DFI UT X58-T3EH8 continued - Part 2

[Razer(x)]

I agree, depending of the CPU-batch, the QPI bus can be the problem too, but that's relatively easy to diagnose/fix. On the other hand, many RAM-sticks can limit a good CPU-batch to run at BCLK210 (1680MHZ RAM). This is not so easy to diagnose and people are reporting some "strange" instability problems (it can pass the Memtest, LinX and PrimeBlend, but fails in SuperPI, or other "strange and unheard" problems).

It is not easy to generalize, but I believe many of these "strange" instability problems are rather related to "too high BCLK for the RAM" than for CPU/QPI/Uncroe (but I agree that this problem exists too). After "the dead of the hyper", there not many sticks in the marked that can run stable @1680MHZ (even with unsafe Vdimm).

An easy way to diagnose the QPI-problem can be to run LinX/PrimeBlend with QPI fast mode, disabled.

My problem is, for example, that i can't get the pc stable with uncore >3000

[Sam_oslo]

My problem is, for example, that i can't get the pc stable with uncore >3000

If Uncore is your "only" problem, then you can solve it easily. You haven't post any detail screen-shot yet, and it is not easy to guess the rest from your short posts either.

I'm afraid your problems are a bit more than that, but if that's the only problem, then you can use x6 multi on RAM, and have 12x or 14x Uncore, and still be under 3000 even at BCLK210. 14x210=2940 < 3000. We ended up with RAM-settings again?

[Razer(x)]

If Uncore is your "only" problem, then you can solve it easily. You haven't post any detail screen-shot yet, and it is not easy to guess the rest from your short posts either.

I'm afraid your problems are a bit more than that, but if that's the only problem, then you can use x6 multi on RAM, and have 12x or 14x Uncore, and still be under 3000 even at BCLK210. 14x210=2940 < 3000. We ended up with RAM-settings again?

In fact now i'm stable at 200x18 with uncore 2800mhz with 1.24v (the lowest vtt, my mobo has a positive drop of 0,03v)

Ok, i don't really care about the uncore, but 'cause of this mobo i can't get my pc stable with ram >1400 mhz

[Sam_oslo]

What about this one?

I think the main problem to run high RAM frequency is strictly related to uncore frequency and quality of IMC of CPU. There are a lot of RAM in the market that are able to run at high frequencies with really low voltages (all the kits equipped with elpida BBSE).

I won't trust this one for 24/7 use, either the stability (because you have tested the RAM all the way, at least on this screen shot) or the reliability/durability (because your are using VVT>1.35v). But yes you are right, when you get to high speeds (around 2000MHZ) then IMC plays much bigger role than it plays around 1700MZ. You need good CPU-batch to run Uncore >4200MHZ.

I know some sticks can run >1600MHZ, but most of these who are reporting about problems here, are running those "god old first generation 1600MHZ" RAM.

In fact now i'm stable at 200x18 with uncore 2800mhz with 1.24v (the lowest vtt, my mobo has a positive drop of 0,03v)

Ok, i don't really care about the uncore, but 'cause of this mobo i can't get my pc stable with ram >1400 mhz

Maybe your RAM (or CPU or other settings which i don't know much about) are limiting your MB? Uncore is sitting in your CPU, and is more dependent on your CPU-batch than your MB. In these speeds will the MB has almost nothing to say about your Uncore speed.

[Razer(x)]

Maybe your RAM (or CPU or other settings which i don't know much about) are limiting your MB? Uncore is sitting in your CPU, and is more dependent on your CPU-batch than your MB. In these speeds will the MB has almost nothing to say about your Uncore speed.

I don't think so. This kit can easily do 2000mhz CAS8 with 1.65v (2h of memtest passed)

[eva2000]

I don't think so. This kit can easily do 2000mhz CAS8 with 1.65v (2h of memtest passed)

memtest86+ v2.11 DOS or windows HCI memtest on DFI UT X58 is weak and not entirely accurate for stress testing memory and stability real tests for stability are hyper pi v0.99b 8x 32M for 64bit systems, 8x 16M for 32bit systems and Prime95 custom large FFTs or blend set at 5400MB for 6GB cofigs. But memtest86+ v2.11 is useful still for figuring out what voltages and settings are lacking or need tuning, depending on which test the errors are showing up at - http://i4memory.com/f79/memtest86-v2-11-a-6/

Also when talking about IMC/ram dependence don't discount the bios itself and how it interacts with different types of cpu and memory type/brand pairings.

