Hitting the wall with a GA P35-DS4

FishD · Nov 1, 2009

I have got up to 3.9gHz, posting, and OS, and even ran Prime. But failing after several minutes in prime.

I'm wondering what further adjustments should/could I make to improve stability at this clock.

Details of setting ad below.

Cheers

EarthDog · Nov 1, 2009

Try raising the vcore. Its below stock according to CPUz (b/c of vdrop).

You also dont show what your memory is running at. Make sure, for now, its 5-5-5-15 with AT LEAST 1.8v (if not the 2.1 needed for 1066).

Grnfinger · Nov 1, 2009

Failing the 1024 test you need to increase VTT or vNB, here is a good table to help you with Prime and errors

1024 - increase VTT or vNB
896 or 448 - vNB
768, 512, 332 - VTT

EarthDog · Nov 1, 2009

Interesting. Where did you get that information? Why is there no mention of Vcore?

FishD · Nov 1, 2009

EarthDog said:
............. also dont show what your memory is running at. Make sure, for now, its 5-5-5-15 with AT LEAST 1.8v (if not the 2.1 needed for 1066).

In my bios I have the option to increase the DDR voltage in small increments. Is this how I set it from 1.8 to 2.1 ie add +0.3v, or is there a setting for 2.1V, or does the MB detect the correct voltage in 'auto' mode. I need to be sure of this before adding 0.3V onto the normal ddr voltage.

Also i noted that even though when I set the VCore to both 2.00v and 2.0625v, cpuZ shows it up as 1.184v idling, and jumps between this and 1.168 under load. I expect this vdroop is the culprit in preventing me getting stability.

Any advice here welcome.

Grnfinger · Nov 1, 2009

EarthDog said:
Interesting. Where did you get that information? Why is there no mention of Vcore?

if you fail any other test its vcore, the rest is listed, it is common knowledge I thought..

I could link you the source as we tested this for many months over XS but it seems they black listed my IP so..but if you test this table you'll see its pretty acurate.. just cause my post count is low dont mean I'm a n00b

redduc900 · Nov 1, 2009

Post the VID of your chip using Real Temp--> "Settings" page--> Max Core VID (assuming it's 1.175V, but because of LLC being enabled it's 1.168V in CPU-Z - disable LLC while checking for stability). Increase the Vcore (like EarthDog mentioned) to a point at which while running Prime, the Core Voltage showin in CPU-Z is slightly higher than the VID. Reset PCI-E OverVoltage Control to Disabled or AUTO, and manually adjust the four primary RAM timings to 5-5-5-15; w/ DDR OverVoltage Control set to Disabled or AUTO (shouldn't need more than default Vdimm at only 870MHz DRAM frequency).

EarthDog · Nov 1, 2009

Grnfinger said:
if you fail any other test its vcore, the rest is listed, it is common knowledge I thought..

I could link you the source as we tested this for many months over XS but it seems they black listed my IP so..but if you test this table you'll see its pretty acurate.. just cause my post count is low dont mean I'm a n00b

I never inferred you were a noob. I have been a member at XS for 4+ years with 1k+ posts and never saw a mention about this (must have missed the thread). I dont have time to test it, otherwise, I certainly would. If what you mention is true, it should save people a lot of time and guessing at such things, including me. I just wanted proof like I do from everyone as a lot of times people come in spewing garbage with no supporting...well, anything.

But I am curious to which specific test this is associated with...Small, Large or Blend? If its all, how is that valid considering each test tests something significantly different than the other? For example, small FFT basically stresses ONLY the CPU. How would adding vNB to that help when nearly 0 memory is being tested? I guess I dont understand the association between the iterations and how they correlate with each voltage change.

(I dont want to get too into this in someone else's thread, so if I have further questions after this I will start another thread or PM you.

