Sign in with
Sign up | Sign in
Your question
Solved

Mixing RAM kits.

Last response: in Memory
Share
a b } Memory
September 1, 2012 1:07:38 PM

I made the not-so great decision to get high/med profile RAM when I initially built my system. While this isn't a problem now, I am sure if I had to upgrade to 16GB later, getting another pair of these would conflict with my Hyper 212 EVO cooler (or it would be a close thing). Just wondering if I used a low profile kit to fill the left-over slots, would there be any problems or downsides?

Existing RAM kit.
G.Skill Ripjaws 8GB (2x4GB) 1600mhz, CL9 1.5v (not overclocked at all, they wont budge)
http://www.newegg.com/Product/Product.aspx?Item=N82E168...

Prospective RAM kit to get.
G.Skill Ares 8GB (2x4GB), 1600Mhz CL9 1.5v.
http://www.newegg.com/Product/Product.aspx?Item=N82E168...

As far as I can tell, only difference is the heat-spreader and slightly different timings.
9-9-9-24-2N (Ripjaws)
9-9-9 (Ares)

Thanks for any input on this matter, I know a lot about computers, but RAM timings still evade me :lol: 

My rig as it matters.
i5-3570k
AsRock Z77 Extreme4
CM Hyper 212 EVO
That Ripjaws kit above.

More about : mixing ram kits

a c 347 } Memory
September 1, 2012 1:29:33 PM

Downside - tons of memory errors; Upside - alludes me. DO NOT MIX RAM!

If needed move-up the CM's fan or flop it over on the other side.

To get the RAM to work @ Rated:
BIOS:
[optional] Press F9 ; Load Optimal Defaults - if you other changes in the BIOS this will lose them.
DRAM Timing Configuration /Enter
Load XMP Setting -> Profile 1
Press F10 ; Save & Exit = Yes

Otherwise explain:
Quote:
(not overclocked at all, they wont budge)
m
0
l
a b } Memory
September 1, 2012 1:44:07 PM

The RAM kit is running at normal frequencies and settings, with no issues I know of.

I tried to overclock them before (not thoroughly, just to see if I could really) and they wouldn't budge from stock settings. Upping the frequency (to 1866Mhz) or lowering the latency (CL8) both failed to POST, with the motherboard speaker telling me there were RAM problems. Resetting the CMOS set them back to stock settings, and it works just fine. Didn't attempt to again after that.

In a nutshell, tried to overclock the RAM, didn't work, couldn't be bothered to try harder, put it back the way it was and let it be, all works.

m
0
l
Related resources
Can't find your answer ? Ask !
a b } Memory
September 1, 2012 1:51:25 PM

I wouldn't mix different sets of ram it will work but i wouldn't recommend it to many cons possible to risk it imo
m
0
l
a b } Memory
September 1, 2012 1:53:52 PM

I am pretty sure that ram doesn't overclock very well anyway it didn't for me either on a rog mobo i did get it stable 1800 MHz but it wasn't easy lol
m
0
l
a b } Memory
September 1, 2012 1:54:09 PM

I know RAM doesn't overclock all that well, and the benefits are small compared to the CPU or GPU.
I mainly tried it just to see if I could, and because all the reviews I'v read on my kit said they were good for it...

I did Frequency and Latency changes independent of each other. So 1600Mhz and CL8, or 1866Mhz at CL9. I know overclocking method, only touch one variable at a time. Both settings described failed to POST. Didnt mess with voltage, modern Intel processors shouldn't have RAM that uses more than 1.5v (or at least that's what Intel say)
http://www.intel.com/support/processors/corei5/sb/CS-03...
m
0
l
a c 347 } Memory
September 1, 2012 1:56:27 PM

You're kidding, Frequency OC: DDR3-1866 9-9-9-28-2N (1.55v~1.60v) or (10-10-10-30-2N) and you might need to up the VCCIO & VTT 1.15v~1.20v (1.20v is max safe); CAS OC DDR3-1600 8-8-8-24 (1.55v~1.60v) and ditto -> you might need to up the VCCIO & VTT 1.15v~1.20v (1.20v is max safe).

Neither kit you listed will OC DDR3-1866 CAS 8 - 'period' - well without the need of crazy voltages.
m
0
l
a b } Memory
September 1, 2012 1:58:22 PM

manofchalk said:
Just wondering if I used a low profile kit to fill the left-over slots, would there be any problems or downsides?

Some motherboard/CPU/DIMM combinations have a hard time working at 1600+MHz even when all DIMMs are identical. In those cases, reducing memory clock to 1333MHz has a high probability of fixing it.
m
0
l
a b } Memory
September 1, 2012 1:59:01 PM

most ram is tested to work as it spesification. So mix some rams fine when they work as the spec (default setting), mixing higher speed ram with slower it will just downclock to it.

