3770K SuperPi 32M/IMC issues

[I.nfraR.ed]

Hi,

 

I have 2 identical 3218B chips in terms of core clocks. Both validate around 6.9, run 1M at 6850 range, 3DMark 03/05 at 6.6+ range (6.5 with 1.74V pretty stable).

However there are 2 differences:

 

1. The second chip works no matter if I restart, shutdown completely or change any values in bios such as active cores and bclk.

First one have CBB issues (00 code) when changing active cores from bios and sometimes after a hard crash, so I need to heat up to -135 in order to save the settings when bios recovers.

Otherwise they are identical for core clocks.

 

2. Thought the second chip would be better and save me some LN2 with lack of CBB, so I can always run full pot without need of heat-freeze cycles. It was looking pretty good and in line with the first chip in terms of IMC capability. In fact all three chips I've bought so far seemed to work with 1300 memory and I don't have better modules to check real IMC capabilities. The oldest cpu and the first 6.9 chip are verified to run 32M at 1300 no problem with pretty tight timings.

 

Now, the issue is that it actually works at 1300MHz memory, both with 2600 divider or with lower divider and raised bclk. No 55 code for memory, no 23 code for IMC. I have some runs at ~1300 8-11-7 pifast and pi 1M. But when tried to run 32M it fails instantly. 1200 is around the max I can run 32M.

Tried CPU PLL low and high, VCCIO, VCCSA low and high, VDIMM low and high and nothing seems to work even for the initial loop. Tried very slack timings as well with no real difference, also tried with a known working good primaries with slackier secondary and tertiary.

I always delete all generated files in pi directory before run. It errors out with "Not exact in round". Same on watercooled cpu, checked it after the session.

 

So, is it the IMC not capable of stable operation at that frequency, cause timings make no difference? Haven't tested much 3770K, so I can't make a clear judgment... that's why any suggestion is welcomed.

 

I really wanted to keep the second chip and sell the other one, but now I can't decide.

 

Plus I can't run 32M at anything higher than 1.8V. It keeps restarting with a black screen and it makes it faster with higher vcore. Doesn't matter the frequency - tried 6.7GHz and 6Ghz. I'm stuck at ~6740MHz with 1.8V. Any tips about this?

 

Sorry for the long post, but I wanted to make it as clear and informative as possible.

Board is Maximus V Gene with 1 broken pin which should be VCC for the integrated video.

Edited October 2, 2012 by I.nfraR.ed

[TerraRaptor]

What is DRAM additional swizzle for READ and WRITE? Check if cpu is capable of 1300 with this option enabled. Try bigger value for Latency Boundry.

[I.nfraR.ed]

Hmm, it seems I've read the documentation wrong. The pin is VCC for the cpu.

Going to check with the other cpu when I get back from work and if it's the same, then I have no option, but change the mobo.

 

Not sure if I read it right, but...

 

Broken pin marked on a photo found with google

 

 

If I read it correctly, it sould be A18.

From the 3rd generation datasheet and the socket land map, page 98

 

 

From the LGA1155 datasheet, pinout diagram, lower right quadrant, page 91

 

 

 

What is DRAM additional swizzle for READ and WRITE? Check if cpu is capable of 1300 with this option enabled. Try bigger value for Latency Boundry.

 

Swizzles were at auto, but I thought they are RAM related, not CPU IMC. The memory is capable of running tight timings 1300 8-11-7, but I tried with slack timings as well. Even tried loading mem profiles from bios.

I set LB at 6 usually, then tighten up further in windows to 1-1-1-1-4-4-1-4-4-0 when needed. Now I got it up to 8, 9 and 14 - no difference.

Edited October 2, 2012 by I.nfraR.ed

[I.nfraR.ed]

Board posts and I'm able to load windows at DDR3-2800 10-12-10 dual channel loose BBSE profile, VCCIO 1.15V and run 1M, so it's not the IMC.

At 2933 divider I get 55 code, but I don't have memory capable of this frequency, at lest not on air.

 

Hmm, ok... probably something got wet, but I dried out the board thoroughly like I always do.

Now it worked without any special setting, just ones I use with all cpus.

 

 

Anyway, going to change with M5F.

Edited October 2, 2012 by I.nfraR.ed

[Vivi]

Hey man!

 

Some imc's go really bad on cold. specially these 3218b cpu's. only from the newer cpu's have i seen imc's actually get better on cold.

 

Have you tried checking your max memory clock on -196? Might have lost some MHZ due to cold

[I.nfraR.ed]

This one - no. Only pifast and 1M at ~1300, but when I tried 32M it was crashing on the initial loop.

On the other 3218B cpu I had no problem at 1300 on cold.