Jump to content
Corsair Community

4GB on Gigabyte GA-EP35-DS4


TomGu

Recommended Posts

I've read this report. However, if one keeps in mind that the load line calibration disabled is intended for use in non overclocked systems and is inserted as a tool for those who wish to find a stability at a lower voltage, then its use is a good thing. Keep in mind that you are overclocking your system. Also keep in mind that this setting is one that has been inserted by system board design engineers who have overclockers in mind. They are far more advanced in the realistic use of this setting and that overclockers are not looking for ten years of CPU use.

 

However, if you are one of those who is looking for the longest term use of your system then I advise you to disabled load line. Disable any and all increases of any and all voltages. Purchase JEDEC standard 1.8v DRAM and enjoy your non overclocked system for longer than the five or more years than I expect my load line enabled 4.0Ghz Xeon 3110 to last.

 

I disagree with Kris. I have seen great differences in voltage necessary to run at load with load line enabled. I have two systems that run 24/7 and which are highly overclocked and are 45nm. I have modded for VDrop and load line for Vdroop. For him to say "Clearly, removing Vdroop gains us nothing and only serves to create problems that are more serious" shows me that he is NOT an enthusiast or he would see the real world differences when enabled or when boards are physically Vdroop modded. Anyone can show how schematics are supposed to work, but that is NOT the real world. Remember the Pentium 4 Prescott? It was supposed to reach 6+Ghz and never made it past 3.8Ghz. Theory showed 6+GHz. Reality showed a far different result.

 

At any rate. I know what I see and what the differential is and you can make whatever choice you wish.

 

You may wish to research real sites of overclockers and enthusiasts who dwell in the real world of products rather than some of these blogging technoboobles who know little of reality but who are bounded in theoretics.

 

A few sites of real world enthusiasts:

 

http://www.xtremesystems.org/forums/showthread.php?t=177546&highlight=Loadline+Anandtech

 

http://www.xtremesystems.com

http://www.ocforums.com

 

And make your own choice. :D:

 

Edit:

 

Power requirements and thermals are tied together btw.

 

http://i259.photobucket.com/albums/hh293/DerekT2008/now.jpg

Link to comment
Share on other sites

  • Replies 69
  • Created
  • Last Reply

Appreciate the explanation but didn't mean to raise an irritating issue.

FWIW, the manual shows loadline calibration to be enabled by default and I have not changed it. Assume the VDrop mod you refer to is the "pencil mod"? Have not done that yet.

Would like to clarify your prior comment: "It's not a bad idea to give the MCH/Northbridge (Memory Controller Hub) a bit of juice greater than stock. Also keep in mind, that when you set to +.1v you also remove "Auto" which is an indiscriminate setting based on an algorithm and activated via DAC. This will keep your MCH at a single value which is better when overclocking highly."

The manual shows "Normal" for the (G)MCH OverVoltageControl default. I interpret "normal" to mean "no change" from the default voltage but it seems you are saying "normal=auto" which allows system algorithms to adjust the MCH (and other voltages) dynamically unless a value is assigned. Please clarify this as I prefer to "set" a value if the default of "normal" is going to allow changes without my knowledge.

Link to comment
Share on other sites

You are running your system with a fairly high overclock and are running with dense DRAM. For that reason I believe a +.1v raise of the MCH will have two characteristics. One will be to stop voltage fluctuations that normally occur on a non overclocked systems and which are not a healthy thing in an overclocked system. The electrostatic ripple effects on a non overclocked system are far different than on an overclocked system and so by setting a lightly higher static voltage, you are somewhat negating the effect. Secondly, you are allowing for a minimally higher baseline voltage which will help the memory controller when awaiting a DRAM refresh. Normal does NOT mean static. Normal means interpretive and the interpretation is based on a non-overclocked system. Which is again, why I disagree with Kris. No one other than CPU design board testing engineers can know what the actual results of the overclocked system are and Kris extrapolates. Well, extrapolation is NOT a scientific principle grounded in reality. It is based on a statistical analysis referenced on possibilities. I would say that the if --> then else extrapolation might well have multiples of "Else" and not a single threaded result. For this reason, theory will never be a surety until actually tested at the relativistic level and of which Kris has no response. However, in the enthusiast arena there is the response of thousands of users and their results. I personally prefer the thousands, nay millions of users who have demanded that their enthusiast motherboards have features that their use have shown to be beneficial. This is why many boards are modded with voltage pots by enthusiasts. It is to gain a control of the static voltage and bypass the multiple voltage identification algorithms that many chipsets now employ and which are good for non overclocked and low end mainboards with low end transistors/capacitors, etc.

