Can I Get Some Help Overclocking My RAM?

[jordan1794]

Well I can hit 4.9 and be stable (on Prime95) with 1.3 vCore so I *think* I have a rather good chip.
However at 4.9 & 1.3 vcore my temperatures on the hottest core are around 82-85, so I run 4.8 @ 1.265 for my 24/7.

Anyone know of a good (free) memory benchmarker? I've been using MaxxMem but I don't know if it's a good one or not.

I guess I'll use IBT to test, and do 2 cores and probably about half my RAM
(I usually have 10-14 GB free of my 16 GB, so 8 GB should stress it without giving me an issue with running out of RAM)

 

[MxPhenom 216]

Well I can hit 4.9 and be stable (on Prime95) with 1.3 vCore so I *think* I have a rather good chip.
However at 4.9 & 1.3 vcore my temperatures on the hottest core are around 82-85, so I run 4.8 @ 1.265 for my 24/7.

Anyone know of a good (free) memory benchmarker? I've been using MaxxMem but I don't know if it's a good one or not.

Just play games or whatever you use your system for.

 

[jordan1794]

Well I'm doing this overclock for two things...
1.Foremost, to learn...overclocking fascinates me and through this I have already learned how to troubleshoot computers far and beyond what my knowledge was prior
2.I want to have my desktop and windows load faster, which the RAM speed has a direct affect on.

I'm not actually worried about my FPS, my GTX 980 does everything I want it to and runs nearly all new games on max settings
(GTA 5 I WOULD be able to...but I very quickly run out of VRAM)

I know these are small gains, that's why I'm not too *upset* if I can't get an overclock, but I want to learn the process so that I know how to do it, why certain settings fail, etc.

 

[MxPhenom 216]

Well I'm doing this overclock for two things...
1.Foremost, to learn...overclocking fascinates me and through this I have already learned how to troubleshoot computers far and beyond what my knowledge was prior
2.I want to have my desktop and windows load faster, which the RAM speed has a direct affect on.

I'm not actually worried about my FPS, my GTX 980 does everything I want it to and runs nearly all new games on max settings
(GTA 5 I WOULD be able to...but I very quickly run out of VRAM)

I know these are small gains, that's why I'm not too *upset* if I can't get an overclock, but I want to learn the process so that I know how to do it, why certain settings fail, etc.

Not really honestly. If you dont have Windows on an SSD, then you are doing it wrong.

 

[jordan1794]

Not really honestly. If you dont have Windows on an SSD, then you are doing it wrong.

I observed it myself

Normally after I enter my password it takes around 2-3 seconds for my desktop icons to appear, and if I attempted to open a window(like google chrome) it would take another 2 seconds or so to do so
When I had my RAM at 2600 they popped up instantly, and there was no lag when I tried to open a window.
I also have an issue where, during the first 30 seconds of a boot, any text that I type would show up on my screen 5-10 seconds AFTER I typed it...running the RAM at 2600 also stopped this from happening.

I currently don't have the luxury of a SSD, but I have a hybrid drive so my boot times are similar, but not quite as fast, as an SSD.

(I go from power off to login screen in around 8-10 seconds)

 

[jordan1794]

So do you guys think it is ok to use 1 GB of RAM to test each change, then if I pass move up to the next speed...but then if I fail to fall back to the previous settings and use 2 GB and then rinse and repeat until I hit 8 GB?
I don't have a lot of free time to wait 1/2 an hour for each test to run every time I change one of my timings.

 

[spaceprowler]

Seriously, OC your ram has no real benefit, your asking for instability and in real world usage you pretty much gain NOTHING !
Even if you perceive icons and what not is loading faster, not really worth it, run them at rated speed and timings ( lower timings are
actually better anyway when using DDR3 or 4 ) !!! In fact, you can even read reviews that DDR4 is really not much faster in
REAL world performance, just a few benchmarks ! OC'ing CPU & GPU = Major difference, RAM = HEADACHE ! But have fun !

 

[jordan1794]

Ok so I haven't been able to test much, but there are promising results!
I have passed IBT with 1 GB, and 2 GB of RAM, with 2133 @ 10-13-12-31
I also passed with 1 GB at 2400 MHz @ 13-15-15-35
I have not yet tried to tighten timings at 2400 MHz, I just made an educated guess at what would SURELY boot
Despite not tightening timings, I saw a 10% increase in my Gflops speed with the 2400 MHz over the 2133.
So I now definitely have a way of checking my RAM stability!
Thanks everyone! Lots of tweaking ahead of me now

 

[jordan1794]

Alright so I'm now at 2400 MHz @ 11-13-14-33
I have another question though...
My friend told me Ares RAM is the lowest quality G-Skill RAM (Not that it's bad, but that everything else is better)

Is this true? If so...I guess I am incredibly lucky that mine can hit 2400...

