- Sep 28, 2005
- 21,042
- 3,522
- 126
you'll lose...
I can gaurentee it. :sneaky:
CB is very HT friendly... so its really almost 4+2 vs 6 + 3 seeing how HT on CB is almost half a real core.
you'll lose...
I can gaurentee it. :sneaky:
CB is very HT friendly... so its really almost 4+2 vs 6 + 3 seeing how HT on CB is almost half a real core.
Rubycon
Madame President
- Aug 10, 2005
- 17,768
- 485
- 126
A 990X is NOT a Xeon. Try running ECC memory and you will see what I mean. 
I prefer pushing BCLK to the limit. I do find the system transaction limit is higher than running at 133. This probably does not affect 97% of typical users, however.
My idea of stress testing is also completely different than others I guarantee that. People running 5GHz* on air are not stable in that regard. They're cheating and if they deny it they are lying. They are not exempt from the laws of physics.
Just because it runs cinebench/vantage/yourfavgame/etc. without giving an error does NOT mean the machine is stable! Running OCCT / IBT/ etc. can reveal errors quickly however they may need extended periods of run time. I like to run for no less than 300 minutes. Pass count is not of much use. If you're using Linpack (LynX) on a Gulftown you want to select 512MB size and change number of times to run to minutes and select 300 or more. If it passes that (five hours later) I'd say it's pretty safe bet that you're stable.
Why so long? Well overclocking is a gamble. ALL components have limits and the manufacturer guarantees a certain speed that is considerably below these limits. The difference between stock speed and absolute stability is called a margin. It's there to allow for warmer room temperatures, dust in the heatsink, etc. By running your equipment faster your margin decreases. This is why it's important to determine what the absolute limit is THEN select a speed that you're comfortable with. Running with little margin for 24/7 can often result in silent corruption or a problem later on as your system ages.
Water cooling the CPU can increase this margin somewhat or allow you to run "on the line" that was established with lesser or air cooling. The system will be more resistant to errata when it's hotter in the room or if dust accumulates on cooling fins, etc.
Long runs also will heat up ancillary components on the board and can reveal deficiencies that may not be a problem with shorter runs. There is nothing worse than believing your system is stable and working on a rendering project that takes 300 CPU hours and in the middle you get a BSOD or computation error! Sadly if that happens and you have to start over you would have finished ahead if you left things STOCK! Do it nice, not twice!
Why even overclock at all then? There's really no purpose, right? Some people call it fun. Some call it cheating. (Q6600 at 3.6GHz comes to mind or Celeron 300A at 466 etc.) It's a hobby. Don't let it make you lose sleep! Most overclocking is harmless but you can kill hardware if you get careless with voltages. With the advent of the LGA platform I'm willing to bet statistics for ruined motherboards from bent socket pins exceeds burned out CPUs from excessive vcore/vtt voltage.
Oh and those Cinebench scores I bet I can beat Aigo's score with a stock system!
________________________
* Gulftown/Nehalem @5GHz
I prefer pushing BCLK to the limit. I do find the system transaction limit is higher than running at 133. This probably does not affect 97% of typical users, however.
My idea of stress testing is also completely different than others I guarantee that. People running 5GHz* on air are not stable in that regard. They're cheating and if they deny it they are lying. They are not exempt from the laws of physics.
Just because it runs cinebench/vantage/yourfavgame/etc. without giving an error does NOT mean the machine is stable! Running OCCT / IBT/ etc. can reveal errors quickly however they may need extended periods of run time. I like to run for no less than 300 minutes. Pass count is not of much use. If you're using Linpack (LynX) on a Gulftown you want to select 512MB size and change number of times to run to minutes and select 300 or more. If it passes that (five hours later) I'd say it's pretty safe bet that you're stable.
Why so long? Well overclocking is a gamble. ALL components have limits and the manufacturer guarantees a certain speed that is considerably below these limits. The difference between stock speed and absolute stability is called a margin. It's there to allow for warmer room temperatures, dust in the heatsink, etc. By running your equipment faster your margin decreases. This is why it's important to determine what the absolute limit is THEN select a speed that you're comfortable with. Running with little margin for 24/7 can often result in silent corruption or a problem later on as your system ages.
Water cooling the CPU can increase this margin somewhat or allow you to run "on the line" that was established with lesser or air cooling. The system will be more resistant to errata when it's hotter in the room or if dust accumulates on cooling fins, etc.
Long runs also will heat up ancillary components on the board and can reveal deficiencies that may not be a problem with shorter runs. There is nothing worse than believing your system is stable and working on a rendering project that takes 300 CPU hours and in the middle you get a BSOD or computation error! Sadly if that happens and you have to start over you would have finished ahead if you left things STOCK! Do it nice, not twice!
Why even overclock at all then? There's really no purpose, right? Some people call it fun. Some call it cheating. (Q6600 at 3.6GHz comes to mind or Celeron 300A at 466 etc.) It's a hobby. Don't let it make you lose sleep! Most overclocking is harmless but you can kill hardware if you get careless with voltages. With the advent of the LGA platform I'm willing to bet statistics for ruined motherboards from bent socket pins exceeds burned out CPUs from excessive vcore/vtt voltage.
Oh and those Cinebench scores I bet I can beat Aigo's score with a stock system!
________________________
* Gulftown/Nehalem @5GHz
Castiel
Golden Member
- Dec 31, 2010
- 1,772
- 1
- 0
- Jan 4, 2010
- 1,989
- 20
- 81
Here's the Cinebench run I did:
The first one is higher because my bus speed was 135 as opposed to 133 which is the second (highlighted in orange) run.
Also, now I am at 4.41GHz "stable" and I was actually able to lower my QPI/DRAM (VTT) voltage to 1.275v while my v-core is still 1.29v.
On that note, is it okay for the QPI/DRAM voltage to be less than the v-core voltage? I know of the 0.5v rule but that refers to the DRAM voltage (for memory) which, for me, is at 1.66v (Corsair Dominator 1600MHz).
My QPI/DRAM voltage before was 1.30v so I'm glad I was able to bring it down by 0.025v!
Last edited:
Rubycon
Madame President
- Aug 10, 2005
- 17,768
- 485
- 126
Yes that looks fine and there is no requirement to set vcore and qpi volts the same.
When you say you are stable what are you using to validate this stability?
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter