which SO-DIMM DDR3(L) for 2/4 bank RAM-upgrade i3/i5, i7 --- power vs. speed vs. size

I know some differences of the various iX CPU SO-DIMM RAM requirements, but can't find information on power consumption or performance benefit of the various options that helps to make up my mind which power,speed, number of SO-DIMM and size of the SO-DIMM to choose.
So is it better to buy 2 x 8GB or 4 x 4 GB for performance and lower power consumtion?

number of SO-DIMM
i7 adresses the RAM without a controller an can adress 2 banks of 2 SODIMM each, while i3/5 have only banks of 2 SO-DIMM each


power consumption:
Depends on the supply voltage the supply voltage (lower is somewhat better) and number of memory-chips, and their manufactering process = structure in nm.

Voltage DDR3(L):
Starting with the iX-4xxxx generation DDR3L RAM with 1.35V Voltage are supported, even though they can be mixed with 1.5V SO-DIMM.
I noticed one disadvantage in doing this in my zBook 15, which is the UEFI BIOS halts boot ,saying it detected unsuitable RAM. After accepting the use of the mixed 1.35V/1.5V SO-DIMM it continious boot and works fine

Here is where my problemg starts, no SO-DIMM specifications I looked at provides a max. power consumption. Of course it emphasizes, when it's DDR3L with 1.35V, that power consumption is lower then DDR3 with 1.5 V, but how much is left to our imagination.


Speed:
The higher the clock, the more shorter each clock cycle and therefore the more a RAM can achieve were it not for the Latency which defined the number of cycles it requires for access CL-tRCD-tRP and CAS ( DDR3 SDRAM - Wikipedia, the free encyclopedia)
Mushkin 4GB has CL11-11-11-11 while for others it's amuch higher CAS CL9-9-9-2x (CAS 24 to 28)

 

2 modules will consume less power than 4 modules (if speed & voltage are the same).

As for speed, some laptops will handle 4 modules with 2T command rate (slower) while using 1T command rate (faster) if only 2 modules are used.

As for RAM density: single-sided 4GB RAM module [with density of 4Gb (512MB) per die, total of 8 dies on module] will consume less power than double sided 4GB RAM module [with density of 2Gb (256MB) per die, total of 16 dies on module]. Modules with the same number of dies, running at the same voltage and speed, will consume the same power. 8GB module with 16 dies will consume the same amount of power as 4GB module with 16 dies (if speed & voltage are the same).

 

Well here's something to help confuse you a little more.

Kingston KVR16S11/8 DDR3 1.5V SO-DIMM 1600 CL11 - Maximum Operating Power 2.580 W
Kingston KVR16LS11/8 DDR3L 1.35V SO-DIMM 1600 CL11 - Maximum Operating Power (1.35V) = 2.721W

Do note the "depends on SDRAM type used"

 

It would be a very interesting read, since it contradicts some of the things I wrote in this thread, if not for the fact that both your links point to the same document...

EDIT:
Nevermind, I found the right PDF.
You are absolutely right - according to these specifications the 1.35V module can consume more power than the 1.5V module, when rated speed, latency, RAM density and capacity remain the same. True mystery...

PS
+rep added

 

Yes, not sure why 1.35V is higher in those specs. Still considering same modules may use SDRAM from different manufacturers perhaps the datasheets for the SDRAM would be more suitable, lot of work though. In my experience manufacturers often emphasize good points so if the modules are lower power than the rest it might get mentioned, good marketing. If it is not so good or worse quite poor, it might not get mentioned at all.

 

I was reading right yesterday some test runs between G.Skill 1.35v and 1.5v RAM on desktops.
The test conclusively showed lower power consumption FOR THE CPU when using lower-voltage ram, with savings between 4 and 13 watts under load.
Power consumption of the modules themselves was near identical, with the 1.35v even pushing further than the 1.5v ram in consumption in a few limited scenarios, but that increase was negligible compared to the CPU savings.

What this means for me, is that the CPU has to consume far more power when operating the memory controller at a higher voltage. 4 to 13W is pretty big, nowhere near negligible. Admittedly, these numbers may vary on a mobile CPU, but I don't see the mobile controller being much more efficient than the desktop one, really.

 

The power consumption between RAM modules are more or less negligible. And considering that document shows "maximum operating power" that's if it's fully loaded. At idle or low power usage is where it's more important figures, but the difference is even more miniscule. I don't have the time or energy at the moment to find the battery life tests between several different modules and the difference was pretty much minutes, probably within error of the test.