 

Your board is not one of those. Your system is not one of those, therefore I personally believe that it is beneficial for you to test all your voltages and manually set them.

Link to comment
Share on other sites

Presently setting DDR2 OverVoltage Control and (G)MCH OverVoltage Control and plan to set CPU Voltage Control when moving from 3.6 Ghz to 4Ghz. Presently not setting (on purpose I thought) PCI-E OverVoltage Control and FSB OverVoltage Control but are leaving them at "normal". Do I need to set a value (even if the lowest optional value) for PCI-E OverVoltage Control and FSB OverVoltage Control to prevent the "system" from dynamically modifying them to values I do not want?
Link to comment
Share on other sites

Many say that you do not, but I set all those voltages to the minimum personally. Call me a control freak :D:

 

Presently setting DDR2 OverVoltage Control and (G)MCH OverVoltage Control and plan to set CPU Voltage Control when moving from 3.6 Ghz to 4Ghz. Presently not setting (on purpose I thought) PCI-E OverVoltage Control and FSB OverVoltage Control but are leaving them at "normal". Do I need to set a value (even if the lowest optional value) for PCI-E OverVoltage Control and FSB OverVoltage Control to prevent the "system" from dynamically modifying them to values I do not want?
Link to comment
Share on other sites

In your post #15 of this thread, you specified a "Refresh to ACT Delay" of 0. However, in the Advanced Timing Control section of the BIOS, there is a column of numbers in white text beside the parameters then there is a column in yellow text in line with the other parameters that have no white column of values, like this (white text is the numbers and yellow text is the word "auto")

 

Advanced Timing Control (White Text) (Yellow Text)

ACT to ACT Delay (tRRD) 3 auto

Rank Write to READ Delay 3 auto

Write to Precharge Delay 6 auto

Refresh to ACT Delay 42 auto

Read to Precharge Delay 3 auto

Static tRead Value 7 auto

Static TRead Phase Adjust 0 auto

 

Pressing "enter" on the "auto" beside Refresh to ACT Delay brings up a dialog showing "Min 0", "Max 255", with instructions to "key in a DEC number". When I key in 0 and press enter, the dialog closes and I am returned to the BIOS screen where the value for Refresh to ACT Delay is still shown as "auto". How do I change the value for Refresh to ACT Delay to 0 as you suggested?

Link to comment
Share on other sites

Was able to run memtest (15 passes, no errors) by using a memtest floppy instead of a memtest cd. Since this memtest was run with the 400 mhz FSB and 800 mhz (4,4,4,12) ram settings, I assume the memtest must now be repeated after adjusting the BIOS settings to 445 mhz FSB, the Vcore to 1.35v, and the DDR2 to 890 mhz (5,5,5,15) before entering Windows and running OCCT on the CPU. Is this the correct sequence of events?
Link to comment
Share on other sites

Set FSB to 445, DRAM at 890, 5, 5, 5, 15 on the latencies, DRAM OverVoltage Control at +.3v, (G)MCH at +.1v, VCore to 1.3225.

Booted to MemTest floppy and ran two passes with no errors.

Rebooted into Windows and took 3-wide screenshot of CPU-Z and shot of OCCT / RealTemp together - both attached.

Ran OCCT test on CPU for one hour. Screenshots halfway through and near end are attached.

Why the large discrepancy between the temps reported by OCCT and RealTemp? Which should I believe?

