OC w/ 2 different RAM types?

[dez93]

Hi all,
Complete OC noob but been doing tons of reading. Nearly ready to start but for this one thing: I have 4 sticks of RAM, in 2 different flavours.

Originally I bought:
http://www.newegg.com/Product/Produc...82E16820148069
Crucial Ballistix 2GB (2 x 1GB) 240-Pin DDR2 SDRAM DDR2 1066 (PC2 8500) Dual Channel Kit Desktop Memory Model BL2KIT12864AA1065, Timing 5-5-5-15, people have had it at 4-4-4-8 on boards.

Then recently I added two of these (I think):
http://www.newegg.com/Product/Produc...82E16820139229
Kingston 2GB 240-Pin DDR2 SDRAM Unbuffered DDR2 800 (PC2 6400) System Specific Memory Model KTL2975C6/2G, CAS:6.

So 2x1gb good spec about 5+ years ago, and 2x2gb cheap spec (tho times have changed so everything is cheaper) a few weeks ago.

If I just had the original sticks, I'm ready to start my OC (E6600, Gigabyte DS3r3.3 board, seemingly very common combo, aiming for 3.6GHz). BUT: what happens with the addition of the slower (probably) memory? What do people recommend? Can I have the quick sticks doing the power work and the bigger slow ones helping with the slow heavy lifting?

Were they a waste of money...?

Thanks!!

 

[KingFatty]

First question would be, how much memory do you need? Can you get by with just the 4 GB of Kingston, so you can just not use the Crucial 2 GB? Will 6 GB be noticeably better than 4 GB? I'm guessing no, so it makes sense to just use one type of ram for more stable performance and potential ability to overclock.

The next question is, if you want to mix ram, what is the highest specifications that both memory types are able to share? This would depend on several factors, like which voltage they can run at, how they can overclock, etc. The reason this matters is that perhaps you will end up sacrificing any relative difference in performance between the ram types, because if you mix them, you'll have to run all of them at the lowest common denominator. As an example, mixed ram may need to be run at slower settings than even the slowest ram, because it may be more unstable than if you just use one type of ram and get better stability and can overclock it more.

However, in the end, it may be that overclocking RAM is just not worth the extra heat and potential instability. I'm not too familiar with DDR2, but it has been shown that with DDR3 memory, overclocking doesn't really give you a noticeable difference in performance anywhere, except maybe in artificial memory benchmarking programs.

If that holds true for DDR2, then maybe you can just not worry about trying to overclock the ram, and instead focus on getting the ram as stable as possible to support a nice fat CPU overclock.

 

[dez93]

My assumption, from the reading, was that one needs to OC the memory as part of the general OC, to attain higher speeds... otherwise why would people bother at all?

Conceptually, then, it'd be worth my while investigating what my likely highest attainable OC is on stock clock RAM, in order to benefit from the 6gb, as opposed to either the 2gb of fast or 4gb medium speed sticks on their own...?

Cheers for the advice dude.

 

[BUnit1701]

I run a mixed setup (2x2GB of DDR2-1066 and 2x2GB of DDR2-800, headspreaders on the 800Mhz RAM prevented using it in the two slots near my CPU, and too cheep to buy 8GB of DDR2-1066) Generally, I run at 800 Mhz, but have run as high as 950 at times. The harsh reality is that you will be limited by the speed of the slowest RAM in your system. In your case, I think I would stick with just 4GB, especially considering the relative age and slowness of the accompanying hardware, speed of RAM will be more valuable to you than volume of RAM.

EDIT: Sorry, had misunderstood, thought you had 4GB of 1066 and 2GB of 800. Start with just the 800Mhz RAM, see how high it will clock. Assuming they play nice together, there should be little to no issues running the 1066 RAM and reduced clocks and timings to match wherever the 800Mhz stuff ends up.

 

[dez93]

Thanks - glad it wasn't me getting confused there!
Ok, i'll give that a go. Cheers.

So, out of interest, which did you mean was more important - speed or volume of RAM? For reference, I dual boot xubuntu & windows XP, and don't play games on it (any more. adulthood. boo.)

 

[BUnit1701]

Speed of RAM is how fast it is, I.E. 800, 1066, 1600, etc. Volume of RAM is how much you have, 2GB, 4GB, 8GB, etc. In your specific scenario (slow accompanying hardware) it is my opinion that faster ram would be more important than more ram, with the caveat that you have at least 4GB.

 

[dez93]

Why the 4gb threshold? Don't get me wrong - i bought the extra sticks precisely because 2gb was becoming limiting (albeit only because of firefox, thunderbird & the occasional Rhythmbox memory holes), but it interesting to hear you put a specific number on it.

 

[BUnit1701]

4GB is the 'lowest acceptable amount' that is commonly recommended for a PC system today. There is no hard and fast number, and it changes over time.

 

[T_Yamamoto]

4GB is the 'lowest acceptable amount' that is commonly recommended for a PC system today. There is no hard and fast number, and it changes over time.

Its more of a 'standard' feature for a computer, sorta like airbags on cars.