Here's some interesting articles:
http://www.xbitlabs.com/articles/memory/display/kingston-lovo-hyperx_3.html#sect1
http://www.pugetsystems.com/labs/articles/Technology-Primer-Low-Voltage-RAM-150/
http://www.tomshardware.com/reviews/lovo-ddr3-power,2650-7.html

And an earlier NBR discussion: http://forum.notebookreview.com/har...grades/684257-1-5v-1-35v-ram-does-matter.html

 

actually the ram in the ultrabook is the warmest component since no fan is spinning over it. Do you guys know which ddr3l 1.35 stays the coolest? Thanks

 

Thank you for the links provides above, but I wonder if SO-DIMMS are compareable to the standard sticks used in desktops?

Only when I knew that Kingston provided data sheets and then onlyforsome of their SO-DIMM, but not for the paired SO-DIMMs, did I find the ECC SO-DIMM http://www.kingston.com/dataSheets/KVR16LSE11_8.pdf ECC SO-DIMM
.
I find it strange that it requires much less
(1.35V) = 2.041 W*
(1.50V) = 2.268 W*

than the power the non ECC SO-DIMM (KVR16LS11/8) requires, whic is

(1.35V) = 2.721W*
one despite the additional chip.

KVR16LSE11/8
8GB 2Rx8 1G x 72-Bit PC3L-12800
CL11 204-Pin ECC SODIMM

KVR16LS11/8
8GB 2Rx8 1G x 64-Bit PC3L-12800
CL11 204-Pin SODIMM

If this values are representative forother brands too, then I wouldadd the additional question if ECC would work instead of a non ECC?

 

On the subject of speed, Haswell absolutely loves memory bandwidth (ie speed), and timings take a distant backseat.

With that being said, I find that stressing the memory controller too much may actually lead to lower CPU efficiency. I have a set of 2x4GB 2133 CL11 ram that can be run at 2400 CL11. Although memory benchmarks improve, CPU benchmarks (SuperPI, wPrime, CineBench, XTU) all show about a 5% decrease compared to running ram at 2133. Considering that Haswell only officially supports 1600 ram, I can't really complain.

 

Regardless of power haswell mobile requires 1.35v it will not deliver 1.5v.

 

The difference in power consumption is fairly simple to estimate. If we assume that the IC is close enough between the 1.35 and 1.5 volt varieties, then we can estimate it based on the fact that power consumption goes up by the square of the voltage.

1.5^2= 2.25

1.35^2=1.8225

Using DDR3 at 1.35 volts uses about 19% less electricity than 1.5 volts, all else equal. That isn't a whole lot, especially when you consider the low power consumption of RAM as a part of the total system power consumption.

 

@Meaker

You may be right for i3- and i5-Haswell which I have not checked, but my i7-4700MQ supports not only DDR3L or DDR3 but also mixed DDR3/DDR3L. In mixed mode both run 1.5 V, the voltage of the DDR3.

My zBook 15 runs with one 4 GB DDR3 from my Elitebook 8560p and the 4 GB DDR3L the zBook 15 came with. Unfortunately it won't boot unless I acknowledge the difference in RAM, but I wrote this before.

FOr the CPU you can easily check other CPU's

Intel Core i7-4600U Mobile processor - CL8064701477000 i7-4600U
Intel Core i7-4700MQ Mobile processor - CW8064701470702 i7-4700MQ

 

Are you sure in mixed mode the ram is running 1.5V? Haswell mobile chipset can only supply 1.35V, so if your 1.5V stick runs, it's because you got lucky with the silicon, and it's not something that you should take for granted.

 

No I'm not so shure any more, and so started searching some more without to much success. It depends on how HP implemented the voltage when the read the SO-DIMMs information.

CPU-Z identifies the RAM as 8GB DDR3 so I don't know what to think. CPU-z does not provide the voltage selected. Has anyone an idea to read the selected RAM Voltage?

I now took the time to look through Intel's Mobile 4th Gen Intel® Core™ Processor Family: Datasheet, Vol. 1 and it says DDR3L and DDR3-RS only, and have to appologize to meaker for not reading through Intelss documentation first and beieving in second source information.

SInce the DDR3 is only a few weeks old, I will open the zBook tomorrow and look at the DDR3 SODIMM chips and compare them to the chips on the DDR3L. I'll post it when I know more.

