very good!

 

[BradleyW]

Thanks.

 

[Bit_reaper]

That's actually a far bigger difference then I would have expected coming in at as much as 10FPS between 1333 and 2400 which considering we are talking 60-70 is pretty substantial. From an bang for buck stand point it looks like it makes sense to try and go for 1600MHz as the 1333 to 1600 is the biggest jump.

You should do additional tests about latency vs MHz. Like is it better to run 1333Mhz CL8 or 1600Mhz CL10 or going really loos like1866MHz CL12. It would probably interest thous who are running less then awesome ram.

 

[frickfrock999]

Awesome job, Brad. +rep.

 

[axizor]

Thank you so much for these results. I'm in no rush to buy 2400 MHz+ 16gb kits. Then again, I'm still only at 1080p, which seems to put me in the minority around here, lol.

 

[BradleyW]

Thank you for the feedback everyone!
Quote:

Originally Posted by Bit_reaper

That's actually a far bigger difference then I would have expected coming in at as much as 10FPS between 1333 and 2400 which considering we are talking 60-70 is pretty substantial. From an bang for buck stand point it looks like it makes sense to try and go for 1600MHz as the 1333 to 1600 is the biggest jump.

You should do additional tests about latency vs MHz. Like is it better to run 1333Mhz CL8 or 1600Mhz CL10 or going really loos like1866MHz CL12. It would probably interest thous who are running less then awesome ram.

Not sure how I would test the latency, but one thing I can tell you is that games are just as smooth between 1600MHz and 2400MHz, expect of course when the added fps helped in CPU intensive areas. I did not see or feel a difference during "normal gameplay" That was my interpretation of it any way.

Quote:

Originally Posted by axizor

Thank you so much for these results. I'm in no rush to buy 2400 MHz+ 16gb kits. Then again, I'm still only at 1080p, which seems to put me in the minority around here, lol.

All tests were conducted at 1080p.

Also, I'm lucky my chip can handle 2400MHz. Most are limited to 1866 on SB-E, some are limited to 2133, I'm limited a little higher than that. 2666 is a no go on my system.

If you all found this study interesting, please check out my other investigations. See sig. Thank you very much.

 

[BradleyW]

@Bit_reaper

oh I see. I did do a mini test on timings but did not report it because I only tested timings on Tomb Raider. Here we go, as always the min recorded fps:

Tomb Raider: Helipad overlooking Shanty Down during sunrise.
Ultimate Preset, FXAA, 1080p:

1600MHz CL10 = 63fps
1600MHz CL9 = 63fps
1600MHz CL8 = 64fps

Then I tested my G-Skill Ripjaws-X rated CL9 1600MHz. Here was the result in Tomb Raider:

G-Skill 4x4GB 1600MHz CL9 = 65fps. (1fps higher than the GEil's)

Maybe the sub timings on the G-Skill allowed increased performance over the GEil's at 1600MHz?

 

[Bit_reaper]

Quote:

Originally Posted by BradleyW

@Bit_reaper

oh I see. I did do a mini test on timings but did not report it because I only tested timings on Tomb Raider. Here we go, as always the min recorded fps:

Tomb Raider: Helipad overlooking Shanty Down during sunrise.
Ultimate Preset, FXAA, 1080p:

1600MHz CL10 = 63fps
1600MHz CL9 = 63fps
1600MHz CL8 = 64fps

Then I tested my G-Skill Ripjaws-X rated CL9 1600MHz. Here was the result in Tomb Raider:

G-Skill 4x4GB 1600MHz CL9 = 65fps. (1fps higher than the GEil's)

Maybe the sub timings on the G-Skill allowed increased performance over the GEil's at 1600MHz?

Hard to say. Did you punch in all the timings by hand? If not it might be that the GEil's where something like 9-10-9-27 while the G-skills where 9-9-9-24 for example. But it might also just be rounding error. Unless you do multiple run and use the average you pretty much have to assume at least +-1FPS variation.

 

[BradleyW]

Quote:

Originally Posted by Bit_reaper

Hard to say. Did you punch in all the timings by hand? If not it might be that the GEil's where something like 9-10-9-27 while the G-skills where 9-9-9-24 for example. But it might also just be rounding error. Unless you do multiple run and use the average you pretty much have to assume at least +-1FPS variation.

Yeah I always did 2 runs and got the same results. I think the GEil's might have been at mixed timings. I will investigate further.
Hey, at least we've found that fast ram...and even timings help in "CPU bottlenecked / Intensive areas".

 

[Bit_reaper]

Quote:

Originally Posted by BradleyW

Yeah I always did 2 runs and got the same results. I think the GEil's might have been at mixed timings. I will investigate further.
Hey, at least we've found that fast ram...and even timings help in "CPU bottlenecked / Intensive areas".

Well at least when running in a very high end system. If it wasn't so late and I wasn't so lazy I do some testing of my own and see how things stack up on an slower machine though my ram does not really want to run at more then 1700MHz CL10

Also the channel count could play a role as the LGA2011 is quad channel opposed to the dual found on AM3+ and LGA1550 or the triple in the LGA1366.

 

[BradleyW]

Yep, lot's of variables at play. But one thing's for sure, CPU intensive areas benefit from fast RAM + decent timings on a....let's say a high end system for now. During the CPU intensive parts of the games, my CPU usage was pegged at 100% and the GPU usage reduced slightly! The fast RAM seemed to help the fps when the CPU had it's hands full.

This is all down to pushy producers (EA), crap API's (Direct-X) and lazy programmers! At least devs and producers have addressed poor CPU usage by adding optimization. However, the overhead remains on most titles. Very frustrating for multi GPU owners.

 

[mrawesome421]

Quote:

Originally Posted by axizor

Thank you so much for these results. I'm in no rush to buy 2400 MHz+ 16gb kits. Then again, I'm still only at 1080p, which seems to put me in the minority around here, lol.

Your not alone bro.

1080p and 1333 RAM here

 

[BradleyW]

I actually love 1080p. It is sharp. even on my 42" screen and FXAA works wonders for bad aliasing on certain games. All my games look great on the PC overall.

I can't see a difference on 1440p or these down sampled images I've seen lately. When you blow them up to their proper size, they look just like 1080p, maybe because I am viewing them in 1080p. Not bashing high res, I'm sure many people enjoy it.

Edit: I might try testing Sleeping Dogs. That can be CPU heavy in some places.

 

[axizor]

Quote:

Originally Posted by BradleyW

Thank you for the feedback everyone!
Not sure how I would test the latency, but one thing I can tell you is that games are just as smooth between 1600MHz and 2400MHz, expect of course when the added fps helped in CPU intensive areas. I did not see or feel a difference during "normal gameplay" That was my interpretation of it any way.

All tests were conducted at 1080p.

Also, I'm lucky my chip can handle 2400MHz. Most are limited to 1866 on SB-E, some are limited to 2133, I'm limited a little higher than that. 2666 is a no go on my system.

If you all found this study interesting, please check out my other investigations. See sig. Thank you very much.

I believe your GPU driver uninstall guides were the first guides I've ever looked at on this site, and have referred to them countless times since joining.

(I would usually print out the first post in the thread then place it next to me when I went offline to reinstall/uninstall drivers)

Your excellent work here does not go unnoticed.