Seeking second-opinions, seat-of-the-pants comment: 4-slot RAM configuration

[BonzaiDuck]

OK here's the story.

I have this absolutely great system -- 2700K @ 4.7, G.SKILL DDr3-1866 "GXM" 2x8GB=16GB @ 9-10-9-28 CR=1. SSD dual-boot Win7/Win10. 2x GTX 970 OC'd to core-clock 1480, memory 7500 in SLI.

Never any problems; event-logs all blue; never misses a beat or a lick.

I decided I wanted just a bit more RAM than 16GB, because at one time my RAM usage went up to 75-80%. I'd placed the order for more RAM before I discovered that unloading a program would get the memory settled back to about 55-60%. The reason RAM usage is this high either way derives from using 4GB for caching disks.

The "GXM" RAM is labeled as "RipJaws X." But there are several RipJaws X kits: some now coded as "GBXL," others coded as "GBXM."

I would have been better off, I think getting a 2x2GB GBXM kit with the same timings and speed: 9-10-9-28 and 1866. Instead, I picked a GBXM kit of 2x4GB with tighter timings: 8-9-9-24 @ 1866. Knowing GSKILL, I was sure I could simply loosen the timings on the 8GB kit to match those on the original 16GB kit.

And -- I can. But I was hoping to run these at CR=1. And so I set the timings for that command rate.

Then I booted into HCI Memtest 64. As soon as it begins to run the "long" test for the first 100% coverage, there are some 152 errors that pop up.

So I imagined a possibility that the VCCIO and possibly even the VDIMM were too low to sustain CR=1. Went into BIOS; changed CR to 2. Now the HCI Memtest-64 program seems to show the RAM behaving as it ought, but it's going to take time to test these. Perhaps would have been better to test the new modules separately, but I have a broken latch in a socket for one of the original modules, the stick needs to be levered out with a screwdriver, and I figured to just install all four modules, test, and cross my fingers.

I would only guess this is a likely outcome for filling all slots: you might need to bump the voltages slightly to obtain CR=1.

At this point, I can't tell if I should send the new modules back for RMA replacement. I should just wait and see how a few passes at HCI Memtest-64 goes.

Any thoughts?

 

[BonzaiDuck]

Somebody could respond to this, but I'm not sure it's necessary.

After the 150 errors, I recall going back into BIOS and noticing that the RAM-speed -- an item separate from a chosen "XMP profile" with a selected speed -- was set to "Auto." It needs to be fixed to the desired speed setting, while Auto would allow the RAM speed to increase with the CPU overclock.

So it may not have been the command-rate after all. And now I'll have to test all over again, just to confirm that CR=1 is viable at the same voltages.

Sorry to distract everyone, if anyone was distracted at all.

 

[zir_blazer]

As a general rule, take that if you are going to use all 4 slots, you are going to be force to run Command Rate 2T. If I recall correctly, the Command Rate is NOT a DRAM Timming, it is instead for the Integrated Memory Controller, which you are pushing harder with twice the electrical load than with only one module per channel.
I don't know how much lower IMC Frequency or more Voltage would compensate to force for 1T Command Rate. But I doubt its worth it.

 

[BonzaiDuck]

As a general rule, take that if you are going to use all 4 slots, you are going to be force to run Command Rate 2T. If I recall correctly, the Command Rate is NOT a DRAM Timming, it is instead for the Integrated Memory Controller, which you are pushing harder with twice the electrical load than with only one module per channel.
I don't know how much lower IMC Frequency or more Voltage would compensate to force for 1T Command Rate. But I doubt its worth it.

I think you're right. But there had been exceptions with G.SKILL kits I'd had. For one, I had two 2x4GB RipJaws "GBRL" DDR3-1600 kits -- the red ones -- spec'd with timings of 9-9-9-24. I and another member here were able to fill all four slots with two such kits and run them at 1866 with 10-10-10-30 (or 32 -- vague recollection). And with the voltage at the 1.5V spec, you could run them with CR=1. I didn't have to run the VCCIO or IMC voltage up beyond 1.12V, and the recommended limit was 1.2V. I think with a little more testing, I might have had stability with 1.10 or a little lower.

But that was the case of two identical kits. In this situation, while the model code and heatspreader design are "identical," I had one kit spec'd at 9-10-9-28 for 2x8GB, and the second kit at 8-9-9-24 for 2x4GB. I even fixed the secondary timings to what (I thought) were the SPD values of the larger kit, but CR=1 is just not in the cards for that. YEt, the XMP profiles on these kits may still have the motherboards pre-setting the CR to 1, even though the XMP "profiles" show "2N."

I may RMA-refund the 8GB GBXM kit with the CL=8. For <=$25, I ordered a 2x2GB GBXM kit that matches the timings of the 2x8GB kit. If I can't get CR=1 with that combination, well -- it's not all that important. Flawless operation and reliability are important.

I really don't want to use 32GB of RAM filling all four slots. I can't use all of it, and the testing would take forever. I think 20GB is probably just right for my needs.