Hi,

Well, after a complete disaster weekend with the new rig build, it seems my Sandybridge CPU may have bitten the dust.

I've had constant struggles with stability and system freezes.

There is a question mark over my motherboard (Asus P8P67 Pro), but I'm not sure if the struggles are due to the chip or the board, or both!

It feels that the stability has degraded over the past 3 days.

I now can't get stable running with any BIOS, either 804, 1003 or 1053, my system freezes in Prime95 after about 1 min now. All settings stock bar the RAM at 1.65v and 2133MHz.

I've seen the feedback from Gibbo on voltages, my RAM won't work at anything under 1.65v, it's recommended rating. I purchased Patriot Sector 5 RAM at 2250MHz based on Gibbo's post here that stated it was compatible with Sandybridge, which I'm struggling to see how it is.

http://forums.overclockers.co.uk/showthread.php?t=18226563

I did manage Saturday and Sunday to run most of the time with the RAM at the correct speeds and timings, under stock conditions and also some overclocked conditions, but would freeze randomly at the desktop or in benches.

Played Crysis and BFBC2 for 30min to 1hr periods ok. Out on the desktop it would freeze randomly, generally after some time.

Temps don't get over 48deg C at load in stock.

So, now the CPU won't TurboBoost above 3500MHz (should go to 3800) if left on Auto, when before it was working on two BIOS revs. Won't work in any BIOS rev now. If I force the Turbo ratio in the BIOS it freezes under Prime load pretty instantly on all BIOS revs.

I've mem tested the RAM and it is ok, but it may be a RAM incompatibility with the board issue mixed in there too, I don't know.

In any case, the system has degraded and I've ran 95% of the time stock settings, bar the RAM as discussed. My highest Vcore was a 30min spell at 1.32v where it ran at 4.8GHz. Temps were really good at 57deg C. Interestingly it ran Crysis a long time ok (a good hour), then exited and was tinkering on the desktop in Internet Exlporer and froze! I tried a 4.6GHz clock with similar results, at Vcore of 1.275v. Then I reverted back to stock settings and trials.

The thread below shows my struggles with the board. Post #52 shows a pic of the 4.8GHz clock.

http://forums.overclockers.co.uk/showthread.php?t=18226900&page=2

So, at first glance it feels like it is a mem compatibility issue, but why won't the Turbo ratio clock up to x38 under load now? I may be clutching at straws here, but is the RAM voltage affecting the CPU's memory controller? 1.65v is supposed to be safe, it also states it in the manual!

I'm going to run the CPU faster but with the RAM on Auto, see what happens.

Feel like returning everything at the mo!!

Hi there

Try setting the RAM to 1600MHz with 9-9-9-28 timings with 1.50v just to rule it out.

We've left a system under prime tonight with memory at 1.65v, see if its still alive in the morning.

Problem is even Sandybridge certified memory is 1.65v when running 2133MHz and higher, so you'd think that 1.65v is safe considering motherboard and memory manufacturers seem to be both saying its safe. I am sure the likes of G.Skill must have tested their memory with Sandybridge to advertise it as Sandybridge suited and with a stock voltage of 1.65v.

Getting ever closer to me cancelling my order

I would advise running intel CPU checker to see if the CPU is infact dieing...

matthab said:

Getting ever closer to me cancelling my order

Click to expand...

If I was buying now, I'd probably wait a couple of weeks for everything to settle down and to hear some definitive answers concerning ram voltage, overclocking, etc. Doesn't seem right this many people having issues

I have bought but my mainboard isnt in till the 17th so plenty of time to cancel.

In my view it should be this hard. People will be getting there CPUs tommorow and I hope no more CPU's die. I HOPE to god this is a batch issue that OCuk have accidentally been given.

Gibbo said:

Hi there

Try setting the RAM to 1600MHz with 9-9-9-28 timings with 1.50v just to rule it out.

We've left a system under prime tonight with memory at 1.65v, see if its still alive in the morning.

Problem is even Sandybridge certified memory is 1.65v when running 2133MHz and higher, so you'd think that 1.65v is safe considering motherboard and memory manufacturers seem to be both saying its safe. I am sure the likes of G.Skill must have tested their memory with Sandybridge to advertise it as Sandybridge suited and with a stock voltage of 1.65v.

Click to expand...

Hey Gibbo,

Thanks for the prompt response.

Well I'm running Prime as I type this, everything on Auto except the Turbo multiplier on x42. Running 4200GHz what appers to be stable.

