Join Date: Jul 22 2004
Location: Brisbane, Australia
Initial notes, findings and tips DFI UT X58-T3EH8
This is a copy of the original post here
so it's easier to view in it's own dedicated thread.
Initial notes, findings and tips
DFI UT X58-T3EH8
I'll continually update this list of notes, tips etc as I come across them.
- Finding max BCLK on DFI UT X58
- Elpida 50mm Hyper ICs are found in Corsair Dominator GT 1866/2000 7-8-7-20/8-8-8-24 and Kingston DDR3-2000Mhz 8-8-8-24 triple channel kits. These are so new that some weird issues have come about for single super pi 32m stability. You can read all about this here. Update: new 3/18 and 4/28 beta bioses have improved Elpida based IC clocking tremendously see here. Along with Corsair Dominator 2000C7 review here.
- 2x i7 920 3836A756 C0 cpus compared: Note that NOT all Core i7 cpus are created equal! This 2nd i7 920 3836A756 is noticeably different from my 1st i7 920 of same batch in various ways,
- It handles way lower cpu vtt voltages needed and clocks to a higher bclk than 1st i7 920
- But when it comes to high mem clocks >DDR3-2000Mhz and high QPI frequencies, it seems it has a more strict relationship between cpu vtt and vcore once above 1.48v cpu vtt and likes vcore to be 0.02-0.08v close to it. My first i7 920 didn't have so a strict relationship allowing greater distance between cpu vcore and cpu vtt voltage.
- If vcore isn't as close to cpu vtt for this 2nd i7 920, then >1.48v cpu vtt will introduce slight instabilities at high mem and qpi frequency clocks. For example, inconsistent super pi 32M stability between 32M pi runs, not booting on bios save/exit after setting higher cpu vtt leaving vcore too far from cpu vtt, BSOD on windows loading and even Memtest86+ v2.11 test #2 and #3 errors and irregularities.
- Update: Now looking back at my first overclocked screenshot on this 2nd i7 920, and the required vcore and cpu vtt voltages must of been a preview of what this cpu needed. Vcore set 1.30625v (SG = 1.32v) and CPU VTT set 1.31 (SG = 1.31).
- Each cpu will have optimal cpu vtt requirements for bclk scaling and remember as you increase cpu multiplier the same bclk which was stable on lower cpu multiplier, might need more cpu vtt on higher cpu multiplier. For example, 19x200bclk might be fine with 1.37v cpu vtt, but 20x200 bclk might need 1.39-1.47v cpu vtt, 21x200bclk might need 1.39-1.55v cpu vtt. All depends on your individual cpu. Also cpu vtt voltage needed will also depend on your uncore frequency. The higher QPI frequency the more cpu vtt volts is needed. The higher the uncore frequency more vdimm is needed.
- 2x i7 920 3851A368 D0 cpus compared: Seems D0 handle better in terms of Bclk to CPU VTT voltage ratios as seen here. Unlike, C0 step cpus I tested, D0 step like O.C. Shut Down Free set to disabled for more stability as opposed to C0 which liked it enabled.
- TDC/TDP & TM Protection (Turbo mode effects): In CPU Feature section of bios the Set VR Current Limit Max and Thermal Management Control options control TDC/TDP and other temperature related throttling for Core i7 cpu.
- TDC = over current protection which default Vcore OCP max is 140A (for CPU PWM IC). At >70A, TDC is tripped, it will disable Turbo mode - thus you may see your cpu multiplier drop by 1x or 2x.
- TDP = thermal design power according to DFI is tripped at around >80W which will also disable Turbo mode
- Set VR Current Limit Max = Enable to disable both TDC and TDP protection functions thus allowing Turbo mode to run without any TDC/TDP protection.
- Thermal Management Control = Tripped when core temp is >90C, Turbo mode will disable. At >105C core temp, will force system shutdown.
- Only when 965 cpu is installed are 2 additional options for TDC/TDP displayed for TDC Enable: x TDC Limit: and TDP Enable: Disabled x TDP Limit: 0
- Freezing/Laggy Mouse & USB/Floppy Bios Flashing: Usually caused by incorrectly reloading a older bios based saved profile from ABS application onto a newly updated bios flashed motherboard. DFI have 01/14 beta bios with an accompanying 01/14 ABS Profile for those that need to fix this issue. Full details here. Solution here.
- CMOS Reloaded & ABS Saved Profiles: My advice is don't go unknowingly loading other folks profiles ABS saved bios settings profiles. Also DO NOT load previous bios version's saved profiles after you have updated your bios - I always save a new version of a CMOS Reloaded profile after I do a bios update based from Load Optimised defaults after bios update.