 

[dez93]

Righty, so now I'm at home & can confirm my spec, it seems like potentially all 4 chips could potentially run at these settings that the guy I'm copying did:

"with final F4 bios I'm doing 9 x 400 / E6600 retail cas 4-4-4-12 / 2 gb 667 @ ddr800 el cheapo ram / all on air and 100% silent

CPU Clock Ratio (Note)____________ [X9] <<<----CPU Multiplier
CPU Host Clock Control_ [Enabled]
CPU Host Frequency (MHz)__________ [400] <<<----FSB Speed (Front Side Buss)
PCI Express Frequency (Mhz)_______ [100]
C.I.A. 2__________________________ [Disabled]
System Memory Multiplier (SPD)____ [2.00]
Memory Frequency (Mhz) 800
DRAM Timing Selectable_______ SPD __ [Manual]
CAS Latency Time_____________ [4]
Dram RAS# to CAS# Delay______ [4]
Dram RAS# Precharge Delay_____[4]
Precharge Delay (tRAS)________[12]
ACT to ACT Delay (tRRD)_______4 _____[auto]
Rank Write to READ Delay______3 _____[auto]
Write to Precharge Delay______6 _____[auto]
Refresh to ACT Delay________42 ______[0]
Read to Precharge Delay_______4 _____[auto]
Memory Performance Enhance__________ [Normal]
High Speed DRAM DLL Settings________ [Option 1]
System Voltage Control____ [Manual]
DDR2 OverVoltage Control__ [+0.200V]
PCI-E OverVoltage Control_ [+0.1V]
FSB OverVoltage Control___ [+0.1V]
(G)MCH OverVoltage Control [+0.2V]
CPU Voltage Control_______ [1.325V] (he also wrote 1.500 above this, don't know what that's supposed to correspond to. Hmm.)

Ive have placed a fat north bridge heatsink on the nb, and also 2 silent fans on the nb heatsink and south bridge. That helps alot ...
joeloe: http://www.xtremesystems.org/forums...-on-1st-page&p=1658836&viewfull=1#post1658836 "

...The question being whether the big/slow/cheap sticks will be able to deal with the 4-4-4-12 timings. KingFatty mentioned about not bothering OCing the memory, but I'm still unsure what effect that will have on the CPU OC - if it doesn't create a bottleneck, then why would anyone ever bother OCing their memory? And since everyone on the boards who was/is overclocking these mobos/chips/sticks WAS doing that*, surely there must be some reason?

And - n00b question alert - I assume the parameters one changes which count as 'overclocking the memory' are the timings (4-4-4-8 etc), and the system memory multiplier, and the memory frequency, and 'DRAM timing selectable'... and leaving all the ACT/READ stuff alone, as he's done already, and not touching the DDR2 overvoltage control?

Cheers for the chat guys.

*admittedly probably 97% of those people are copying the 3% who really truly know what they're doing, so that's not an independent 'everyone'!

 

[KingFatty]

There could be a constraint where it depends on how you are overclocking your CPU.

If you change the CPU multiplier and leave the FSB speed unchanged, then your CPU will be overclocked. Yet, the FSB speed has not changed.
In contrast, you could also include a change for the FSB speed. That will affect the ram. Now, what happens is you will also need to decide whether to multiply or divide the FSB speed by a different factor to affect what the ram ultimately runs at, using the system memory multiplier. Above, I think that would mean using 400 FSB multiplied by 2 for 800 total. But, if you increased your FSB to say 450, you could then try to run the ram at 450x2=900, or use a different memory multiplier like 1.75. With that lower memory multiplier, the ram would be 450x1.75=787.5. So that's slightly less than 800.

But you can see how you have flexibility on how to overclock, whether you rely on just the FSB changing, or just the CPU multiplier, or both. And maybe your ram can take higher speeds etc., but if not, and if you also increase the FSB, you can use the memory multiplier to go easy on the ram, by reducing the overall frequency product of FSB and memory multiplier.

Keep in mind some CPUs lock the multiplier, so your only option is to increase the FSB, and then you end up exceeding the memory's capacity because the memory is also run at a factor of the FSB, so you are forced to use the memory multiplier to go easy on the ram.

My point is that you want the CPU maxed out, so don't sacrifice CPU overclocking just to try to max out your RAM. IN the example above, you'd much rather run the CPU at 450 FSB, and use the ram at 787, instead of leaving the CPU at 400 and running the ram at 800. It gets interesting because you are stuck with using the memory multiplier at only certain increments, so you could end up with a slower memory speed and that's OK.

As far as whether a memory overclock is going to help, it may turn out that overclocking DDR2 memory yields better results than DDR3, I'm not too familiar with DDR2. However, I think you should definitely overclock it, to the extent that you focus on maxing out the CPU overclock, and then settle for the fasted memory speed considering the constraints above with the FSB and memory multiplier.

 

[dez93]

Firstly, dude, thank you. Mate this is so helpful, and it's so rare to have someone write it all out, rather than trying to wok it out oneself through inference.

