There must be an intersection or overlap between over-clocking myths and over-clocking rules-of-thumb.
For instance, somebody on some forum somewhere -- coulda been Anandtech -- suggested that if you over-clock with a net increase in speed or voltage of 20%, you might expect your rig's longevity to be something like 80%. That is, the processor-- with an expected lifespan of ten years, might last eight.
I've developed a different view about this based on factory specs: "If you adopt some particular over-clocking strategy, the part that is stressed the most -- in relation to its factory spec limitations -- will fail sooner than everything else."
My experience over-clocking has only spanned some two years, while my experience building PCs extends over maybe fifteen years. So I'm new to this.
And my system is now a generation and a half old. Socket-478. Northwood migrated to Prescott. 800 Mhz FSB. SATA-150 drives.
What I chose to do in the end was to drop the multiplier on a Prescott 3.2E to 2.8 Ghz, and then to over-clock the front-side bus back to 1,000 Mhz so the processor ran at 3.5 Ghz and the memory modules (DDR500's) ran at their full spec.
Now there may be different opinions about this from more knowledgeable luminaries in this forum, but I don't believe that this pushed the processor to its stress limit. The net effect was to run the processor at 3.5 Ghz, or 300 Mhz over its recommended speed. The memory was run within its rated speed. And the motherboard was pushed beyond its recommended limit by 25%.
A month ago, what appeared to be memory errors crashed my system after six months of smooth running. But I replaced the memories with a dual-core kit rated and reviewed to run even faster. And after extensive error-free MEMTEST86 testing, S&M version 1.7.6 showed memory error -- an "address" error -- at a lower setting than with the original modules. It also "passed" the processor. Some more validation may be warranted, but it looks like the motherboard was beginning to "go south."
In hindsight, what could I expect to happen? And, in hindsight, it would seem that I might have been better off getting the fastest processor I could find with "decent" over-clocking potential, so I could spread the risk more evenly among components. For instance, I might get low-latency memories that were proven to be flexibly over-clockable at those latencies, and choose only to run the FSB up by an addition in Mhz only one-third of that achieved before.
Tell me -- tell us -- about your relatively short-lived over-clocking project, and whether your parts failure follows the hypothesis I put forward above. It isn't likely that each and all such failures will show that the part most-stressed by an over-clocking strategy is the part shortest-lived, but I'm betting that a large enough statistical sample will bear it out.
For instance, somebody on some forum somewhere -- coulda been Anandtech -- suggested that if you over-clock with a net increase in speed or voltage of 20%, you might expect your rig's longevity to be something like 80%. That is, the processor-- with an expected lifespan of ten years, might last eight.
I've developed a different view about this based on factory specs: "If you adopt some particular over-clocking strategy, the part that is stressed the most -- in relation to its factory spec limitations -- will fail sooner than everything else."
My experience over-clocking has only spanned some two years, while my experience building PCs extends over maybe fifteen years. So I'm new to this.
And my system is now a generation and a half old. Socket-478. Northwood migrated to Prescott. 800 Mhz FSB. SATA-150 drives.
What I chose to do in the end was to drop the multiplier on a Prescott 3.2E to 2.8 Ghz, and then to over-clock the front-side bus back to 1,000 Mhz so the processor ran at 3.5 Ghz and the memory modules (DDR500's) ran at their full spec.
Now there may be different opinions about this from more knowledgeable luminaries in this forum, but I don't believe that this pushed the processor to its stress limit. The net effect was to run the processor at 3.5 Ghz, or 300 Mhz over its recommended speed. The memory was run within its rated speed. And the motherboard was pushed beyond its recommended limit by 25%.
A month ago, what appeared to be memory errors crashed my system after six months of smooth running. But I replaced the memories with a dual-core kit rated and reviewed to run even faster. And after extensive error-free MEMTEST86 testing, S&M version 1.7.6 showed memory error -- an "address" error -- at a lower setting than with the original modules. It also "passed" the processor. Some more validation may be warranted, but it looks like the motherboard was beginning to "go south."
In hindsight, what could I expect to happen? And, in hindsight, it would seem that I might have been better off getting the fastest processor I could find with "decent" over-clocking potential, so I could spread the risk more evenly among components. For instance, I might get low-latency memories that were proven to be flexibly over-clockable at those latencies, and choose only to run the FSB up by an addition in Mhz only one-third of that achieved before.
Tell me -- tell us -- about your relatively short-lived over-clocking project, and whether your parts failure follows the hypothesis I put forward above. It isn't likely that each and all such failures will show that the part most-stressed by an over-clocking strategy is the part shortest-lived, but I'm betting that a large enough statistical sample will bear it out.