But when we O.C. or runing dual/multi chanel it's a diffrent story. A match ram for each channel is required...

edit: my old rig has a diffrent set of ram, my laptop also using diffrent pair of ram (one that OEM, one i'm added later). still works fine..... :bounce: 
m
0
l
a c 347 } Memory
September 1, 2012 2:07:15 PM

manofchalk said:
I did Frequency and Latency changes independent of each other. So 1600Mhz and CL8, or 1866Mhz at CL9. I know overclocking method, only touch one variable at a time. Both settings described failed to POST.

Didnt mess with voltage, modern Intel processors shouldn't have RAM that uses more than 1.5v (or at least that's what Intel say)
http://www.intel.com/support/processors/corei5/sb/CS-03...

* If you simply set the DRAM Frequency and CAS Timings then no doubt that will fail. Any OC (exceeding Rated) will require additional voltage -- same with the CPU, RAM or GPU once the stability threshold is reached.

* Not true, Frequency & CAS Timings are very much tied together. See Chart.

* The damaging voltage is the CPU's Voltages (VCCIO & VTT). Exceeding 1.35v has been proven to degrade your CPU, and 1.35v or higher actually pretty quickly.

DRAM to CAS Chart (I know I need to update):
m
0
l
a b } Memory
September 1, 2012 2:21:45 PM

Yeah for oc ram my buddy has i am pretty sure not certain i think it is Mushkin Enhanced Redline holy crap oc like a beast when i was at his house one day he was showing me how impressive he could oc his ram lol
m
0
l
a b } Memory
September 1, 2012 2:26:07 PM

As I suggested somewhere above, didn't really research thoroughly into RAM overclocking before trying it. Was more just a "what-if I could" thing, so I tried. I'm not concerned if my RAM stays the way it was intended. It seems my clumsy attempt at it hasn't broken anything, so no harm done.

Anyway, from what I have gathered is that RAM overclocking is a lot harder than CPU/GPU, mixing RAM kits is bad idea and that I should just suck it up and buy a 16GB kit if/when I need 16GB (or change my cooling, get a slimline fan or something. Probably the simplest solution :lol: ).
m
0
l

Best solution

a c 347 } Memory
September 1, 2012 2:38:10 PM

The HSF Fan can butt up to the tops of the RAM and as I mentioned if need flip the fan to the other side (pull) or raise it a few mm's.

Try the RAM OC settings I posted, you'll probably have better luck. I wouldn't even think twice have a DRAM Voltage of 1.55v~1.60v with 1.20v or lower VCCIO/VTT.
Share
a b } Memory
September 1, 2012 2:48:03 PM

OK then, will try to overclock to your settings when I have an hour or so spare.
(its currently 12:45am here in Australia, just off optimal timing for computer tweaking IMO :lol: )

Just a question, will a higher frequency or lower timings give better performance? I realize the performance boost with OC'ing RAM in general is only small, but which would be better (have a feeling its the timings)?
m
0
l
a b } Memory
September 1, 2012 2:57:02 PM

Best answer selected by manofchalk.
m
0
l
a c 347 } Memory
September 1, 2012 3:00:19 PM

That's a loaded question, some benchmarks respond better to reduced latency while others increased frequency. Gaming is more affected by latency on the SB/IB and Multitasking frequency.

The largest impact is from CPU Clock, example CPU Clocks (Stock vs 4.8GHz) with DDR3-1600 CAS 9 (fixed):
m
0
l
a b } Memory
September 1, 2012 3:06:48 PM

manofchalk said:
Just a question, will a higher frequency or lower timings give better performance? I realize the performance boost with OC'ing RAM in general is only small, but which would be better (have a feeling its the timings)?

That depends on the specific benchmark/application.

Code that does streaming processing of large linear data sets benefits most from higher bandwidth (clock) since most of the RAM access time is linear burst transfer.

Code that does lots of conditional/non-linear processing such as compiling code or executing scripts benefit most from lower latency since a lot of RAM access requires closing/opening memory rows, switching memory banks, etc. before the first byte of newly requested data can be accessed.
m
0
l
a b } Memory
September 1, 2012 4:06:12 PM

manofchalk said:
As I suggested somewhere above, didn't really research thoroughly into RAM overclocking before trying it. Was more just a "what-if I could" thing, so I tried. I'm not concerned if my RAM stays the way it was intended. It seems my clumsy attempt at it hasn't broken anything, so no harm done.