- Bios Flashing: Best method for flashing bios on DFI boards is to use a bootable USB Flash drive as per method outlined on DFIClub forums. I've used this method on DFI UT P35-T2R, DFI, LT X38-T2R and DFI UT X48-T3RS as well without problems. DFI UT X58-T3EH8 worked fine as well.
- Thermalright Ultra 120 Extreme Push-Pull config: - been asked many times how I set up my push-pull config with TRUE120. The pull fan on TRUE120 is directed on the Flame Freezer heatsink and the push fan is positioned lower down so air flow goes under TRUE120 heatsink towards PWM area See photo here.
- Smart Guardian Temps: Ignore SG cpu temps as they under report by a country mile! ~10-18C under reported at idle and ~15-35C under reported during load compared to coretemp/realtemp and everest ultimate OSD. The rest of reported temps and voltages in SG seem to be okay. Haven't put a multimeter to the measure points yet located here.
- Memory tip #1: If you're having problems reaching a particular memory frequency overclock in dual or triple channel, try playing musical chairs with memory modules and test them in different dimm slots for the same channel. So label each of your 3 memory modules with a bit of masking tape #1, #2, #3, and test them in different combinations in dimm slot 2+4+6 - you may find a particular combination allowing much easier and higher memory overclocks
- Memory tip #2: Play with DRAM PWM Switch and VTT PWM Switch frequencies as well. My 2x1GB OCZ PC3-14400 Platinum D9GTR, 2x1GB Cellshock PC3-14400 D9GTR and G.Skill PC3-12800HZ D9JNM based modules handle better with like Reduce 15% over Nominal setting but my Samsung HCF0 based G.Skill PC3-12800HZ and Corsair DDR3-2133C9DF like Nominal. For my Corsair Dominator GT 2000 CAS7 Elpida 50mm Hyper based ICs, they handle best with Nominal as well but some instances Increase 15% worked well too.
- Memory tip #3: Subtiming for tWTR in bios refers to Memset/CPU Tweaker's Write to Read (same rank) value and usually will have a displayed off set value of between +4 to +14 from what is set in bios depending on ram main timings, mem multiplier, mem divider and mem frequency used. So set tWTR in bios to 18 may display a Write to Read (same rank) value of 22 @DDR3-2263Mhz 9-9-9-24 on 10x mem multiplier - a +4 offset. But set tWTR to value of 5 may display Write to Read (same rank) value of 16 at DDR3-2000 8-8-8-21 with 10x mem multiplier - a +11 offset. Set a tWTR of 8 @DDR3-1600 7-7-7-18 with 8x mem multiplier = Write to Read (same rank) value of 18 which is a +10 offset.
- DRAM Voltage Safety: Following the guidelines, for safe high vdimm usage, you need to keep CPU VTT (Uncore) voltage within 0.5v of DRAM memory voltage. Seems the bios has a clever feature, if you try to initially set DRAM memory voltage to >0.5v difference from CPU VTT, the bios will automatically increase CPU VTT voltage to ensure there's always a 0.5v difference between DRAM memory voltage and CPU VTT!
- Memory tip #4: Seems DFI bios tWR value doesn't correspond to Memset/CPU Tweaker's displayed tWR value but instead corresponds to the displayed tWTP value. Haven't figured out the offset value but on 8x mem divider and 7-8-7-20 1T timings, setting tWR to 16 = tWTP of 28, tWR to 14 = tWTP of 26, so looks like there's at least a +12 offset.
- Memory tip #5: I'm finding the optimal difference for DRAM memory voltage and CPU VTT to be between 0.24v and 0.31v for 2x1GB Corsair DDR3-2133C9DF Samsung HCF0 modules. I've only used mismatched dual channel kits in triple channel config, but I find as your use more modules or at higher memory frequencies the optimal difference get's wider i.e. with 3x1GB HCF0/D9GTR difference between 0.3v and 0.45v seems better. 3x2GB Corsair Dominator GT Elpida 50nm based ic memory seems to like 0.20-0.28v difference most of the time hence why they generally require higher CPU VTT than other Samsung/Micron ICs. Of course it depends on the bclk and QPI frequency used since higher QPI frequency requires more CPU VTT (uncore volts). Finding that balance allows you to clock memory higher as well as use less Vcore for stability.
- Memory tip #6: Incorrect detection of total memory available: Are your 6GB or 3GB kits being detected as 1GB, 2GB, 4GB in bios/memtest86+ etc when you first install them or when you start to overclock ? This could happen for various reasons other than you having a faulty memory module or two. The reason could be due to non-optimal VDIMM/CPU VTT balance as mentioned in previous memory tip.