 

[jordan1794]

Ok so my final settings are 2400 MHz @ 10-12-13-32
If I get any lower I won't bother updating this thread though.

Thanks again for all the help! IBT with 2 cores and 8 GB is stable, and so are all my games.

 

[Henkenator68NL]

Above is xmp


OC .....
First start testing efficiency of XMP settings. .
how?
Search for the " share your aida64 Memory setting here" thread on tpu
1 download trial of aida64 and run cache memory test. Gives great insight in actual read, write , copy and latency of your dram.

2e use tool like MemtweakIt or something similar
3e use Maxxmem2 (download at https://hwbot.org

Note al results = screenshot

Run for example Intel XTU (tuning util) Save result (very depending on dram efficiency ).

After doing all that....

First lock all timings in bios (2e 3e and miscellaneous ).
Tighten latency boundary at first!

Rerun all mentioned tests... compare

Soon you will realize there is more gain there than in oc-ING your dram....

Below is optimized xmp

 

[Henkenator68NL]

Alright so I'm now at 2400 MHz @ 11-13-14-33
I have another question though...
My friend told me Ares RAM is the lowest quality G-Skill RAM (Not that it's bad, but that everything else is better)

Is this true? If so...I guess I am incredibly lucky that mine can hit 2400...

Well your friends is a fortune teller I take it...

How good a dram sick really is can only by figured out by testing.

But the main thing - for dram- is that is not about the sticks. .. it's about those black things on them: called IC's .

There are only a few IC manufacturers Like Samsung, Hynix, PSC, jMicron etc.

So it depends on the quality of those ic's.
Manufacturers use proces of binning:
They rigorously test a batch of IC's. The best ones go to more expensive kits ,binning = time consuming = expensive.

So in general cheaper sticks will not perform like expensive ones. But like with cpu's this is a silicon lottery to.

hope this gives some back ground information

 

[jordan1794]

Above is xmp


OC .....
First start testing efficiency of XMP settings. .
how?
Search for the " share your aida64 Memory setting here" thread on tpu
1 download trial of aida64 and run cache memory test. Gives great insight in actual read, write , copy and latency of your dram.

2e use tool like MemtweakIt or something similar
3e use Maxxmem2 (download at https://hwbot.org

Note al results = screenshot

Run for example Intel XTU (tuning util) Save result (very depending on dram efficiency ).

After doing all that....

First lock all timings in bios (2e 3e and miscellaneous ).
Tighten latency boundary at first!

Rerun all mentioned tests... compare

Soon you will realize there is more gain there than in oc-ING your dram....

Below is optimized xmp

I used MaxxMem to test, and my current settings have my latency down to 46.9, at the stock XMP it was 52.6.
I'll have to download AIDA64 though, and test with that to see what my gains are.

I think 2400 will be best however, because I CANNOT get my CAS latency lower than 10, even at 2133...
And my tRCD (second timing) and tRP(third timing) I can't get below 11,so I think 2400 MHz at 10-12-13-32 SHOULD be faster than 2133 @ 10-11-11-32
But only testing can tell. I already know my latency at 2133 @ 10-11-11-32 is 49.23, so the 2400 MHz is far better in that aspect. Guess I should check the other speeds to ensure complete coverage.

EDIT:
So AIDA64 says my latency is 43.2
Guess I'll revert to my xmp and test next.
Also, AIDA64 won't let me view my write or copy speeds, it says "TRIAL VERSION"
:/

Guess I'll only know the difference in read speed.

 

[Henkenator68NL]

You will be suprised to hear... the biggest wins are actually in the 3th timings !! (partially covered by setting the latency boundary) but manually you should key in the auto values. after testing all is oke you can try to lower
the Write to Write delay tWWDD and tWWDR,
the Read to Write tRWDD and tWWRD
Write to Read tWRDR and tWRDD,
and also Read to Read tRRDR and tRRDD

names might be slightly different in your bios

 

[jordan1794]

So here are my comparisons between the stock 2133 XMP and my 2400 overclock:

Read:+2,893
Write:+3,980
Copy:+3,387
Latency:-5 ns

I think that's a pretty good increase...around 10%+ better on each reading.