Perhaps it is the RAM being incompatible with the board. I'm going to do some more tests. Which RAM do you recommend Gibbo? I pressume the XMS3 stuff? You have no stock though!!

This is why I dont run high speed memory in boards that say they support it.

I recommend the Corsair Vengeance 1600mhz stuff. Its 1.5v and is a nice clock, C8 or C9.

Gibbo, running at 1600MHz, 9 9 9 28 and fixed Turbo ratio at x38, seems to be running ok at the mo, will leave it running over night.

On this basis, what is the fastest memory that works?

matthab said:

I have bought but my mainboard isnt in till the 17th so plenty of time to cancel.

In my view it should be this hard. People will be getting there CPUs tommorow and I hope no more CPU's die. I HOPE to god this is a batch issue that OCuk have accidentally been given.

Click to expand...

Hope its not a batch issue, the job lot Intel shipped into the UK would all be affected as all our stock has originated from the same place as every other etailor that purchased from genuine Intel authorised dealer, which would mean I think at this stage a UK wide issue if not European wide.

I think compared to how many we shipped last week and have sold we've had 3-4 issues and so far I believe only 1 RMA. Then the ocuk bundles and systems have been fine.

But we shall do more testing ourselves as it would be nice to actually discover if something is killing the CPU's exactly what it is so we can pre-warn anyone not too do that.

Loko said:

Gibbo, running at 1600MHz, 9 9 9 28 and fixed Turbo ratio at x38, seems to be running ok at the mo, will leave it running over night.

On this basis, what is the fastest memory that works?

Click to expand...

Well you wanna see if you can get the CPU back towards at least 4.2GHz stable and then towards 4.6GHz.

Then if its stable you know its a memory compatibility issue.

The best memory in my book is memory that will run at 1600MHz with tight timings, like say 7-8-7-24 or even lower with just 1.50v.

So far C8 vengeance seems best and we've got some new lines coming in later this week.

But in regards to high frequency stuff, I'd not like to comment as the older generation I5/I7 dislike high frequency memory and maybe Sandybridge is the same. CPU's do seem to be getting more delicate to high frequency memory which makes one wonder why do memory manufacturers still make it and sell it when on the old I5/I7 it was more down to the CPU's onboard memory controller if you could run those speeds rather than the memory itself. Well maybe Sandybridge is the same, early days yet, but just a gut feeling.

I definetly got a bad cpu, however i think a lot of the issues here are teething problems with a new platform.

But in regards to high frequency stuff, I'd not like to comment as the older generation I5/I7 dislike high frequency memory and maybe Sandybridge is the same. CPU's do seem to be getting more delicate to high frequency memory which makes one wonder why do memory manufacturers still make it and sell it when on the old I5/I7 it was more down to the CPU's onboard memory controller if you could run those speeds rather than the memory itself. Well maybe Sandybridge is the same, early days yet, but just a gut feeling.

Click to expand...

Agree completely, I have discovered to my cost with my current rig high clocked memory is notorious and theres very little gains in it.

Gibbo said:

Well you wanna see if you can get the CPU back towards at least 4.2GHz stable and then towards 4.6GHz.

Then if its stable you know its a memory compatibility issue.

The best memory in my book is memory that will run at 1600MHz with tight timings, like say 7-8-7-24 or even lower with just 1.50v.

So far C8 vengeance seems best and we've got some new lines coming in later this week.

But in regards to high frequency stuff, I'd not like to comment as the older generation I5/I7 dislike high frequency memory and maybe Sandybridge is the same. CPU's do seem to be getting more delicate to high frequency memory which makes one wonder why do memory manufacturers still make it and sell it when on the old I5/I7 it was more down to the CPU's onboard memory controller if you could run those speeds rather than the memory itself. Well maybe Sandybridge is the same, early days yet, but just a gut feeling.

Click to expand...

Ok Gibbo, I'll leave the RAM as is and then start bumping the CPU up to 4.6GHz. Thanks for the help.

We'll have to wait and see how things develop from the RAM perspective.

matthab said:

Agree completely, I have discovered to my cost with my current rig high clocked memory is notorious and theres very little gains in it.

Click to expand...

Its the sole reason why in 6GB and 12GB kits we offer very little, because no matter how good the kit, its down to the CPU and most I7 CPU's simply can't run such frequencies, you need to get a great CPU to do it.

I find its best to buy 2000MHz stuff and just run it at 1600MHz with much tighter timings.

