Samsung 30NM 4x4GB Ceiling @ 1866 CL8

Joined
Dec 22, 2007
Messages
987
I recently picked up two sets (16GB) of the Samsung Wonder RAM.

I've managed to OC to 1866 8-9-9-24 1T @ 1.5V (rig details in sig) and run Prime 95 and Memtest for 24 hours each without issue.

I can't even touch 2133 at a worthwhile latency. I was @2133 CL/CAS 10-12-12-26 2T stable but in comparison to 1866 CL 8 I was of course getting more bandwidth but at the cost of higher latency.

I have seen many reports of people getting 2133 CL 9 and even 8. I'm going to assume that these users had only 8GB and that my instability at those speeds is due to having all 4 slots full. Is this an accurate assumption or did I get sets of underperforming sticks?


I'd prefer to stay @ 1.5V max, for gaming would I bet better @ 1866 CL8 or 2133 CL10?


I've been referencing this chart made by cyclone3d.

http://hardforum.com/showpost.php?p=1040182029&postcount=1

Code:
					RAM SPEED			
		1066	1333	1600	1866	2000	2133	2400
CAS	6	5.629	4.501	3.750	3.215	3.000	2.813	2.500
	7	6.567	5.251	4.375	3.751	3.500	3.281	2.917
	8	7.505	6.002	5.000	[COLOR="Yellow"]4.287[/COLOR]	4.000	3.751	3.333
	9	8.443	6.752	5.625	4.823	4.500	4.219	3.750
	10	9.381	7.502	6.250	5.359	5.000	4.688	4.167
	11	10.319	8.252	6.875	5.895	5.500	5.157	4.583
	12	11.257	9.002	7.500	6.431	6.000	5.626	5.000
	13	12.195	9.752	8.125	6.967	6.500	6.095	5.417
 
Well, if you are artificially imposing limits of 1.5v you of course will not get any faster.

For the same reason I'm keeping my CPU under 1.4V I had hoped to keep my VRAM under 1.5V

Just about everything I've read suggests staying below these voltages for CPU and memory respectively on SB platforms in order to ensure long-term stability and longevity.

I'm also worried about heat with the dimms. I have not active cooling on them and as they are ULP sticks they have no heatsinks of any type.

To be honest I'd love to hit 5.0+ and 2133 @ CL 8/9 but going higher on the voltage makes me nervous. This build needs to last me until 2015
 
Yeah, 1.5v limit for the samsung sticks is kind of silly. When I was running them at ~2300, I am pretty sure I was pushing ~1.6 volts through them with the default timings.

The timings table for those Samsung sticks is quite different from other RAM that I have tested with. Some of the timings are a lot looser, and some are tighter when comparing to other sticks running at the same speed.

You will want to verify in the Intel specification documents, but I am pretty sure that as long as you keep the IMC voltage within 0.5v of the RAM voltage, you should be fine.

As for that chart, it is only for latency, and even at that, it is not a hard set list.

You also have to take into account the sub-timings. Just setting the main timings is not going to get you the highest throughput and lowest latency.

From my testing, Sisoft Sandra seems to give the most reliable (repeatable) benches so it is easier to tell what effect the settings have.
 
1.5v is the default voltage of DDR3.

Essentially you are running stock voltages, there is no way that will harm your processor. You will be fine at 1.6v. Above 1.65v is when I would start getting concerned about long term stability.

Also, those DIMM's don't even get warm to the touch at 1.65v, which is what I run mine at.
 
Yeah, 1.5v limit for the samsung sticks is kind of silly. When I was running them at ~2300, I am pretty sure I was pushing ~1.6 volts through them with the default timings.

The timings table for those Samsung sticks is quite different from other RAM that I have tested with. Some of the timings are a lot looser, and some are tighter when comparing to other sticks running at the same speed.

You will want to verify in the Intel specification documents, but I am pretty sure that as long as you keep the IMC voltage within 0.5v of the RAM voltage, you should be fine.

As for that chart, it is only for latency, and even at that, it is not a hard set list.