Anyway, from what I have gathered is that RAM overclocking is a lot harder than CPU/GPU, mixing RAM kits is bad idea and that I should just suck it up and buy a 16GB kit if/when I need 16GB (or change my cooling, get a slimline fan or something. Probably the simplest solution :lol: ).
I know bud i read your post now that i really think about it was a pointless post on my part about the ocing ram thing ;) 
m
0
l
a b } Memory
September 2, 2012 12:56:19 AM

Now I am getting really confused. I knew that a while ago, the CPU and RAM had to be overclocked in tandem (kind of), but I thought now they can be overclocked separately? I only got interested in computers around the time Ivy Bridge and Nvidia's Kepler GPU's were being released, so not that long ago.

Or are you showing that the CPU is acting as almost a bottleneck to RAM read/writes (because its essentially instant), so overclocking the CPU will "widen" the bottleneck and allow the RAM to perform?
Because perfect, already got an overclocked CPU (could clock it higher, but then I would be messing with voltage, which I am not yet entirely comfortable with on the CPU).
m
0
l
a b } Memory
September 2, 2012 2:10:43 AM

manofchalk said:
Or are you showing that the CPU is acting as almost a bottleneck to RAM read/writes (because its essentially instant), so overclocking the CPU will "widen" the bottleneck and allow the RAM to perform?

No, RAM is still "slow as hell" compared to the CPU. The reason CPUs manage to perform so well regardless of memory performance in most benchmarks is because CPUs, compilers, algorithms, libraries, etc. have become more efficient at working around DRAM's high latency and low bandwidth by making better use of L2/L3 caches and prefetching capabilities that it does not matter as much anymore unless you are applying trivial computations to really large data sets in which case memory bandwidth may end up being the bottleneck.

As for why OCing used to involved memory but now doesn't, in the 'age' of north bridges, normal CPU multipliers were locked with the memory clock derived directly from the FSB clock so the only way to overclock the CPU was to increase the FSB clock which also increases the memory clock. With modern overclocking CPUs (AMD and Intel's K-series chips), the CPU core multiplier is unlocked, which makes overclocking a simple matter of setting the core multiplier which does not affect anything else.

FSB overclocking was in large part popular due to the lack of core multiplier access. FSB overclocking is practically dead but now people can pay extra for unlocked chips that enable them to use safer overclocking through multipliers.
m
0
l
a c 347 } Memory
September 2, 2012 2:50:10 PM

manofchalk said:
Now I am getting really confused. I knew that a while ago, the CPU and RAM had to be overclocked in tandem (kind of), but I thought now they can be overclocked separately? I only got interested in computers around the time Ivy Bridge and Nvidia's Kepler GPU's were being released, so not that long ago.

RAM and CPU can be overclocked separately it's just the gains of RAM OC by itself has a much smaller impact over performance gains in comparison to the CPU OC. To suggest a CPU isn't a bottleneck isn't true, it's more about Instructions Per Second (MIPS). The analogy is a car with wheels Rated at 1000MPH with a 200MPH engine and wondering how fast the car will go.

Peak Transfer Rate = transfer rate of (memory clock rate) × 4 (for bus clock multiplier) × 2 (for data rate) × 64 (number of bits transferred) / 8 (number of bits/byte).

DDR3-1600 = 800MHz * 2 * 64-bit * 1/8 = 12800MB/s = 12.5GB/s

CPU Max Memory Bandwidth:
Sandy Bridge (LGA 1155); i7-2600K - http://ark.intel.com/products/52214/Intel-Core-i7-2600K...(8M-Cache-up-to-3_80-GHz)
21GB/s = 21504MB/s = XMHz * 2 * 64-bit * 1/8
X = 1344MHz = DDR3-2688 and there are DDR3-2600 Kits.

Sandy Bridge Extreme (LGA 2011); i7-3930K - http://ark.intel.com/products/63697/Intel-Core-i7-3930K...(12M-Cache-up-to-3_80-GHz)
51.2GB/s = 52428.8MB/s = XMHz * 2 * 64-bit * 1/8
X = 3276.8MHz = DDR3-6553.6 and you'd need DDR4/DDR5 RAM to saturate; the RAM used on the GPU you mentioned is pretty close

Ivy Bridge (LGA 1155); i7-3770K - http://ark.intel.com/products/65523/Intel-Core-i7-3770K...(8M-Cache-up-to-3_90-GHz)
25.6GB/s = 26214.4MB/s = XMHz * 2 * 64-bit * 1/8
X = 1638.4MHz = DDR3-3276.8 and there are DDR3-2800 Kits and folks have OC the RAM >DDR3-3000

My point is the current DDR3 technology has reached it's limits, and soon DDR4 will be introduced to keep the gap closed between the CPU's & RAM's Max Memory Bandwidth.

