BonzaiDuck
Lifer
- Jun 30, 2004
- 16,775
- 2,114
- 126
What you say doesn't deviate much from my thoughts, since we're speculating.
But think about it. They initially want to release processors priced at (say) $85, $100, $115.
Just to construct a scenario, let's say that a week's production is 10,000 CPUs. [I don't know how long it takes to knock those things out, but suppose, anyway . . . ]
Maybe the normal sample size to assure 2% accuracy (+/-) is 20, or 25, or 35 randomly sampled units. So they set up 35 bare-bones test systems -- maybe with their own motherboards -- pull last week's sample, and in a few hours time, they've plugged in this week's sample and set all to preconfigured, known settings. Then they stress-test for the remaining of the day. If the batch shows more than the anticipated failure, they downgrade the processor.
I think you'd be right, also, about the "margin of safety," -- that they test them -- better yet -- QC the manufacturing process to assure a performance level a notch above what they stamp them. And this is good for us geeks and nerds. They give us extra Lebensraum to "play."
I'm thinking now about this comparison I'm doing on the E2140 and E2180 systems -- same mobo, same memory, equivalent PSUs and cooling. And I'm thinking that these things are FSB-limited within the Intel voltage spec. You can't get above that FSB limit unless you punch up the VCore.
That's why I don't anticipate any certainty of running the E2140 to 3.0 Ghz as 9 x 333. I have trouble running the E2180 to 3.0 Ghz as 9 x 333 -- both configurations with the same VCORE. But at that very same VCORE, I CAN run the E2180 @ (stock=) 10 x 300 = 3.0 Ghz -- no problem. I just don't get the memory bandwidth, although I think I can tweak the latencies on the DDR2-667's to 3,3,3,8 -- and maybe even CMD= 1T. Then, we'll see about bandwidth.
Before I do that, of course, I'll have to reset this second (E2180) configuration to stock defaults and install the operating system, the drivers and the test software.
The lesser E21x0 models are multiplier-limited -- therefore locked deliberately in the manufacturing process, and I'm betting if you could unlock that multiplier, the E2140 would also be stable at the same voltage and 10 x 300 Mhz.
I THINK that possibility supports my theory about what they're doing, and I'd like to replace the intuition with more logic. But it's just a possibility -- not an observation, other than knowing that the multipliers are deliberately locked. It will be interesting to see what results I get to try the E2140 at the settings that gave me trouble with the E2180. Because if I DO have the same problems, it suggests -- even if inconclusively -- that the E2140 and E2180 are one and the same.
I should contract myself to Federal Trade Commission. Maybe I've invented a new specialty called "Forensic Over-Clocking." Now -- let me tell you how I solved the Kennedy Assassination . . . . .
But think about it. They initially want to release processors priced at (say) $85, $100, $115.
Just to construct a scenario, let's say that a week's production is 10,000 CPUs. [I don't know how long it takes to knock those things out, but suppose, anyway . . . ]
Maybe the normal sample size to assure 2% accuracy (+/-) is 20, or 25, or 35 randomly sampled units. So they set up 35 bare-bones test systems -- maybe with their own motherboards -- pull last week's sample, and in a few hours time, they've plugged in this week's sample and set all to preconfigured, known settings. Then they stress-test for the remaining of the day. If the batch shows more than the anticipated failure, they downgrade the processor.
I think you'd be right, also, about the "margin of safety," -- that they test them -- better yet -- QC the manufacturing process to assure a performance level a notch above what they stamp them. And this is good for us geeks and nerds. They give us extra Lebensraum to "play."
I'm thinking now about this comparison I'm doing on the E2140 and E2180 systems -- same mobo, same memory, equivalent PSUs and cooling. And I'm thinking that these things are FSB-limited within the Intel voltage spec. You can't get above that FSB limit unless you punch up the VCore.
That's why I don't anticipate any certainty of running the E2140 to 3.0 Ghz as 9 x 333. I have trouble running the E2180 to 3.0 Ghz as 9 x 333 -- both configurations with the same VCORE. But at that very same VCORE, I CAN run the E2180 @ (stock=) 10 x 300 = 3.0 Ghz -- no problem. I just don't get the memory bandwidth, although I think I can tweak the latencies on the DDR2-667's to 3,3,3,8 -- and maybe even CMD= 1T. Then, we'll see about bandwidth.
Before I do that, of course, I'll have to reset this second (E2180) configuration to stock defaults and install the operating system, the drivers and the test software.
The lesser E21x0 models are multiplier-limited -- therefore locked deliberately in the manufacturing process, and I'm betting if you could unlock that multiplier, the E2140 would also be stable at the same voltage and 10 x 300 Mhz.
I THINK that possibility supports my theory about what they're doing, and I'd like to replace the intuition with more logic. But it's just a possibility -- not an observation, other than knowing that the multipliers are deliberately locked. It will be interesting to see what results I get to try the E2140 at the settings that gave me trouble with the E2180. Because if I DO have the same problems, it suggests -- even if inconclusively -- that the E2140 and E2180 are one and the same.
I should contract myself to Federal Trade Commission. Maybe I've invented a new specialty called "Forensic Over-Clocking." Now -- let me tell you how I solved the Kennedy Assassination . . . . .
Facebook
Twitter