Offler

How much watts need this board? Since i changed the PSU (due RMA) the system is unstable even on default frequency. It is not the graphics which i suspected, because the GPU OCCT test passed, only CPU tests are failing. Maybe the CPU power cable? i use only 4 pins...

Eva please could you test your x48-t3rs with the PCI-Spread spectrum and Jmicron IDE controller? I need to know if my board is screwed or there is some problem with the boards in general.

Edit:
Just connected extra power to the PCI-E onboard (the FDD connector). Even the Ac 550w is now stable. The symptoms are very very strange. 3.3v lowered to 3,0v, but system has not crashed in 5 minutes after full OCCT burn test. Before it was the CPU which became unstable, but now extra power for PCI-E slot makes CPU much more stable.

It works like a charm I maybe try some other tests now but seems very promising in attempts to OC the quad core

Offler

Now i have my Corsair PSU back home so i can continue with my work.

Since i connected both additional power connectors many things changed. Right now i have switched the ram modules from bank 1 and 3 to 0 and 2. Now the memory works on its default voltage under heavy load without any trouble. Trying to work with 8gb ram could be now very complicated while the banks have very different characteristics.

The issue with the Jmicron IDE controller has been confirmed more than once. I can only recommend to disable this device during overclocking. I will try to consult this issue with Oscar Wu. Maybe he will respond to my mail about bit older board.

eva2000

X48-T3RS is packed away so can't test it sorry

Offler

Since we in Europe have summer here, and the weather changed the current temperature reached 30-35 degrees here.

The Thermal Design power of my CPU is 71,4 degrees, but under full load the core reaches temperature of 82 degrees now. The Core react on the temperature unfortunately.

For three days since I installed Corsair again i was trying to replicate stability which i had before, but before it was bit colder. The OCCT was failing with system crash, BSODs or errors. I believed that problem is with graphics, but tests with graphics only passed, only full power supply tests were failing. Then i did some other tests with LinX and all have passed, BUT the instalbility indicator with bad reports for voltages repeated. Everything was pointed on CPU or memory, but all linx tests passed, until today.

The test with 512mb of memory will not test memory very intensly, but CPU will get a bit, and single test take only 8-9 seconds and is quite accurate. Takes up 100% of cpu time for 4-5 seconds and burns up the core to 82 degrees. After just a few tests i was completely sure that CPU core is the problem now.

At all it is certain that Ram Frequency 1600mhz (or memory divider 1:2) is also bit unstable comparing to slower memory configurations and makes system overally less stable than others. this has been criticised on the X48 more than once and it is quite interesting that solution for the problem lies on CPU voltage or CPU cooling. Resulted voltage for CPU is 1,375v for now, which is much more than before. VTT voltage is still same.

Will try different fan, just have to check some Air cooled systems.

edit: Reference from other OCers.
At different forum i found 2 years old thread. The guy there had same CPU and Board, he reached 3.7ghz on CPU at 1.375v, while at fsb 400 the 1,29v should be enough for CPU. There is no mention whether he tested the CPU with OCCT or if he has CPU spread spectrum on (i think its off).

Good to know that my previous OC potential was reached by someone else at similar voltages. Definitely the CPU cooling has to be better.

eva2000

hehe as i said way back in this thread DFi UT x48-t3rs, X3360

Offler

This is the most extreme test i ever did.

Ambient temperature is 40 degrees. Voltage has been set to 1,375v. It is quite strange, but GPU still has its 84 degrees as usual. CPU runs on 76 degrees. Fan has been replaced. TV tuner runs with Windows Media Center (intense on HDD and memory).

If the system will be stable (since CPU is not as stable as i wish) it will be quite good for further OC.

Cooler is Thermolab Baram with one Noctua 12cm. Very similar to cooler you advised.

Offler

So i have now superstable settings like this:
CPU voltage 1,375v
VTT 1.21v
RAM: 1.68v (1,65 real)

FSB:400
CPU spread spectrum: 0.30%

This means that frequencies generated on CPU MUST be same as on stock frequency and do not contain any noise, or are not different. My secondary indicator - VBAT shows 3.3v at all times, and the test at 40 degrees has Passed.

CPU voltage with Spread spectrum ON: 1.375v
CPU voltage with Spread spectrum OFF: 1,275v

This difference shows how much can the spread spectrum consume, but also shows how much is CPU tolerant to different character of its own frequency. Lower CPU voltage also means lower frequency amplitudes and temperature of course.

Now to bit OCing of X48. I decided to go forward very slowly. +1% each step of FSB. That means +4Mhz on FSB and +1Mhz on PCI-E. Here are steps i actually tested. I decided to set frequency of memory to 400/1066. I am testing just the CPU and X48.

100% - 400Mhz FSB - 100Mhz PCI-E - Base and well tested frequency
101% - 404Mhz FSB - 101Mhz PCI-E - Works fine without any other settings
102% - 408Mhz FSB - 102Mhz PCI-E - NB voltage increased. VTT inreased
103% - 412Mhz FSB - 103Mhz PCI-E - Dead Frequency. System cannot POST
104% - 416Mhz FSB - 104Mhz PCI-E - Voltage to SB added.
105% - 420Mhz FSB - 105Mhz PCI-E - Dead frequency. Graphics cannot start on OS with this PCI-E
106% - 424Mhz FSB - 106Mhz PCI-E - Bit more voltages, but still works Fine with CPU spread spectrum.

CPU skew rates are still default.

Offler

Check the screens i sent. I have detected some voltage trouble at 12v line. What do you think? By the way seems as the problem which caused system to crash instantly during heavy load.

The 12v PCI-E connector was not used this time. At all i had to add bit more volts to NB - it seems that even its power source was not so stable as i expected. Now it works much better.

OCCT VS Windows Media center cross tests:

The DFI board has one hidden flaw. When the memory is set to faster rate as "More relaxed" it will most likely lead to instant crash - black screen, graphics fan running on full, and sound still working. My experience show that TV tuner can make this issue more ocurrent while playing the TV, but there was different behaviour.

Sometimes the system crashed into "normal" blue screen, or the system seemed like overloaded until i closed the Windows media center. After that i was able to open the TV once again using different app.

There are two layered reasons for crash. First is the board itself, second is the Windows media center - most likely only the SW, not the TV tuner.

Why should the board require such low memory performance?

The "Board issue"
- Senses voltage on both NB and VDIMM. When both voltages go to the lower values due higher utilization nothing can happen, or when the both values are lowered at the same time the system will crash
- Performance level: When set to "Auto" everything works better. At this situation board assigns Perf level 7 to channel 1 and Perf level 8 to Channel 2. In any other configuration the system will most likely crash.
- Also senses voltages on 12v rail and on 3.3v rail.
- Does not react with frequency. Even when lowered the system can crash due this issue.
- Senses all memory timings.
- Will be "visible" only when both CPU and GPU are fully utilized.

Attached Images

	12v1.png (36.1 KB, 1 views)
	12v2.png (35.7 KB, 1 views)