Are my temperatures during OCCT (67/63 or 57/53, depending on which set you believe) too high? Is a 20 degree increase from idle to load abnormal? Is the quick return of the core voltages to "normal" when OCCT lowered the CPU to "idle" during the final 4 minutes of the test a good sign or is that to be expected?

Based on the VCore voltage graph generated by OCCT (attached), the voltage was between 1.29 and 1.30 except at the beginning and end of the test. If so, it is valid to infer the system might be stable with a lower VCore, maybe 1.30 or even lower?

Would reducing the VCore voltage to 1.30 or lower reduce the core temps significantly under load?

Thanks for the help.

cpuz_3wide_4005.thumb.jpg.5df4bf52119e4c11cc2fc1552e0f07ee.jpg

occt_realtemp_4005_beforeOCCT.jpg.d7bb361c3e877b82400ba61a9b5080f3.jpg

occt_realtemp_halfwayocct_4005.jpg.b330ceebc3b44c246e8bea01eb59d0ee.jpg

occt_realtemp_endocct_4005.jpg.d9da57fb5eb1d5e9c53a8fb519cc38d2.jpg

2008-04-26-05h53-VCore.png.bf9da8d354a8c1fca6d59383c62aec3d.png

Link to comment
Share on other sites

The issue with thermals is that OCCT and most other software is following previous tjunctions and not the correct ones for 45nm. Real temp is correct at the maximums, but it does seem that there is no clearly correct idle temps, so follow the Real Temp full load temps.

 

Research:

 

http://www.xtremesystems.org/forums/showthread.php?t=179044

 

You have a good CPU if you can make 4.0Ghz fully stable with 1.30V. Keep in mind that the stock maximum is 1.25V so you are hardly above that voltage. You can try for lower voltages if you wish but you will not find any real difference in temperatues (my personal opinion, not fact).

 

It seems to be a good overclock. I would now run OCCT for ~12 hours for a full stable test. It seems that there can be errors in the later hours when you have a very low Vcore. Be certain this is not the case with a good 12 hour OCCT.

 

1.3225V in BIOS and ~1.3v in Windows gives you a Vdrop of .0225v which is quite minimal.

 

You should be happy for such a clocking and thermals. I find 1.344v (Windows) with 58C full load temps at 4.05Ghz. I am quite pleased with this result and your CPU is better.

 

 

Set FSB to 445, DRAM at 890, 5, 5, 5, 15 on the latencies, DRAM OverVoltage Control at +.3v, (G)MCH at +.1v, VCore to 1.3225.

Booted to MemTest floppy and ran two passes with no errors.

Rebooted into Windows and took 3-wide screenshot of CPU-Z and shot of OCCT / RealTemp together - both attached.

Ran OCCT test on CPU for one hour. Screenshots halfway through and near end are attached.

Why the large discrepancy between the temps reported by OCCT and RealTemp? Which should I believe?

Are my temperatures during OCCT (67/63 or 57/53, depending on which set you believe) too high? Is a 20 degree increase from idle to load abnormal? Is the quick return of the core voltages to "normal" when OCCT lowered the CPU to "idle" during the final 4 minutes of the test a good sign or is that to be expected?

Based on the VCore voltage graph generated by OCCT (attached), the voltage was between 1.29 and 1.30 except at the beginning and end of the test. If so, it is valid to infer the system might be stable with a lower VCore, maybe 1.30 or even lower?

Would reducing the VCore voltage to 1.30 or lower reduce the core temps significantly under load?

Thanks for the help.

Link to comment
Share on other sites

Kind words but you and RamGuy (early on) get the credit for any progress.

Had concerns about "overtesting" the CPU - actually found a thread online about it and the associated "degradation" some think they have experienced from extensive / extended testing.

Understanding one must test to confirm a stable system, but since benchmarking is not my pursuit, I am trying to find a "low" VCore where the system will support my normal computing activities (no gaming) but might show an error on the most rigorous benchmarking tests over an extended time period.

Set VCore to 1.28125 at boot this morning (wanted 1.28 but BIOS did not offer it) and the system booted normally.