Loko said:

Gibbo, running at 1600MHz, 9 9 9 28 and fixed Turbo ratio at x38, seems to be running ok at the mo, will leave it running over night.

On this basis, what is the fastest memory that works?

Click to expand...

Hi mate,
I have an identical thread running, same problems to the wire...
It could be problems with the chip... I run a p67a ud7 so it might not just be your board.
I would reccomend a few things that I did to ressurect my chip... I now have it running stable (intel burn test & gaming stable @ 4.3ghz)

1. If your going to run your ram at 2133 and 1.65 volts (I have similar ram and I'm running it at 1.66v and 2133) then I reccomend you up the QPI VTT a bit (1.2V) as it needs the juice at 2133.

2. Most cpus after a "burn in period" will need a modest vcore bump mine went from stable at 4.4ghz 1.28vcore to 1.32vcore stable at 4.3ghz...

3. Turn off turbo boost - its ****e... set your multi to desired overclock (in my case 4.3ghz) and let speedstep handle the underclocking...

4. Turn on Load Line Calibration to max and put the Vcore slightly less than you think it will need... e.g in bios I have mine set to 1.30V and its around 1.32-1.33V in windows CPUZ and HW monitor verified...

5. Set your VCCSA VCCIO +PLL to the stock figures off the top of my head they will be something like 0.925 1.025 and 1.82v. You can google them by typing I5 2500k VCCSA VCCIO CPU PLL...

6. Make sure your timings are not to tight for the ram.

7. Set your BCLK to 100mhz manually.

8.Any other voltage settings do no select normal or auto - define them yourself (type in the stock voltage e.g 1.82v for PLL) this will ensure that your board is not putting in the wrong values or bumping them itself...

People will probs tell me I'm silly for doing this to my chip, but this was the only way I got it stable... It came back from only botting at x16 multi...

lollyhayes said:

Hi mate,
I have an identical thread running, same problems to the wire...
It could be problems with the chip... I run a p67a ud7 so it might not just be your board.
I would reccomend a few things that I did to ressurect my chip... I now have it running stable (intel burn test & gaming stable @ 4.3ghz)

1. If your going to run your ram at 2133 and 1.65 volts (I have similar ram and I'm running it at 1.66v and 2133) then I reccomend you up the QPI VTT a bit (1.2V) as it needs the juice at 2133.

2. Most cpus after a "burn in period" will need a modest vcore bump mine went from stable at 4.4ghz 1.28vcore to 1.32vcore stable at 4.3ghz...

3. Turn off turbo boost - its ****e... set your multi to desired overclock (in my case 4.3ghz) and let speedstep handle the underclocking...

4. Turn on Load Line Calibration to max and put the Vcore slightly less than you think it will need... e.g in bios I have mine set to 1.30V and its around 1.32-1.33V in windows CPUZ and HW monitor verified...

5. Set your VCCSA VCCIO +PLL to the stock figures off the top of my head they will be something like 0.925 1.025 and 1.82v. You can google them by typing I5 2500k VCCSA VCCIO CPU PLL...

6. Make sure your timings are not to tight for the ram.

7. Set your BCLK to 100mhz manually.

8.Any other voltage settings do no select normal or auto - define them yourself (type in the stock voltage e.g 1.82v for PLL) this will ensure that your board is not putting in the wrong values or bumping them itself...

People will probs tell me I'm silly for doing this to my chip, but this was the only way I got it stable... It came back from only botting at x16 multi...

Click to expand...

Thumbs up!

When testing with the Asus board today these boards seem overly keen to autotune everything, as above set as much to manual as possible and with the post above you can do so.

Again big thanks for posting this up think it will help a lot.

Gibbo how much weight do you have to ask Asus to fix it? (+other manufactures)

Or is this issue Intels area?

I don't seem to have QPI VTT voltage in the Asus P67 Pro BIOS?

Just to point out that the Patriot 2250 Ram you have together with the Patriot 2400 ram I have is on the Asus compatibility list for the P8P67 Pro (In the manual that came with my board). It even shows that it should be at 1.66v.

I'm also wondering if the PLL voltage change in the Asus 1053 bios may be causing issues.

Gibbo, what bios are you running on the test P8P67?

I'm going to flash mine back to 1003 before I install my new chip arriving today.

Remeber the Asus 1053 bios is BETA so it may not be 100% stable. I personally wouldnt be using it with such a new chip.