So, to make sure I've understood it as much as anything:

Your proposal:
CPU Clock Ratio (Note)____________ [X8] <<<----CPU Multiplier
CPU Host Frequency (MHz)__________ [450] <<<----FSB Speed (Front Side Buss)
PCI Express Frequency (Mhz)_______ [100]
System Memory Multiplier (SPD)____ [1.75]
=Clock speed 3600Mhz & Memory Frequency 787.5Mhz

My first proposal:
CPU Clock Ratio (Note)____________ [X9] <<<----CPU Multiplier
CPU Host Frequency (MHz)__________ [400] <<<----FSB Speed (Front Side Buss)
PCI Express Frequency (Mhz)_______ [100]
System Memory Multiplier (SPD)____ [2.00]
=Clock speed 3600Mhz & Memory Frequency 800Mhz

"don't sacrifice CPU overclocking just to try to max out your RAM" [same CPU OC in both]
"focus on maxing out the CPU overclock" [done, same in both] "then settle for the faster memory speed considering the constraints above with the FSB and memory multiplier"

So would you recommend the second of those two scenarios, given that CPU speed is the same?
Incidentally, from recollection and a quick google somewhat supporting that, i don't think the memory multiplier GOES to 1.75 - i think increments are 2, 2.5, and up.

But overall it sounds like the spec i'm aiming for could/should be supported by all my RAM sticks, from what you've said.

Thanks again dude!

 

[KingFatty]

Ok now to get a bit more complicated.

It's possible your CPU can achieve a higher overclock by using a high FSB, and a lower multiplier, when compared to using a lower FSB, and higher multiplier. Or the reverse could be true. Ideally you'd determine the optimal setting, playing with both settings and picking the values that result in highest overall product of FSB and multiplier for the CPU.

Just throwing an example, think how maybe your CPU, using x7 multiplier, could reach a FSB of 460 for 460x7=3220, but if you try for the x8 CPU multiplier, maybe the CPU could only handle 400 FSB for 400x8=3200. So you'd get a faster overall value with x7. Then you'd maybe have to adjust the memory multiplier to keep the ram stable. But remember, with fsb at 460, there is a chance the ram would be unstable unless you lowered the memory multiplier. So then you balance overall system performance and think, is that mere 20 extra MHz of CPU speed worth lowering your ram speed?

Anyway, I just want to point out that your CPU can be stable at multiple different combinations of FSB speed and CPU multipliers. If you happen to max out CPU performance for two relatively similar settings, use the one that also benefits RAM performance as well. Sometimes it's good to sacrifice some ram overclocking to achieve a higher CPU overclock, but if it's close, that's fine to pick whatever.

The point being is that you can just aim for something you like, or you can create a nice little spreadsheet to keep track of all the combinations. It's just a matter of how much time you want to set the bios settings, run a stability test, and repeat.

With your example above, you see that your proposal is better because the ram runs at a faster speed, and both scenarios have the same CPU performance. If the RAM can handle it, go for that.

But, are you running a test for making sure your RAM is stable? There is a free program called memtest, you install it to a USB stick (or CD/DVD) and boot your computer off that media to run the memtest and verify your RAM settings are stable.

So really, you can just use the rated specs of your RAM as a starting point, and then push it as far as you like by testing each setting using memtest to verify your settings are stable. If it passes memtest, it is very likely to be reliable during everyday use (assuming your computer gets good ventilation, and doesn't overheat the memory sticks when you are loading the CPU and GPU to max).

 

[dez93]

Great tip re: using memtest first to check stability before going further. According to a check I did will RAMMon (free), the sticks are running at 800 currently, so that bodes well for this proposed overclock!

So would you suggest changing the 4 CAS/RAS timing settings to those proposed by the guy I'm copying (if they're different to what I'm currently running), then running memtest again?

Cheers!

 

[KingFatty]

Yes, and again, I wonder if trying to reach the bleeding edge of perfection for ram overclocks is worth some effort, but not too much, as it may not affect your gaming or general usage very much at all. Personally, I tend to get a bit anxious if my ram is on the brink of becoming unstable, because that can lead to silent errors in your data when the ram contents get scrambled due to instability, and you, say, save a file from RAM to the hard drive storage. usually error correction will catch that, but sometimes things slip by. But again, you should read up on DDR2 specifically, as my advice is more general and not specific to DDR2, as DDR2 may show more of an effect than DDR3 where you don't really see any gaming difference in performance for RAM overclocking.

But you sure see it when you overclock the CPU.

But look at this page and how overclocking DDR2 doesn't seem to do much for these older games that are already above 100 FPS:
http://www.anandtech.com/show/2162/6

Note this very telling quote at the end:

There is also the reality that Core 2 Duo and AM2 really don't need the highest memory speeds to perform best. What they need is low latency DDR2-800 or possibly 4-4-3 DDR2-1067 to get the best performance possible. The unfortunate reality that we see again and again in memory tests is that the super high memory speeds are great for bragging rights and flexible overclocking, but they really don't do much for increasing actual real world performance on either the C2D or AM2 platforms.

 

[dez93]

Cracking research!

So what I'm hoping for then, given:
"What they need is low latency DDR2-800"
is that my four 800 sticks can stably handle the tight timings proposed. Fingers crossed.
Thanks dude!