You also have to take into account the sub-timings. Just setting the main timings is not going to get you the highest throughput and lowest latency.

From my testing, Sisoft Sandra seems to give the most reliable (repeatable) benches so it is easier to tell what effect the settings have.

Good info, thanks. I read the exact same info about the relationship between IMC and RAM voltage.

I have never messed with the sub-timings before, guess I've got some reading to do.

1.5v is the default voltage of DDR3.

Essentially you are running stock voltages, there is no way that will harm your processor. You will be fine at 1.6v. Above 1.65v is when I would start getting concerned about long term stability.

Also, those DIMM's don't even get warm to the touch at 1.65v, which is what I run mine at.

Mine were certainly warm to the touch @ 1.5 but I only checked when Memtest was running which is probably why, as that test is more demanding than just about anything you could do throughout normal operation.

I suppose I could bump to 1.55 and shoot for 2133 CL8/9.

As far as CPU temps go, Max after 24 hours of Prime95 was 61C on 1 core, others were 55-58C. I could probably do 5.0 @ 1.4 but with LLC set to extreme things could get out of hand.

2133 CL 8-ish & 5.0+ GHz would definitely make me smile.
 
Found more good (old) info from Gibbo at overclock.net.

The numbers he provided are fairly safe though I know many people who run outside the VCore & VRAM suggested below.

I'll be playing around with VCore as I'm @ 1.37 but under load can hit 1.4V though I'm not sure how accurate CPUZ and HW Monitor are.

Sandybridge maximum safe voltages

Core Voltage - Not recommended too exceed 1.38v, doing so could kill the CPU, we therefor recommend a range of 1.325-1.350v if overclocking.
Memory Voltage - Intel recommend 1.50v plus/minus 5% which means upto 1.58v is the safe recommended limit. In our testing we have found 1.65v has caused no issues.
BCLK Base Clock - This is strictly a NO, anyone using base clock overclocking could/will cause damange to CPU/Mainboard. (Set manually to 100)
PLL Voltage - Do not exceed 1.9v!!

Processor - Basically we recommend customers not to exceed 1.35v to play it safe, all our bundles are set at 1.3250v or lower, any competitors offering bundles above 4.6GHz you should be enquiring as to what voltage they are using as we believe anything over 1.38v will limit CPU lifespan and anything over 1.42v will likely kill the CPU or severely limit its lifespan.

Memory - Intel recommend 1.50v plus/minus 5% which means 1.60v is the ideal safe maximum, but we have found in our testing all 1.65v memory is fine. We have also found most new 1.65v like Corsair XMS3 will run at its rated timings with just 1.50-1.55v which is well within Intel specifications. So people upgrading to Sandybridge you can still use your old DDR3, but we do recommend you run it at 1.60v or less. We are shipping most of our bundles which feature Corsair XMS at 1.50v-1.55v at rated timings. We've also discussed with Asus and MSI regarding voltages for memory and they also confirm in their testing 1.65v caused no issues with reliability.

Base Clock - To put it simple if you value the life of your components, do not overclock using base clock!

PLL Voltage - Again do not exceed 1.9v!
 
I need to get back to testing mine again, I've always had problems getting it to OC and be stable, though that could be BIOS to for all I know. Guess I should see how far out mine is, though I hate to screw with something that's working for the most part.
 
Updated my EFI, the version I was using was about a year old. Looks like it helped, I'm about to do 2000 10-10-10 easily now, I had to fight to get 1866 previously. Only done 4 or 5 hours of memtest so far, but 2133 11-11-11 would fail quickly.
 
Updated my EFI, the version I was using was about a year old. Looks like it helped, I'm about to do 2000 10-10-10 easily now, I had to fight to get 1866 previously. Only done 4 or 5 hours of memtest so far, but 2133 11-11-11 would fail quickly.

I just updated last night from original release bios to the newest version and was able to do 2133 stable @ CL 9. I didn't have time to run extensive testing but I'll be doing so tonight.

there were 4 or so bios revisions which noted increased memory stability so I'm optimistic.