Further, in the real world there's little gains to be found from much faster RAM (Higher Frequency to Lower CAS Latency). There's been a few decent Articles but here's one that was compiled for the Sandy Bridge and the difference to the Ivy Bridge is very small; you must consider ROI (Return On Investment) and/or Diminishing Returns On Investment:
Article - http://www.bit-tech.net/hardware/memory/2011/01/11/the-...
Article - http://www.tomshardware.com/reviews/quad-channel-ddr3-m...

manofchalk said:
Or are you showing that the CPU is acting as almost a bottleneck to RAM read/writes (because its essentially instant), so overclocking the CPU will "widen" the bottleneck and allow the RAM to perform?
Because perfect, already got an overclocked CPU (could clock it higher, but then I would be messing with voltage, which I am not yet entirely comfortable with on the CPU).

By overclocking the CPU I am essentially increasing the MIPS or it's ability to process more 'data' more quickly. This isn't a case of a couple variables in a linear 'formula', but instead you get into an odder 'curved' architectural, application, environmental, etc driven results. In the case of Applications running on a 'RAM Drive' the application's structure and coding becomes a bottleneck.

Crystal Ball (Moore's Law Theory) - you're getting into computers IMO at a bad time, Intel is no longer threatened by AMD so those folks that 'hate' AMD CPU are causing their own demise. So if we got in a time machine and compare CPU 5 years in the future they won't be as proportionally as fast as the difference between 2007 and now. Innovation comes from competition and Moore's Law is driven by the inherent assumption of competition, so you'll start to see the liner gains over time to start to level-out. Intel since the AMD FX (Bulldozer) failure has postponed CPU releases and it's Tick/Tock CPU scheduled release time-frames. My hope is that AMD's next CPU line can kick the crap out of Intel to re-spark competitive innovation. No doubt, the SB-E has 8-cores and Intel disables 2-cores, but if the AMD 8-core FX was faster then today you could by an 8-core SB-E and the IB-E would be right around the corner but instead might be cancelled, and I foresee Haswell release being dragged-out. Pray AMD's next release is fantastic and folks flock to buying them in droves. Imagine how bad your phones would be if Apple never invented or re-invented the cell phone - iPhone. Question yourself if there would there even be an Android phone?

Competition is the consumers best friend for innovation.
m
0
l
a b } Memory
September 2, 2012 3:27:43 PM

jaquith said:
So if we got in a time machine and compare CPU 5 years in the future they won't be as proportionally as fast as the difference between 2007 and now. Innovation comes from competition and Moore's Law is driven by the inherent assumption of competition, so you'll start to see the liner gains over time to start to level-out.

Moore's "Law" is about transistor count, not performance. Transistor count is still going up due to larger caches, new instructions, extra cores, more powerful IGP, more and faster integrated IO standards, etc. but throughput per core has somewhat plateau'd due to stock clock speeds hitting somewhat of a brick wall just above 3GHz.

Even AMD which has lots of catching up to do and everything to gain from doing so is very far from doubling performance every 18 months as used to be the case 15 years ago. The "performance scaling brick wall" problem is not only an "Intel does not need to compete anymore" phenomenon. Traditional performance scaling simply no longer works and until programmers become comfortable writing massively multithreaded applications and games, massively multithreaded/multicored CPUs make no sense either.

I have a feeling quad-core CPUs around 3.5GHz are going to be the mainstay of personal computing for a very long time. Only thing to do in the meantime is hope they will become more affordable. (That last bit is aimed at Intel since AMD's quads are already pretty cheap.)
m
0
l
a c 347 } Memory
September 2, 2012 3:49:26 PM

InvalidError said:
Moore's "Law" is about transistor count, not performance.

No Shi*, the transistor count (performance correlation) future CPU's has been STALLED (slowed)! If you read what I stated, innovation through competition is in trouble. The Tick/Tock is all about lithography = transistor count and the time between Intel's CPU's release has been delayed. Therefore, the Transistor's 5 years ago -> Today -> 5 years in the future will not adhere to Moore's Law any longer.

Intel has no motivational need to trip over themselves to release better & faster products.

Imagine the CPU's today if there were no AMD...more than likely we'd be lucky to have anything more than P4 or Core 2 Duo. Oligopolies are bad and Monopolies are devastating to innovation.
m
0
l
a b } Memory
September 2, 2012 4:16:32 PM

jaquith said:
Intel has no motivational need to trip over themselves to release better & faster products.

Are you saying that AMD is also intentionally 'failing to deliver faster and better products as well?

Both AMD and Intel are hitting the same brick wall called the x86 legacy. The only fix for that would be to ditch the x86 instruction set and all the kludges that come with it such as highly non-orthogonal instruction set, huge renamed register files to work around the very limited and highly non-orthogonal register set, too many redundant addressing schemes, etc.

With Microsoft taking an interest in ARM platforms, maybe we'll have ARM-powered desktop PCs in 5-10 years.
m
0
l
!