Personally, I find once you figure out the memory side of things, cpu side falls into place much more smoothly. I'm able to repeat almost all my memory and cpu clocks I managed on DFI UT X58 on a completely new motherboard, Asus P6T6 WS Revolution once I figured out the memory site and bios settings/voltages needed for that board. Still early days on new board. But believe me when I say memory is the key to cpu clocks/stability as well.

[Sam_oslo]

I find HyperPi 32M to be the most "picky" on RAM/IMC too.

I have been trying today (based on my @4GHZ(19x211) 1.15v, 11 & 8 hours Primestable setup), and event tho I have had many successful Superpi 32M and Prime-smallFFT runs up to BCLK220, but Hyperpi 32 will eventually fail over BCLK212, no matter what i could do so far.

At BCLK212 it will run about 8-10 passes, and then 1 core drops out. At > BCLK213, the core drops out after 1-2 passes. Raising the VVT or loosing the RAM-timings didn't help . Even tho this setup has passed several hours Prime smaFFT at 4.1GHZ(216x19) 1.168v, but BCLK212 seams to be the BCLK-wall for RAM/IMC-stability at x19 multi on my setup.

I have been experimenting with QPI:Uncore 1:1 at x18 lately, and it was good to see it passing Hyperpi 32M.

@4GHZ (211x19) 1.15v , Hyperpi 32M:

[v_parrello]

I won't trust this one for 24/7 use, either the stability (because you have only the tested GPU-RAM) or the reliability (because your are using VVT>1.35v). But yes when you get to high speeds (around 2000MHZ) then IMC plays much bigger role than it plays around 1700MZ. You need good CPU-batch to run Uncore >4200MHZ.

I know some sticks can run >1600MHZ, but most these who are reporting about problems in this treats are running those "god old first generation 1600MHZ RAM".
....

Sorry what does mean your sentence "because you have only the tested GPU-RAM"? I have tested CPU and RAM not GPU and RAM (OCCT was CPU test with large data set for RAM, and prime was blend test for CPU and RAM).

The test demonstrates RAM could easily run at 2100 MHz with low vdimm voltage, obviously is not for 24/7 use because of high VTT voltage.

Then do you agree with the fact that the problem are not RAM but CPU uncore and IMC?
RAM are able to go over 2100 MHz easily and to be stable, the problem is CPU that are not able to manage so high RAM frequencies because of high uncore and Integrated Memory Controller (IMC) quality that need high VTT voltage to be stable.

The IMC always plays bigger role: there are CPU that are not stable (with prime, linx and similar application), with 3200 Mhz uncore and need a lot of VTT to be stable. Is not a RAM question. The same kit of RAM could be not stable with low VDIMM and low VTT with a CPU at 1600 Mhz and 3200 Mhz uncore, with another CPU the same kit is stable with same settings.

[Sam_oslo]

Sorry what does mean your sentence "because you have only the tested GPU-RAM"? I have tested CPU and RAM not GPU and RAM (OCCT was CPU test with large data set for RAM, and prime was blend test for CPU and RAM).

The test demonstrates RAM could easily run at 2100 MHz with low vdimm voltage, obviously is not for 24/7 use because of high VTT voltage.

Then do you agree with the fact that the problem are not RAM but CPU uncore and IMC?
RAM are able to go over 2100 MHz easily and to be stable, the problem is CPU that are not able to manage so high RAM frequencies because of high uncore and Integrated Memory Controller (IMC) quality that need high VTT voltage to be stable.

The IMC always plays bigger role: there are CPU that are not stable (with prime, linx and similar application), with 3200 Mhz uncore and need a lot of VTT to be stable. Is not a RAM question. The same kit of RAM could be not stable with low VDIMM and low VTT with a CPU at 1600 Mhz and 3200 Mhz uncore, with another CPU the same kit is stable with same settings.

Sorry I misunrestood/exchanged your OCCT-shot, but edited my sentence now. I sow your high VVT and and got disturbed.

This is impressive speed, and is always fun to bench/test at high speeds, but I would run at leasts 100% of 8 instance memetest (or something similar) to make sure it is stable for 24/7 use. And I'm glad that you are not pushing that much VVT to your "expansive" sticks for 24/7.

As I have already said, The IMC will play a bigger role at high DDR3-speeds. Are you suggesting that most instability are most likely related to IMC?

[v_parrello]

...
Are you suggesting that most instability are most likely related to IMC?

For sure!
RAM are able to run at specified frequency and timings declared, the problem is that IMC CPU is not able to make the same.