Specific example by what I mean: If I have a 3x2GB DDR3 Tri channel kit of memory rated for DDR3-1600 at 1.65v. If I left CPU VTT at default the board sets say 1.2v CPU VTT, that means there's a 0.45v difference between CPU VTT and VDIMM. With this config say only 4GB is detected. I proceed to bump CPU VTT to 1.3v where there is a 0.35v difference. Reboot system and guess what 6GB is detected. So okay this particular Core i7 + 6GB DDR3-1600 kit liked a difference of 0.35v. Now say you push BCLK and Memory clocker higher, this will need more CPU VTT volts for QPI frequency and more VDIMM for memory clocks and Uncore frequency to maintain stability. You may find pushing just VDIMM from 1.65v to 1.75v resulted again in only 4GB detected - well that could be due to CPU VTT - VDIMM difference being pushed out of it's optimal range at 1.75-1.3 = 0.45v difference. So you raise CPU VTT from 1.3v to 1.4v to maintain a 0.35v optimal difference and you may find you have more stability and 6GB is now detected properly.
Now the problem is every Core i7 memory controller will have a different optimal CPU VTT/VDIMM relationship with every type of memory out there - even down to individual module's PCB design (just my theory). Even same brand/model memory may differ in what it's optimal CPU VTT/VDIMM difference is ranging between 0.01v to 0.50v (max safe difference).
To complicate things, what is set in bios volts does not necessarily equal exact same real volts if you use a multimeter to measure CPU VTT and VDIMM. So 1.2v CPU VTT set in bios could be 1.16v or 1.23v real. And under load CPU VTT could either droop or rise slightly i.e. +/- 0.01-0.02v.
This isn't the only reason for mis-detection of memory size, it could be one memory slot/channel is faulty as well or one memory module in tri channel kit is faulty.
Or due to memory module PCB differences, that the Core i7 memory controller likes the modules in a tri channel kit in a particular arrangement in the memory slots. You may find the 1st memory tip above in playing musical chairs with your 3 memory modules, that one or more combinations allow proper detection of your 3GB or 6GB kits
- Multimeter Voltage Read Points: These are located on bottom right corner area between power/reset connector header and last green SATA ports. Photo here.
- Dimm slot distance measurements:
- Distance between each of the 3 channels (1 green + 1 yellow) is ~3.5mm
- Distance between dimm slot 2 and dimm slot 4 with Corsair Domainator DDR3-2133 modules installed is ~14mm
- Distance between dimm slot 1's retention lever and dimm slot 5's retention lever is ~50mm
- Approximate dimm slot width capacity up to ~7.2mm - so that's how wide your memory modules can be in terms of thickness of heatspreader with modules side by side i.e. 6x1GB or 6x2GB configurations
- Corsair Domaintor DDR3-2133 heatspreader thickness ~6mm
- G.Skill PC3-12800HZ heatspreader thickness ~7mm
- Gary from Anandtech.com reported 12/31 bios with 10x multi 1333mhz dram frequency has a bug which requires more vdimm than prior bioses to run, and DFI is working on the fix. Try 12/18 bios or 01/06 and same 10x multi for mem.
- Memory LowGap: is for mapping local memory (add-on-cards) to system memory address size. It most like AGP port Aperture size or PCIE1gen 250M fixed mapping size. Extend it can solve when if "add-on cards" request a bigger size as like we would like to enable Qual SLI on NV new 295, it must be extend over 2048M. Can confirm that with 2x HD4870x2 in Crossfire I had set my LowGap to around 2176 at least. With Highpoint 3520LF PCI-E x8 256MB cache controller added into HD4870x2 Crossfire, I needed at least LowGap of 2560 otherwise windows boots to a blank window and doesn't go any further. As to how this effects windows available memory, with single HD4870x2 installed and 3x1GB memory, if I set LowGap to 2560, WinXP Pro SP3 32bit sees 1.5GB available memory. If I set LowGap to 1024, WinXP Pro SP3 32bit sees 3.0GB available memory IIRC.
- S3 Sleep: confirmed to work fine with Vista Ultimate SP2 Beta 64bit - takes about 8 seconds to resume from S3 Sleep to hit windows desktop (including time to enter my password to log back in!)