I've got a 4770k & MSI GD-65 I'm sitting on until Wednesday when my current setup is being sold to a friend so I'll be starting all over.

I'm really hoping I can get 4.5GHz stable on the new chip with 2200/2400 stable RAM though I'm doubting the 4x4GB Samsung 30nm will attain this stable.
 
I run mine at 1600 8-8-8-24 default voltage 1.3-something.
Are you noticing a performance difference at 2133 or just doing this for fun?
 
I run mine at 1600 8-8-8-24 default voltage 1.3-something.
Are you noticing a performance difference at 2133 or just doing this for fun?

Am I noticing a difference? No, I don't have a GPU at the moment so I can't run anything intensive to gauge the seat of the pants feel.

However, here's some quick-n-dirty benchmarks I've run with Maxxmem at multiple settings.

Please note these tests were run with 4x4GB which makes attaining faster speeds more difficult than in a 2x4GB configuration as stability is reduced.

I would love to maintain CL9 @ 2133 but I bumped the voltage to 1.575 just for testing purposes. To be honest, my current config is already sold and being delivered on Wednesday so I'm probably not going to bother tweaking any more.

Album

1600 CL8
vwW0Alo.png


1866 CL8
658ERg2.png


2133 CL11
NVTek8Z.png


2133 CL10
N9lLpLn.png


2133 CL9
wQ9T39O.png
 
Last edited:
2133 CL9 and 1T with 4 slots?! That was darn good memory, I'm glad I got a bunch because M379B5273DH0-YK0 is no longer available.
 
2133 CL9 and 1T with 4 slots?! That was darn good memory, I'm glad I got a bunch because M379B5273DH0-YK0 is no longer available.

Yes sir it is!

I'm currently running 4x4GB @ 2133 CL9-10-11-24 1T @ 1.565V

I'll be setting up my 4770K & Z87 board this weekend. Depending on how my CPU OC is affected by my RAM OC I'll be at 1866/2133/2400.

I'll run the same tests on the new board @ 2400 CL 10/11 and post them here. I doubt CL10 is going to happen but it would certainly be nice.
 
Yes sir it is!

I'm currently running 4x4GB @ 2133 CL9-10-11-24 1T @ 1.565V

I'll be setting up my 4770K & Z87 board this weekend. Depending on how my CPU OC is affected by my RAM OC I'll be at 1866/2133/2400.

I'll run the same tests on the new board @ 2400 CL 10/11 and post them here. I doubt CL10 is going to happen but it would certainly be nice.

I'd be interested to see if the new HD 4600 GPU can benefit from faster memory. Supposedly it's 50% faster than the previous HD4000.
 
I'd be interested to see if the new HD 4600 GPU can benefit from faster memory. Supposedly it's 50% faster than the previous HD4000.

You will certainly find out!

I don't have a GPU so I'll be OCing the CPU/RAM this weekend and trying to play some games on the integrated 4600.

HD3000 on my 2500k failed to launch BF4 but I was able to run Dead Space 3 @ 1024x768 low with about 15FPS, lol.

This doesn't bode well though:
abGhtfA.jpg
 
Correction, I'm running mine at 8-8-8-22, I believe that's the sweet spot. The timings on the back of the package are the best guideline for this part. I used memtest for at least 48 hours to find the upper limit. It'd be interesting to see your GPU benchmarks stock 1600 vs. 2133
 
Yes sir it is!

I'm currently running 4x4GB @ 2133 CL9-10-11-24 1T @ 1.565V

I'll be setting up my 4770K & Z87 board this weekend. Depending on how my CPU OC is affected by my RAM OC I'll be at 1866/2133/2400.

I'll run the same tests on the new board @ 2400 CL 10/11 and post them here. I doubt CL10 is going to happen but it would certainly be nice.

Let us know how this goes. I have had about the same success with your timings at 2133 CL9 @ 1.56V. Thinking of pushing for 2400 CL 10.
 
Last edited:
Back
Top