 

PS CPU-z SPD report is:

DDR3 is Hunday HMT351S6CFR8C-PB in Slot 2 Week/Year 10/14
• VDD=1.5V +/- 0.075V
•VDDQ=1.5V +/- 0.075V
•VDDSPD=3.0V to 3.6V
•Functionality and operations comply with the DDR3 SDRAM datasheet

DDR3L is Hunday HMT351S6CFR8A-PB in Slot 4 Week/Year 06/14
• Power Supply: VDD=1.35V (1.283V to 1.45V)
•VDDQ = 1.35V (1.283V to 1.45V)
•VDDSPD=3.0V to 3.6V
•Backward Compatible with 1.5V DDR3 Memory module

however not which voltage has been selected for theSO-DIMM

 

I don't repeat myself very often but I know this issue can be confusing.

Haswell desktop can deliver 1.35V and 1.5V

Mobile haswell can delivery 1.35V only, locked, no chance of changing this without a hardware mod kind of locked.

Supported memory: DDR3L-1333, DDR3L-1600, LPDDR3-1333, LPDDR3-1600 (All of this listed is 1.35V)

You can run 1.5V ram because you won the silicon roulette and your chips can run at stock at 1.35V, they are undervolted but still stable, this is not guaranteed as much as overclocking is not guaranteed.

You would need to run extensive stress testing to be sure it's 100% stable.

 

Are you saying CPU-Z is not reporting the voltage as highlighted in the red box below?

 

I think you will find most laptops use a fixed voltage regulator so you get that voltage regardless.

Not likely there is any ADC used in most laptops to be able to read RAM voltage. SPD just shows what voltage the SO-DIMM nominally operates at and there can be more than one voltage. What you actually get though is what the laptops hardware is set for, it's not CPU controlled. With Haswell laptops that is nominally 1.35V.

 

Intel has enforced 1.35v to help intel notebooks look better battery wise, however small the difference.

 

As has been noted many times, 1.35V RAM doesn't affect battery life significantly.

The Intel notebooks have superior battery run times because of the platform (cpu/chipset) not because of a nominal voltage drop.

 

@snn47, depending how far you want to go you might find this useful for calculating power draw.

Power Calc - Micron Technology, Inc.

 

Actually Tiller, I believed this at first but check out this article. Patriot Memory for Ultrabook PC3-12800 8GB LoVo SODIMM Review - Benchmarks - Test System Setup and Battery Life Test

A single 8gb DIMM at 1.35V at 1600mhz managed to eek an extra 6% battery runtime compared to a single 4gb DIMM 1.5v at 1600mhz. So it stands to reason that 2x4gb DIMMs at 1600mhz 1.35v should yield roughly 10-12% extra battery runtime compared to 2x 1.5V modules.
This might not sound like much but its a free 12% with minimal performance compromises (full 1600mhz JEDEC speeds). Laptop processors generally don't benefit from high speed RAM as much, unless there's significant iGPU gaming involved, as their desktop brethren.

Whether this extra 12% is worthwhile is dependent on the OP since it obviously has a greater impact when the laptop can routinely pull more than 6hrs as this could potentially mean an extra 43 minutes of activity.

For a Laptopm, I would be setting my expectations with the following mindsets.
1. Absolute performance: get the G.SKILL 2133mhz RAM, costly and hard to find but gives you the best of power efficiency and performance. Also OCs very well.
2. Price conscious: 1600mhz G.SKILL 1.35V, great power consumption and good enough performance. Good price too for the fact they are giving you RAM that is being manufactured on the most cutting edge node.
3. High Capacity: Generally will yield better overall utility than high speed, when the application calls, you will feel the lack of RAM more severely than a lack of memory bandwidth.

I don't generally like the 1866mhz kits since they are so close in price to the 2133mhz kits. Plus, even though their latency is lower, high bandwidth tends to be better utilized by any Intel CPU circa Sandybridge and later.

 

How do you figure it would double the %? Whether 1 stick vs 1 stick or 2 sticks vs 2 sticks, the % would remain the same, it's not additive. So 2x8GB @ 1.35v vs 2x8GB at 1.5V would still only result in 6% improvement.

Also that test has to be taken with a grain of salt. They are two different brands, two different size chips, and I'm sure Lenovo didn't buy according to best power saving, but on cost. Most 1.35V modules can run at 1.5V as well. If they could take the same module and run it at 1.5V then run it at 1.35V and check the battery life then. Or at least same brand 1.5v and then 1.35V. Plus I'd have to say battery run times, 2-3% difference is within margin of variance.

If anything that test should show that buying a quality name brand memory can improve performance reduce power consumption slightly.