- Added 01/01/09 - Updated versions greater than setfsb 2.1.88 for DFI X58 (no need to check the Ultra checkbox )
- Vcore Droop: For Vcore droopage, according to CPUZ v1.49:
- Vdroop Enabled = 0.016-0.032v droopage under load
- Vdroop Disabled = 0.00-0.008v droopage under load
- Uncore frequency to DRAM frequency ratio: DFI UT X58-T3EH8 bios automatically scales UNCORE frequency in bios by a ratio of 2:1 as you increase your DRAM memory frequency in bios. For example, set DRAM memory frequency at DDR3-1067Mhz and bios automatically moves UNCORE frequency to DDR3-2133Mhz. You can move the DRAM frequency down again and it will keep UNCORE frequency as previously set if you want to try UNCORE frequency greater than 2:1 ratio to DRAM memory frequency. Example bios screen shot.
- With Vcore set at AUTO, with Turbo enabled vcore defaults to 1.24v in bios, with Turbo disabled vcore defaults to 1.15v in bios.
- Turbo mode: For i7 920 stays at 21x for load tests in winxp sp3, for Vista i just installed (configuring now) if you leave default power management profile enabled it drops to 12x multi at idle and 21x at load as EIST needs to be enabled for Turbo on DFI. But set Vista power management to high performance and it stays at 21x as well under load. Some Win7 OS folks having 21x Turbo multi issues with i7 920 and W3520 cpus - read here for work arounds and an outline of the issue.
- O.C. Shut Down Free automatically gets set to disabled when CPU VTT set to >1.55v
- Much easier bios replacement procedure now that DFI has a bios holder socket.
- LAN1 and LAN2 in bios don't correspond to same LAN1 and LAN2 in manual diagram. Disabling LAN2 in bios will disable LAN1 port from manual's diagram
- The manual has neglected to explain the memory configuration combinations for single, dual and triple channel and it's same for all X58 boards.
- High PWM (System) Temps tip #1:It does seem the NB heatsink has a heatpipe that extends to the mosfet heatsink from these 2 photos of the cpu socket area - side view and top view. It would also explain why, adding a fan over NB heatsink and dropped SYSTEM (and PWM) temps at idle from 57C to 44C and load temps from 70C to 55C for SYSTEM and 46C to 39C for CHIPSET temps as seen here. Update: Got confirmation Smartguardian SYSTEM temp = PWM temps and CHIPSET temp = NB temps.
- High PWM (System) Temps tip #2: As mentioned above, if you're experiencing high SYSTEM (PWM) temps under cpu load, it's highly recommended to have a fan over NB and PWM areas. But it's also highly recommended to install the included FLAME FREEZER heatpipe based heatsink with decent TIM paste. I used Artic Ceramique on my FLAME FREEZER heatpipe which attaches to I/O connector area.
- USB keyboard/mouse/devices - system hang: If you experience system hangs with USB devices connected, try changing USB power source from 5V to 5Vsb as per image below.
Received some clarification on the PCI-E slot assignments for multi-GPU configurations.
- PCI-E #1 and #4 are x16 when nothing is installed in PCI-E #3.
- When PCI-E #3 is occupied, then PCI-E #1, #3, #4 become x16/x8/x8. This applies to both CrossfireX and SLI.
Example, with 2x HD4870x2 double slot video cards in PCI-E #1 + #4, you'd block off access to PCI-E #2 x4 slot and PCI #2 slot but still have access to PCI-E #3 (yellow x16 slot - photo below has Highpoint RR3520LF PCI-E raid controller installed) and PCI #1 which is empty below
Based on 02/06 beta AND 02/17 beta
Genie BIOS Setting
Set VR Current Limit Max: Disabled
Thermal Management Control: Disabled
EIST Function: Disabled
CxE Function: Disabled
Execute Disable Bit: Disabled
Virtualization Technology: Disabled
TDC Enable: Disabled
x TDC Limit: 0
TDP Enable: Disabled
x TDP Limit: 0
***** Logical Processor Setting *****
Intel HT Technology: Enabled
Active Processor Cores: All
Channel Interleave Setting: AUTO
Rank Interleave Setting: AUTO
Memory LowGap: 1536M
DRAM Command Rate: AUTO/1N/2N etc
