Theoretical Question on DDR3 2133/1866/1600/1333/1066

[XeoNoX]

My board supports DDR3 2133/1866/1600/1333/1066. I have a 2500K. I see all these people buying CAS 9 at 1.5v. i notice most people buy 1600 (PC12800) cuz its most common in most current systems. However i have noticed DDR3 2133 (PC3 17000) prices dropping pretty close to the 1600 (PC12800).

Qustion 1: Now OC question is i notice most people tinkering with CAS and the volts to achive higher speeds (in OCing). Lets say someone buys 1600 (PC12800) CAS9 1.5v and they bump the ram to 2133 CAS 11 @ 1.65v. Would u guys agree that it was better off buying a stick of 2133 CAS 9 (PC3 17000) @ 1.5v at same price???

Question 2: using the same scenario above, my THEORY is that wouldnt it it be possible to underclock the 2133 CAS 9 (PC3 17000) @1.5v RAM down to 1600 and achieve tighter times as lets just say CAS 8 timings?

Question 3: In buying 2133 CAS 9 (PC3 17000) 1.5v rated ram on a system that would normally run 1600 (PC12800), wouldn't this give me more ROOM for getting a better OC of my CPU??

 

[Fourstaff]

Yes, if you get ram for the same price, its better to get the faster one at lower voltage (with same latencies).

Yes, its possible to underclock, but you might not be able to achieve CAS 8

Its the same, if both of the sticks can go up to 2133, then your overclock room is the same. If you get one stick at 1600 and another at 2133, and then found out that the 1600 doesnt go all the way up to 2133 then obviously your OC room is smaller

Regardless, Sandy Bridges doesn't show much (if at all) gains after 1600 Mhz, CAS9 so all your theoretical questions are more or less useless with regard to S-B

 

[brandonwh64]

1600mhz Cas7 1.5V is something I would prefer.

 

[lilhasselhoffer]

1) RAM is rated for a maximum tested speed, with a given timing. You could purchase a batch of RAM rated at 1600, but puch it up to 2133. The manufacturer does not guarantee this though.

2) Yes, 1600 pushed to 2133 will be looser timings compared to a 2133 stick you buy with the same initial timings as the 1600. You're asking if a looser timing influences RAM performance perhaps? If so, it does.

3) RAM tweaks balance greater frequencies, with looser timings. THe dirty methodology is to divide frequency by timing to get performance. As such, RAM would have to have double the frequency to justify halving the timings (1200-2400 implies times go from 2-4 for the sake of example numbers). This delicate balance is what RAM overclocking is all about.

4) YOu can clock RAM to the moon. If your PCH or chipset does not support that frequency then it's all for naught...

 

[BlackOmega]

My board supports DDR3 2133/1866/1600/1333/1066. I have a 2500K. I see all these people buying CAS 9 at 1.5v. i notice most people buy 1600 (PC12800) cuz its most common in most current systems. However i have noticed DDR3 2133 (PC3 17000) prices dropping pretty close to the 1600 (PC12800).

Qustion 1: Now OC question is i notice most people tinkering with CAS and the volts to achive higher speeds (in OCing). Lets say someone buys 1600 (PC12800) CAS9 1.5v and they bump the ram to 2133 CAS 11 @ 1.65v. Would u guys agree that it was better off buying a stick of 2133 CAS 9 (PC3 17000) @ 1.5v at same price???

Question 2: using the same scenario above, my THEORY is that wouldnt it it be possible to underclock the 2133 CAS 9 (PC3 17000) @1.5v RAM down to 1600 and achieve tighter times as lets just say CAS 8 timings?

Question 3: In buying 2133 CAS 9 (PC3 17000) 1.5v rated ram on a system that would normally run 1600 (PC12800), wouldn't this give me more ROOM for getting a better OC of my CPU??

