Can RAM quality limit my overclock

[ClaudeWalker]

'm just getting into water cooling and I don't want "performance" ram, I just want ram that won't limit my overclocking of the CPU. I'll happily under clock ram if need be. BUT, can my CPU overclock be limited by my RAM?

Thanks,
JOe K.

 

[tolis626]

'm just getting into water cooling and I don't want "performance" ram, I just want ram that won't limit my overclocking of the CPU. I'll happily under clock ram if need be. BUT, can my CPU overclock be limited by my RAM?

Thanks,
JOe K.

Well, I don't think the quality of the RAM itself can limit your CPU's overclocking potential. Quite the opposite, slower RAM speeds usually mean you can hit higher clocks on the CPU, but not by much these days. I mean, I can run my RAM as low as 1600MHz CL10 or as high as 2400MHz CL10 or my daily 2200MHz CL9, and I can't see a difference in CPU clock speeds. On my chip 4.8GHz is practicaly unachievable (At reasonable volts at least), so I'm "stuck" with 4.7GHz. Memory didn't seem to make that much of a difference there. Maybe I could run my CPU at something like 0.005V lower, but I'll take such an increase in voltage over slow RAM any day of the week.

If you're set on only overclocking your CPU though, underclocking and undervolting your uncore (cache) yields better results than underclocking your RAM. Trial and error will be necessary though.

With all that said, I think with what "good" DDR3 costs these days, it's not worth going with the budget-friendly and slow kits. Spend something more and get something decent. Fancy RAM heatsinks and flashing lights may not be your thing, but still there are great kits out there that look simple and serious. Keep it sane on the clocks and you'll be good to go.

PS : I wouldn't get anything lower than 1600MHz CL9. That's like bare minimum these days IMO.

 

[BonzaiDuck]

Acknowledging Tolis626 in his answer -- as well.

It once seem the case that RAM settings and CPU overclock settings were knitted together, ending with socket LGA-775.

Now -- these days -- it would seem that you can set aside memory-tweaking, setting it to stock or XMP specs, and focus entirely on the CPU.

But I've noticed some things with my sig-rig processor -- couple gens behind the latest Haswells.

By this example I explain it. My RAM settings of 1866, 1.5V 9-9-9-24 2T were effective when I first got the system to be rock-solid at 4.7 Ghz with certain VCORE offset and "Extra-Turbo" voltage settings.

Later or most recently, I set the RAM to the XMP profile giving DDR3-1866 speed, 1.5V, and "auto" timings (for everything) which yielded effective 9-9-9-24 1T (command-rate). Suddenly -- the processor needs more voltage to sustain stability testing, and the stop-codes were mostly "09C" -- indicating IMC or bus voltage deficiency. It was remedied by either bumping up VCCIO, settings affecting VCORE -- or more generally -- both.

I got rid of the 09c stop codes after bumping up VCCIO to an effective 1.12V; I got rid of all instability by adding about 10 mV to VCORE. I'm confident I could go back to CMD-rate= 2, drop the VCORE and VCCIO back to earlier settings, and all would be wonderful.

This is a very small impact of RAM settings on my general overclock. So to the OP's question, I could answer "YES," and I could answer "NO."

I've said here or at other Anandtech forums/topic-areas and just recently: You can get good RAM like these newer low-profile Crucial Ballistix 1600's running at 1.35V and 8-8-8-24. I'm guessing you can overclock them (at least!) to 1866 and 9-9-9-24 without bumping up the DIMM voltage. Maybe you could even get to 2133 with looser timings!

But given the time and trouble in testing large amounts of RAM, I'd just as soon spring for higher-rated RAM and pay the slight difference, then settle on running them at those specs -- with exception of the command-rate.

But again, and referring freshly to my example, I don't think your RAM specs are going to MUCH affect your CPU clocking potential. And basically, with my example, I'm not sure the command-rate change qualifies as "limiting CPU clock potential."

 

[DrMrLordX]

I would like to point out that, depending on the processor, your RAM speed and timings will affect how much benefit you gain from overclocking. Faster memory = better scaling.