CAS Latency Time (tCL): 9
RAS# to CAS# Delay (tRCD): 9
RAS# Precharge (tRP): 9
Precharge Delay (tRAS): 24
REF to ACT Delay (tRFC): AUTO
Write to PRE Delay (tWR): AUTO
Rank Write to Read (tWTR): AUTO
ACT to ACT Delay (tRRD): AUTO
Row Cycle Time (tRC): AUTO
Read CAS# Precharge (tRTP): AUTO
Four ACT WIN Time (tFAW): AUTO
O.C. Shut Down Free: Enable O.C.S.D.F
CPU VID Control: AUTO
CPU VID Special Add Limit: Enabled
CPU VID Special Add: Auto
Vcore Droop Control: Enabled
DRAM Bus Voltage: 1.605v
DRAM PWM Switch Frequency: Nominal
DRAM PWM Phase Control: 2 Phase Operation
CPU VTT Special Add: AUTO
CPU VTT Voltage: 1.21v
VTT PWM Switch Frequency: Nominal
VTT PWM Phase Control: 2 Phase Operation
CPU PLL Voltage: 1.80v
IOH/ICH 1.1V Voltage: 1.11v
IOH Analog Voltage: 1.10v
ICH 1.5 Voltage: 1.5v
ICH 1.05V Voltage: 1.05v
DIMM 1/2 DQ/DQSTB Bus VREF: -00.0%
DIMM 3/4 DQ/DQSTB Bus VREF: -00.0%
DIMM 5/6 DQ/DQSTB Bus VREF: -00.0%
ADDR/CMD VREF Control: Disabled
x DIMM 1/2 ADDR/CMD Bus VREF: 110
x DIMM 3/4 ADDR/CMD Bus VREF: 110
x DIMM 5/6 ADDR/CMD Bus VREF: 110
CPU QPI Drive Strength: Normal
IOH QPI Drive Strength: Normal
Exit Setup Shut down: Mode 2
O.C. Fail Retry Counter: Enabled
O.C. Fail CMOS Reload: Disabled
PPM Function: Enabled
Turbo Mode Function: Disabled
1 core Max Turbo Ratio: 22x
2 core Max Turbo Ratio: 21x
3 core Max Turbo Ratio: 21x
4 core Max Turbo Ratio: 21x
CPU Non-Turbo Clock Ratio: 20x
* BCLK/UCLK/QPI Controller Settings *
QPI Control Settings: Enabled
QPI Link Fast Mode: Enabled
QPI Frequency: AUTO/4.8GT/s/5.866GT/s/6.4GT/s
CPU Base Clock (BCLK): 133 Mhz
Boot Up CPU Base Clock: AUTO
PCIE Clock: 100 Mhz
DRAM Frequency: 6x/8x/10x/12x/14x/16x Memory multipliers
UnCore Frequency: 2x to 32x Uncore multipliers (optimal to keep Uncore at 2x DRAM Frequency)
CPU Spread Spectrum: Disabled
PCIE Spread Spectrum: Disabled
Multimeter Measured Voltages
Using the provided multimeter measure points on the DFI UT X58-T3EH8 pictured below.
I will compile a list of bios, in memtest86+ and in windows idle/load voltages and update the table as I have time to test each of the bios set voltages.
I'll start with basic table for AUTO vcore, 1.590v VDIMM, 1.21v CPU VTT with rest of the voltages at default with VDroop Enabled (default intel spec'd droopage).
Then manually set vcore and VDIMM to 1.20v and 1.65v respectively with VDroop enabled (intel spec) and VDroop disabled (reduce intel spec droop).
While most volts remained fairly spot on, 3 voltages varied:
- VCORE with VDroop enabled. Real voltage seems to undervolt. With AUTO vcore under volt by 0.013v at idle and a further 0.015v under load (or 0.028v from what is shown in bios. But manually set vcore at 1.200v, idled spot on at 1.200v but under load with Vdroop enabled drooped to 1.178v (0.022v droop). CPUZ reports idle 1.216v and 1.200v load while DMM real voltage reports 1.200v idle and 1.178v load. So under load, CPUZ over reports vcore by 0.022v and SG over reports by 0.012v.
- VCORE with VDroop Disabled. Here both bios and Smart Guardian reported vcore is over reported ALOT! 1.20v bios set vcore results in bios and SG reporting 1.23v idle vcore, CPUZ reports 1.232v idle. Under load SG and CPUZ report 1.21v and 1.216v respectively. However, real vcore measured via DMM is 1.206v idle and 1.203v load. So under load, CPUZ over reports vcore by 0.013v and SG over reports by 0.007v.
- DRAM memory voltage seems to overvolt my 0.005v from what is set in bios.
- CPU VTT voltage undervolts by 0.012v to 0.014v at idle but jumps 0.009v higher under windows load but still is 0.005v under volted under load compared to what is set in bios. Note: for the requirements for VDIMM to be within 0.5v of Uncore (CPU VTT), it didn't change if you look at under load voltages. Bios reports 1.66v vdimm and 1.21v CPU VTT = 0.45v difference. DMM Measured 1.655v vdimm and 1.205v for CPU VTT = 0.45v difference under load or 0.459v difference at idle.