1) IF the ram is the same price and capacity between 1600 CAS9 and 2133 CAS 9, you always want the higher frequency RAM even if your board doesn't support it. And this is where it ties to question 2.
2)If you get the 2133 CAS 9 sticks, and drop the frequency down to 1600MHz, you'll most likely be able to tighten the timings WAY up. Like CAS 7 and possibly even 6. IF you drop it lower still to, lets say, 1333, you may be able to get CAS 5 out of them.
The thing is, higher frequency does not necessarily equate to better performance. For instance, 1600 CAS 6 would either be on par with, or even slightly faster than, 2133 CAS9. Simply because it sends the signal to the RAM faster and thus it would be a lot more responsive.

3)Yes, lower the frequency and leaving the CAS alone can and does affect your overall CPU OC. However, if you lower the frequency and tighten the timings, it can place just as much load on the integrated memory controller (IMC) as the higher frequency.
I personally have found that it's better to lower your CPU frequency slightly in lieu of higher RAM frequency and tighter timings because the system will be overall faster and a lot more responsive.
RAM timings can affect a lot. For example on my AM3 rig, I noticed an almost 1000 point gain in 3dmark06 just from messing with the CPU-NB and RAM frequencies and tightening the timings (1600MHz, 7-7-7-21-28-1T @ 2650MHz CPU-NB).

 

[Anusha]

one thing to remember is all these timings are given in clock cycles. CL8 isn't necessarily faster than CL10. Say,

1600 CL8 vs. 2133 CL10.
1600 CL8 -> CL in terms of time is 5ns
2133 CL10 -> CL in terms of time is 4.688ns

2133 CL10 is faster in terms of latency and it has more bandwidth. clearly, it is the better choice.

but if we are talking about 1600 CL7, then the CL is 4.375ns. While it has better latencies than 2133 CL10, 2133's bandwidth might make it faster.

But if you are talking about 1600 CL6, then it is 3.75ns, which might be preferred to the 2133 CL10 because latency is significantly better.

if you are confused about the calculation, it is
(1/clock speed in Hz) * 10^9 for ns * CL clocks == CL clocks / speed in GHz (gives result in ns)

But this is theoretical. Practically, it might depend on the platform and the applications. Better to check out reviews.

 

[cadaveca]

if you are confused about the calculation, it is
(1/clock speed in Hz) * 10^9 for ns * CL clocks == CL clocks / speed in GHz (gives result in ns)

But this is theoretical. Practically, it might depend on the platform and the applications. Better to check out reviews.

Good advice. there. Thanks very much for highlighting that it's theoretical.

AMD's bandwidth efficiency is about 64%.

Intel's bandwdith efficiency is about 82%

So, at the same clocks, Intel gives near 20% more bandwidth.

 

[XeoNoX]

thanks guys that was such great info, i understand now, so my theories werent too far off

now its just time for me to figure out the price/performance difference, and equate it into the rest of my pricing plan for best possible performance for the cheapest price/performance ratio.

So i decided to go with 2x4GB DDR3 1600 (PC3 12800) 9-9-9-24-2N 1.5v for $37 shipped, not too shabby i figure ill have some tweaking room to mess with it.

 

[Anusha]

thanks guys that was such great info, i understand now, so my theories werent too far off

now its just time for me to figure out the price/performance difference, and equate it into the rest of my pricing plan for best possible performance for the cheapest price/performance ratio.

So i decided to go with 2x4GB DDR3 1600 (PC3 12800) 9-9-9-24-2N 1.5v for $37 shipped, not too shabby i figure ill have some tweaking room to mess with it.

but remember one more thing. reviews might say you can OC it to say 1866 or even 2133 with loser timings and higher voltages, but don't ride by that assumption that you will get the same results. if you want to be sure, get RAM with higher rated clocks.

for example, i have 1600MHz CL9 Vengeance RAM. most reviews say you can hit almost 2000MHz with them. i can't. i cannot even hit 1866 with CL11 and 1.65V. my mistake for trusting the reviews. (it's not their fault but...) don't make the same mistake. sure, the performance delta isn't noticeable, but you know inside that you could have gone with faster RAM. regret is far worse than losing a little bit more money.