Fast RAM never (technically) makes your system faster. It simply removes a potential bottleneck to CPU performance. Ideally, all software would fit in L1 cache, and you'd never touch external memory. In the real world, that never happens, so the CPU sometimes has to grab data from higher levels of cache or system memory. System memory offers lower bandwidth and higher latency than cache, making it the second least desirable place to grab data (the least desirable place is from memory swapped to a mechanical hdd, yech).

If you are looking at Intel processors, they have extremely sophisticated cache setups that will help your processor work well at many clockspeeds, even if your RAM isn't extraordinary. All you really need to do is make sure you avail yourself of all available memory channels to avoid a bandwidth bottleneck, and then you should be okay. Still, someone who takes the time to reduce overall system memory latency by increasing memory clock and/or reducing timings will get better scaling out of their chip as they overclock.

AMD chips have slower cache, so system memory can make a bigger impact on their ability to scale with higher clockspeed.

Regardless, if you are planning to overclock, you should expect the first few hundred mhz of overclocking to provide more performance than the last hundred mhz you tack on at the end. Faster memory/tighter timings can mitigate or eliminate that effect, as can larger/faster L2/L3 cache.

 

[BonzaiDuck]

I would like to point out that, depending on the processor, your RAM speed and timings will affect how much benefit you gain from overclocking. Faster memory = better scaling.

Fast RAM never (technically) makes your system faster. It simply removes a potential bottleneck to CPU performance. Ideally, all software would fit in L1 cache, and you'd never touch external memory. In the real world, that never happens, so the CPU sometimes has to grab data from higher levels of cache or system memory. System memory offers lower bandwidth and higher latency than cache, making it the second least desirable place to grab data (the least desirable place is from memory swapped to a mechanical hdd, yech).

If you are looking at Intel processors, they have extremely sophisticated cache setups that will help your processor work well at many clockspeeds, even if your RAM isn't extraordinary. All you really need to do is make sure you avail yourself of all available memory channels to avoid a bandwidth bottleneck, and then you should be okay. Still, someone who takes the time to reduce overall system memory latency by increasing memory clock and/or reducing timings will get better scaling out of their chip as they overclock.

AMD chips have slower cache, so system memory can make a bigger impact on their ability to scale with higher clockspeed.

Regardless, if you are planning to overclock, you should expect the first few hundred mhz of overclocking to provide more performance than the last hundred mhz you tack on at the end. Faster memory/tighter timings can mitigate or eliminate that effect, as can larger/faster L2/L3 cache.

Absolutely. Positively. Sometimes, you can almost visually perceive how a profound memory tweak can improve things at the higher clock speed. But you can benchmark the system using the same memory settings and different CPU speeds, and it may not seem that the last 100 Mhz matters all that much.

I think it's the case that each processor gen's IMC has its own range, though. Correct me if wrong. But -- for instance -- I think my SB-K and Z68 system doesn't get much better in a leap from 1866 Mhz RAM and 2133. If I'm correct, what are the expectations for the Haswell?

 

[Deders]

It once seem the case that RAM settings and CPU overclock settings were knitted together, ending with socket LGA-1156/1366.

Fixed

 

[DrMrLordX]

Absolutely. Positively. Sometimes, you can almost visually perceive how a profound memory tweak can improve things at the higher clock speed. But you can benchmark the system using the same memory settings and different CPU speeds, and it may not seem that the last 100 Mhz matters all that much.

I think it's the case that each processor gen's IMC has its own range, though. Correct me if wrong. But -- for instance -- I think my SB-K and Z68 system doesn't get much better in a leap from 1866 Mhz RAM and 2133. If I'm correct, what are the expectations for the Haswell?

Not sure, but there are some memory benches for Haswell out there somewhere. Really it's not going to be absolutely uniform between different Haswell flavors since different Haswells have different amounts of cache per core. It will also depend on how many threads the CPU has to handle, and uncore speed.

As an extreme example, system memory performance is probably not a big issue for Haswell-E when handling a single-threaded app, since the available L3 for that one working core would be enormous. On the other end, we have the G3258, though most people seem to agree that memory speed/timings aren't that big of a deal for that chip either. Heck many people are content to run it at DDR3-1333 (non z87/97) or DDR3-1600 (z97 Mate).