This seems to be a "low" VCore for the system's current performance level (4 GHz) so I am thinkg of 1 -2 hours of OCCT to "confirm stability"at this setting.

I realize that 1 -2 hours of OCCT (or Prime95) is NOT measuring by the "gold standard" that ultimate overclockers use and am

interested to know if you feel this is a major mistake.

Thanks again.

Link to comment
Share on other sites

I am thinkg of 1 -2 hours of OCCT to "confirm stability"at this setting.

I realize that 1 -2 hours of OCCT (or Prime95) is NOT measuring by the "gold standard" that ultimate overclockers use and am

interested to know if you feel this is a major mistake.

 

It depends on how you use the system. For a system running 24/7 or longterm tasks and rebooted only days or weeks, there can issue singular errors that are cumulative and degenerative.

 

It's up to you where you go, but be clear, lower voltages can create eletromagetic effects that are as dangerous as overvolting. There MUST be a sweet spot and it is NOT that your system "seems" stable. If it is somewhat stable, then it is very possible that your system is suffering from electrostatic migration, transitory discharge (higher voltage) or drop/droop instabilities leading to corrupt data and longer term physical damage as both the overvolt and undervolt can raise issues that are cumulative and degenerative. This is doubly so in the system that is seldom turned off for any real length of time.

 

My advice is not to run into the obsessive action of undervolting and not testing for a single extended run. If your CPU can not take this kind of testing, then it is dysfunctional. However, you have learned enough now to go your own way and accept the ramifications of any choices you decide to make.

 

I personally believe that overclocking and undervolting has damaged many CPUs. DOUBLY so when undervolting with such a high overclock as yours and not testing for long term instability. Look at the more extreme overclockers. They perform SuperPi, etc. Not long term stabilities. Do you think most test for 12 hour stability for no reason? For fun? LOL

 

If you are longer term instable, then it is certainly possible that longer term degeneration will come about. Remember that these 45nm CPUs are far too new to know about any of the longer term degenerative effects that were seen with 65nm.

 

I can make 4.5Ghz unstable (longer term) and 1.28v @ 4.0 unstable (longer term) as well. I need more voltage for longer term stability and this will lengthen the life of the CPU as long as it is well cooled for load.

Link to comment
Share on other sites

Good reply, and helpful.

Downloaded and ran SuperPi and it completed 24 interations of [32M] in 13 min 45 sec (see attached). However, I am unsure what this reveals. The system is "short-term" stable? If yes, then we still need a 10-12 hour OCCT or Prime95? If yes, which?

One comment concerning "well cooled for load" - I disabled "CPU Smart Fan Control" in the BIOS then turned the rheostat knob on the Zalman fan speed control up to max which increased the fan speed from about 2150 to about 2450. Not a huge change but maybe every little bit helps.

superpi.jpg.7a1bec3f52ca92b1284d533dfc3e762f.jpg

Link to comment
Share on other sites

I would not think that a Dual Core CPU could even be considered short term stable with a single 32m 24 iteration run. You can consider it stable enough to work a multi-thread with your two cores at ~50% for 13 minutes.

 

It all depends on your definition of stable, either short term or long term. Short term stable to me is 100% of all cores working at ~100% of full load for a good two hours. Long term stable to me is 100% of all cores working at ~100% of full load for ~12 hours.

 

If your computer might be damaged running a Prime95 test for 12 hours, then it should be RMA'd because there are serious issues with that singular CPU. I perform a 24 hour Prime95 burn in on all my personal machines when I build them. First at stock, then at final overclock. Any issues that arise means RMA of whatever part fails. However, I have NEVER seen a CPU fail in this way. In other words, in all the overclocking I have done, there has never been a fail on a CPU with a 24 hour Prime95 burn in. I have been overclocking since the 286 and working with computers since the Commodore VIC 20. Work! That's what these CPUs were created for by the way and if your overclock is done well, then you have no worry. But stability is paramount with the long term viability of the CPU. Those who tell you that testing the CPU will damage it are, in my view, not at all understanding of the hardware. Poor overclocks, poor cooling, over-volting, undervolting are the killers of hardware. Those who correctly upclock and test for long term stability will not have issues. This is my opinion, whether considered by you or not. :D:

 

Good reply, and helpful.