 

[Henkenator68NL]

If I am correct you still are on command Rate 2T ???? switch that to 1T and lower the tRAS to 26, that should work without problem! than see the test result once more

 

[jordan1794]

If I am correct you still are on command Rate 2T ???? switch that to 1T and lower the tRAS to 26, that should work without problem! than see the test result once more

Well I read that AMD LOVES 1T, but sometime Intel chips can hate it...
I wanted to get to a final overclock for 2T before I started working on 1T, but I guess I'm pretty much there.

I'll start messing around with 1T later tonight maybe, depends on if my wife kills me or not for spending so much time overclocking.

 

[cadaveca]

Might wanna consider a vCache and/or VCCSA(System Agent) boost.

Juice that ram up, man, up to 1.85V should be fine.

 

[jordan1794]

Might wanna consider a vCache and/or VCCSA(System Agent) boost.

Juice that ram up, man, up to 1.85V should be fine.

Currently have my vring manually set to 1.050 (Was the stock setting, was preparing to increase it as I overclocked my uncore, but I did not need an increase to be stable at 4.4)
System agent offset is +0.150 (so it ends up being 1.2)
I'm about to jump into my BIOS, but my Gigabyte BIOS is drastically different from all the screenshots/terminology I find from googling so it takes me a while to decipher what setting does what xD

 

[cadaveca]

OK, so if VCCSA is 1.2V, I feel that is a bit too high.

vRing (cache) could be the same as vCore/vCPU, no problem, so you've got a fair bit of overhead there.

So, using AIDA as you have been, adjust each, one at a time, and then boot back into OS and check the cache bandwidth numbers. You'll find that caches bandwidth lowers in AIDA when stability is starting to go out the window. So, if an increase gets slightly better results, you are on the edge of stability.

 

[TheHunter]

If cpu reboots while stress testing it could also be ram power current too low.. I need 120% ram current @ 2400mhz+

VCCSA also higher, I found out 0.060v+ offset is enough @ 4.7ghz, 0.040v+ offset @ 4.6ghz..

Also vring (cache) affect ram OC capability, higher multi less chance to get higher ram OC at higher cpu OC.


Anyway, I bought 2133mhz with tight timings to OC to at least 2400mhz+, which is possible (saw one review it can do max up to 2690mhz CL11-13-13-35) but now I see there is very little difference, by gaming mostly none.

Well there is by lets say generic benchmarks 2133 vs 2400, e.g. Cinebench15 (5-7cb points) or 3dmark physics tests (100-300points) because of higher ram badwidth (32 vs 36gb/s), but overall tighter 2133 advanced timings 2nd and 3rd usually makeup for the bandwidth with lower latency. Actually its kind of a draw most of the time.


I did a lot of testing when I first got it and anything over 2600mhz already needs 2T no matter what, stock 2666mhz or OC'ed to 2666mhz and this raises latency further - also can sometimes affect cpu L2 latency a bit in Aida64.. Also any mobo so far sets 2666mhz+ too loose, unless you know how to tweak manually and even then its not 100% guarantied.

I posted this once @ another forum, might help you a bit..

Tests:

Spoiler

nvm different Aida64 versions and cpu OC's, ram bandwidth variation is minimal.


2133mhz CL9 - 1T


2400mhz CL10 - 1T



2400mhz - 2T



2400mhz - auto before asus fixed 2400mhz adv timings (early bios 1.5years ago)





2666mhz CL11 - 2T


2666mhz done right on some Asrock mobo (of course manually tweaked), but I couldn't do this, wish I could, but yeah its very rare..



These timings are the most crucial



for comparisons 2133mhz vs 2400mhz advanced latencies


by 2666mhz are a lot higher, its mostly tRRD, tRTR and tRTP that its the most crucial and also the most sensitive by overall stability

ideal are

4, 8, 6, but very tricky, 4, 8, 7 is more stable but already suffers by read speads, 4, 8, 8 is the most stable and also the slowest, anything higher suffers the most and by 2666mhz+ most are 5, 8, 8 or even 6, 8, 9.. => 26 - 28gb/s


check tRTR timings
2133


2400





e.g. geekbench3 memory bandwidth info, 2133 vs 2400 by one 4.5ghz by other 4.7ghz cpu OC

A quick gaming performance test is Resiedent Evil 5 benchmark - fixed benchmark (dx9, 720p, noaa, rest max) so its cpu bottlenecked can show good results too if any, mostly max (5-10% in benchmark) and min fps (40-50% in benchmark), avg depends up to 4-5fps.. But you will need to go to game cfg.ini file and manually change jobthread=8 instead of 4 so it also stresses cpu more..

Same thing in Lost Planet2 (same reso, settings, dx11 mode, dx11 - low; jobthread=8 in cfg.ini) - testB, its also fixed cpu test.