 

Well I'm basing my estimate on 2 assumptions which I hope are quite conservative but feel free to correct me if my reasoning sounds wonky:
-1x8gb aftermarket stick managed to beat 1x4gb stick in power consumption by 6% despite having a double density of RAM (hence maybe a 75% increase in power consumption).
So I'm reasoning that 2x4gb sticks (using the same RAM as the 8gb stick) would yield only slightly more power consumption than this single 1x8gb stick at the same voltage.
Hence a 2x4gb 1.35V stick yields ~5% improvement over a single 1.5V stock 1600mhz 4gb Lenovo OEM stick.
Thus 2x4gb 1.35v sticks theoretically should yield 10% savings over 2x4gb 1.5V sticks.

-Any extra savings due to voltage or manufacturing improvements is gravy on top as this particular 1.35v module which was reviewed is quite old.

I agree that 2-3% is within margin of error but 6% with double density modules (hence at least 50-75% more power consumption) is harder to ignore and worth investigating.

 

Wouldn't double amount of RAM (8GB vs 4GB) significantly reduce the use of swap file, hence the power draw of the HDD, therefore reducing the overall power draw, despite double RAM density?

 

Having "double density" RAM would not increase power draw that significantly. There can be a 16 chip 2GB module and a 16 chip 8GB module, same number of chips, just higher density. In the same way that a 1TB SSD won't draw 4x the power of a 256GB SSD, the principle is the same. There may be some slight differences in power consumption but not that significant. The difference would still be the 6% they saw using 1 module or 2 modules.

 

+the technology e.g.a denser structure/size of memory cells, could reduce power consumption similar to reducing clock speed or voltage. But from what I found you can't easily put a percentage on it like x %.

But why lists Corsair ECC SODIMM RAM to consume less power then the standard SODIMM. As to ECC support, c't wrote for desktop CPU, that while ECC support is not enabled for most Haswell by Intel in the various Haswell CPU, that they work also with ECC RAM, just not giving the benfit of ECC.


I found the time to do more tests and checked the SPD of my SODIMM with CPUZ, therefore I now have to correct myself because I have a I7-4700MQ in my F0U59ET#ABD and not a i7-4600 as I remembered. Either I was looking looking at to many differnt notebooks before I ordered mine or I'm getting old..

Intel lists as required memory DDR3L-1333,1600 ARK | Intel® Core™ i7-4700MQ Processor (6M Cache, up to 3.40 GHz).
No support for 1.5 V SODIMM nor slower RAM, nor for any of the faster available DOSIMM like 1866mhz 2133mhz. It's doubtfull that faster clockable SODIMM will result in a faster Bus clock, even if we reprogramm the SPD, but who knows maybee HP would recognize/allow faster clocked SODIMM then 1600. For lack of such fast RAM I cannot try.

Performance should in my feeling nit increase at higher clock since this RAM normally has a higher latency when compared to lower clocked RAM, and changes in duration of clock cycle are now to small to offset having one or more additional cycles needed at a higher clock frequency before access is achieved.

I removed my two 8GB Mushkin PC3-8500 SODIMM 7-7-7-20 1.5V from my 8560 and inserted it into the bottom slot 1 and 2. During boot it stopped with the message wrong RAM again, after acknowledging the error it booted into w7, showing 12GB if I mixed a 4 and an 8 GB and 16 using the two 8 GB sticks.

Windows RAM performance index of 7.8 for both the RAM it came with and my old PC3-8500 1.5V RAM! CPU is 7.9, Graphics (game) 6.7 (6.7), and HDD 7.9. Doubling the memory might could be called a reason for the lower clock rate, but I would say it's the smaller latency, but again who knows what µ$ programmers smoked when they defined this performance index.

For the RAM slots CPUZ reported Bottom 2 as Slot 2 and Bottom 4 as Slot 1. Could someone please verify the Slot numbers as I got it reported?

Therefor instead of less I have additional confusion.

-CPUz reported all RAM to have the speed of the slower RAM, I did not expect that the performance index is the same despite the slower clock?

-Kingston HyperX Impact 16GB SO-DIMM Kit, DDR3L-2133, CL11?

HyperX Impact SO-DIMM Memory. Does anyone know how/where to check the max allowable RAM clock in the zBook?

-Is the memory voltage generated by the main board and has the BIOS therefor control of the voltage or the CPU?

The next time I'll use my power meter for power consumption.

 

How do you guys know the voltage the memory operates to?
Using a multimeter on the sodimm pins should give me the voltage (1.35V or 1.5V)?


Sent from my iPhone using Tapatalk

 

haswell mobile always provides 1.35v for instance.