Downloaded and ran SuperPi and it completed 24 interations of [32M] in 13 min 45 sec (see attached). However, I am unsure what this reveals. The system is "short-term" stable? If yes, then we still need a 10-12 hour OCCT or Prime95? If yes, which?

One comment concerning "well cooled for load" - I disabled "CPU Smart Fan Control" in the BIOS then turned the rheostat knob on the Zalman fan speed control up to max which increased the fan speed from about 2150 to about 2450. Not a huge change but maybe every little bit helps.

Link to comment
Share on other sites

In the Prime95 "Run A Torture Test" dialog box, there are options for:

Small FFTs (maximum FPU stress, data fits in L2 cache, RAM not tested much);

In-place large FFTs (maximum heat, power consumption, some RAM tested);

Blend (test some of everything, lots of RAM tested);

Custom requiring settings for MIN FFT, MAX FFT, Run FFTs in-place, Memory to use, & Time to run each FFT size.

Which of the above is correct for testing my configuration?

Also, in the "Options/CPU/CPU Settings and Information" dialog, is it necessary to specify the amounts of "daytime and nighttime memory" or are the defaults acceptable?

Any other configuration parameters to be set before running Prime95?

Thanks.

Link to comment
Share on other sites

Is it necessary to run two iterations of Prime95 simultaneously if you have a dual core processor? Seen some postings that version 25.4 of Prime95 has the ability to test both cores with only a single iteration of Prime95 running but the latest version at mersenne.org is 24.14.
Link to comment
Share on other sites

This is what I was telling you about. Had you listened to me and performed the proper voltage settings and tested, you would by now know your sweet spot. You wasted much time playing around with voltages far too low for stability. This is why I was forceful. I knew that you had dropped the Vcore far too low. Set Vcore to 1.32 and retest.
Link to comment
Share on other sites

I've received my E8400 and installed it on my EP35-DS4 with Corsair TWIN2X4096-6400C4DHX.

 

OCCT stable at 9x400 with the following voltage setting :

 

DDR2 OverVoltage Control = +.3V

PCI-e OverVoltage Control = Normal

(G)MCH OverVoltage Control = + 0.1

CPU Voltage Control = Auto

 

The MCH Overvoltage Control value is maybe too high ( knowing that the first value option start at + 0.0225 in the bios ).

 

Do you think it would be preferable to set the MCH at + 0.0225 ?

 

Thanks for your advises.

 

NB : Sorry for my english as it is not my mother tongue.

Link to comment
Share on other sites

  • Corsair Employee
Do you think it would be preferable to set the MCH at + 0.0225 ?

Yes the suggested setting is +.05 Volts for the Memory Controller Voltage so if its stable at + .02 Volts then by all means use it.

Link to comment
Share on other sites

Derek -

With Vcore at 1.325, Prime95 has now been running for eleven hours with no errors.

Want to let it run a while longer? If yes, how much longer?

Please advise concerning the two posts immediately above as my (G)MCH Overvoltage Control is set to +.1v which, according to Ram Guy, may be higher than necessary.

Note: my manual shows the available increase range for (G)MCH OverVoltage to be +0.025V ~ +0.375V at 0.025V increments - can't see if this agrees with the BIOS as my computer is tied up running Prime95.

Link to comment
Share on other sites

Go with RAM GUY's stated voltage on MCH.

Derek -

With Vcore at 1.325, Prime95 has now been running for eleven hours with no errors.

Want to let it run a while longer? If yes, how much longer?

Please advise concerning the two posts immediately above as my (G)MCH Overvoltage Control is set to +.1v which, according to Ram Guy, may be higher than necessary.

Note: my manual shows the available increase range for (G)MCH OverVoltage to be +0.025V ~ +0.375V at 0.025V increments - can't see if this agrees with the BIOS as my computer is tied up running Prime95.

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.


×
×
  • Create New...