)

Grnfinger · Nov 1, 2009

EarthDog said:
I never inferred you were a noob. I have been a member at XS for 4+ years with 1k+ posts and never saw a mention about this (must have missed the thread). I dont have time to test it, otherwise, I certainly would. If what you mention is true, it should save people a lot of time and guessing at such things, including me. I just wanted proof like I do from everyone as a lot of times people come in spewing garbage with no supporting...well, anything.

But I am curious to which specific test this is associated with...Small, Large or Blend? If its all, how is that valid considering each test tests something significantly different than the other? For example, small FFT basically stresses ONLY the CPU. How would adding vNB to that help when nearly 0 memory is being tested? I guess I dont understand the association between the iterations and how they correlate with each voltage change.

(I dont want to get too into this in someone else's thread, so if I have further questions after this I will start another thread or PM you. )

Sorry I should have stated this is with Blend test...and why its not 100% its pretty close.

I'm not offended by your post I like evidence as well when ppl make claims, but if you have some free time test it out, you'll find it works.

I have over 1500 posts at XS I was cought up in the mass banning cause I am a member at RRTech so they never gave me a chance to try and defend myself

.
Anyways sorry to hijack the post

FishD · Nov 1, 2009

redduc900 said:
.................. disable LLC while checking for stability)........

Thanks for the useful tips.

Reset the memory voltage, the PCIE and set the memory timings to manual 5.5.5.15.

The VID is indeed the same as on my picture above, 1.175v, and I have set the VCore to 1.23125 in the bios. With the LLC enabled this gives me 1.216 rest, and drops to between 1.200 and 1.84 under load. With LLC disabled, it drops further to between 1.68 and 1.84, which is where I was getting instability. (Both speedFan 4.39 and cpuZ 1.52.1 agree on the measurements above).

I have been able to run prime for 30 mins without any error (i stopped it to make changes). So upping the VCore has helped, but I think i am doing better with LLC enabled. I will re-run prime for 30 mins to ensure i am back to where i was before the memory timings etc.

I am more confident now that i can reach 4.0gHz stable, with the current rig.

Cheers

redduc900 · Nov 1, 2009

With LLC disabled, Vdroop needs to be compensated for by increasing Vcore.

FishD · Nov 2, 2009

redduc900 said:
With LLC disabled, Vdroop needs to be compensated for by increasing Vcore.

Hi redduc,

Ive got the rig running stable now, at the settings of my last post, and have ran several hours with prime. Still doing some more testing. Even ran geoforms with my gpu overclocked along with it. My temp can hit 65c on one of the cores occasionally under stress, so I need to look a bit more at my cooling.

I still don't see the point (but i'm sure there is one) in having to up the vCore further, by switching off LLC, as this is going to raise temps, etc. Thought the objective of OC'ing is to do it with the minimum voltage of all components. I apologise in advance for my ignorance on this matter.

Cheers,

EarthDog · Nov 2, 2009

b/c it has the potential to be unstable b/c of the vdroop is why you should bump it up a notch.

EarthDog · Nov 3, 2009

65C is fine for that chip. Just keep it under 72C when stress testing and you know you will be fine for everything else you do.

EDIT: I would remove that link in your signature. Outside links aside from Heatware are not allowed.

FishD · Nov 3, 2009

Latest setup 04032011. Note have had to reduce CPU to 4.0, to get staiblity. Not sure of the cause.

Hitting the wall with a GA P35-DS4

FishD

Member

Attachments

EarthDog

Gulper Nozzle Co-Owner

Grnfinger

Member

EarthDog

Gulper Nozzle Co-Owner

FishD

Member

Attachments

Grnfinger

Member

redduc900

Inactive Moderator

EarthDog

Gulper Nozzle Co-Owner

Grnfinger

Member

FishD

Member

redduc900

Inactive Moderator

FishD

Member

EarthDog

Gulper Nozzle Co-Owner

EarthDog

Gulper Nozzle Co-Owner

FishD

Member

Similar threads