[torin3]

The new memory I've gotten for my 4-way classified seems to be working well to restore my overclock. However, I noticed I'm running 11-11-11-29 1T. The ram can run about this speed a 7-8-7-24 2T. Would I be better off to change to 2T and tighten up the timings? I'm running about 1800mzh on the ram.

[ChasR]

This is 2133 MHz ram? If it is, I'd try tightening the timings at 1T and see how low you can go. I'd think you could run close to spec timings at 1T with the ram underclocked.

[torin3]

Originally Posted by ChasR

This is 2133 MHz ram? If it is, I'd try tightening the timings at 1T and see how low you can go. I'd think you could run close to spec timings at 1T with the ram underclocked.

Actually it is 1600 Mhz ram. The modules that Onefstsnake recommended.
http://www.newegg.com/Product/Produc...82E16820231335

And it looks like my memory is a bit off. It is supposed to be 7-8-7-24 2T at 1600 Mhz. It is 1.5V and I'm running it at 1.62V.

I've been letting the timings go on auto. I can probably get them tighter, but I was wondering if dropping them to 2T would be that much of a handicap.

[ChasR]

I haven't done any testing on the effects of command rate on folding performance. I would think you could do better than 11-11-11 though.

[harlam357]

Humm... yeah, I think most kits XMP profile (or better) is rated at 2T. Gosh, I haven't run 1T in anything since my old 939 setups. Memory is so fast (and inherently the most flakey piece in a system) that it's the last thing I want to really fool with in a rig. However, I think when DDR2/3 memory is run at JEDEC specs it is capable of running 1T based on the JEDEC SPD.

EDIT: Opps... sorry, my name is not ChasR.

['Cuda340]

I was under the impression that going from Command Rate T2 -> T1 was like tightening the CAS Latency by one........(In terms of performance) Say from 9-9-9 to 8-9-9

I have not done any testing.......

[ChasR]

My RipJawsX 2133 has an XMP rating of 9-10-9-28 2T, per G. Skill and 9-11-9-29-1T per CPU-Z. The ASUS p8p67 pro reads that as 9-10-9-27-2T for some reason. In CPU-Z command rate is blank for the JDEC SPD speeds. So what does that mean? Beats Me.

I'll see about doing some testing. It would be a lot easier if it didn't take so long to complete a frame. On regular SMP, clicking off frames in 3:00 or less, it doesn't take long to get results. At 22:00+/frame, I forget what I'm doing, before I get results. .

[torin3]

Originally Posted by harlam357

EDIT: Opps... sorry, my name is not ChasR.

No problem. I've edited the title. I just wanted to catch ChasR's attention, since it was his comments that got me looking at my memory as the reason my frame times were slower than other people with 12threads at the same clock speed.

I'm now getting about 61K ppd or around 21-22min tpf with my current ram speed and 4.28Ghz.

I think I'll try and tighten them up to 9-9-9-26 1T (and bump the juice a bit, 1.65v probably) and see how that does.

['Cuda340]

So far i have not had any DDR3 that did not run "T1" at whatever it's rated speed was..........(All were triple channel kits)

[ChasR]

On bandwidth benchmarks command rate can make a measured difference of up to 25% in bandwidth. In the real world of applications, that's usually only 2 to 3%, the same as reducing latency by one tick, as you point out. How it relates to folding, I will one day find out. It may be impossible to make myself perform a scientific test, using the same frames of the same WU because I'd lose too many ppd. We may have to settle on an unscientific test, but that may not mean anything if we're looking at a 1 or two percent difference.

[freeagent]

I actually have a triple set of some gskills that only run at 2T

From my observations playing around with my current setup, I get strong performance with a high bclk (191-196bclk) with my ram at 7-7-7-20. If I go to say 200, and run the ram at 1600 8-8-8-24 performace tapers off noticeably, until I get the mem speed up to around 1750mhz at 8-8-8-24 wich gives me a slightly lower latency then at say 1560 at 7-7-7.

[onefstsnake]

9-9-9-24-2T @ 1829mhz and 1.65v
7-8-7-24-1T @ 1700mhz and 1.65v

The top was just a quick and dirty setup. Both using the same ram that you have.

[torin3]

9-9-9-26 1T 1746Mhz right now. Seems to be handling it fine.

Though I'm wondering why the board won't boot at 173 bclk. Boots fine at 172 bclk, and I can move it up after I'm in windows.

[onefstsnake]

blk hole?

My r3e couldnt run 226-230 but could run 235

[torin3]

Well, I'm now at 9-9-9-26 1T at 1 lower on bclk and I'm getting 20:40 frame times when the system is idle.

[87dtna]

At 1742mhz you should atleast be able to run 8-9-8-24-1t on stock voltage, and could most likely run 7-8-7-24-1t with 1.65v.

Whats your north bridge and QPI multipliers at? Make sure they aren't too high and causing instability.
Whats your VTT at? With ram speed at 1700+ I would imagine you'd need atleast 1.25v VTT if not 1.30, in that area, to be stable.

[torin3]

Originally Posted by 87dtna

At 1742mhz you should atleast be able to run 8-9-8-24-1t on stock voltage, and could most likely run 7-8-7-24-1t with 1.65v.

Whats your north bridge and QPI multipliers at? Make sure they aren't too high and causing instability.
Whats your VTT at? With ram speed at 1700+ I would imagine you'd need atleast 1.25v VTT if not 1.30, in that area, to be stable.

I'll have to wait until I get home to check the multis, but I believe my VTT is 1.35V.

[torin3]

Ok, I forgot to check the other settings when I was rebooting today, but I'm now at 8-8-8-24, and I can get to 174bclk with it before Intel burntest/linpack fails. I'm at 1.65V in bios and 1.67V reported for RAM memory.

Should I up the voltage a bit? I'll know in the morning how it compares to 9-9-9-26 at 179bclk. I was getting 20:37 TPF when the computer was idle.

[torin3]

20:50 TPF. But it is 150 Mhz slower.