Two of the last three results returned were invalid. I would back off the overclock a bit more (say 10MHz on FSB) to see if the errors stop. If they do, I think you have your answer. If they don't .... Well, let's just hope they do!! :).
Gary,
It has been a week since I lowered the FSB. I have had no errors and seven workunits have been completed sucessfully. Must have overclocked just a hair too much for E@H. Thanks for you help.
Yes, I am overclocking the A64. Std bus speed is 200 using a 10x multiplier which give you about 2000 MHz. Currently I have the bus set to 275 using a 9x multiplier which gives me 2483 MHz (about a 25% overclock). CPU is running 36 degrees C (99 F) which is well within acceptable bounds. I have run Memtest86+ with no errors for three hours and even backed down by 5 MHz from the highest stable setting just to be sure. I can drop back some more but I am not having issues with Predictor or Seti. The harddisk is a SATA drive so I don't think that I am clocking the system beyond its bounds.
AW,
I think it more likely that your setting are overstressing the memory. 265-275 FSBis really pushing it, because you don't have the option to uncouple the FSB and memory bus on the A64, memory controller is embedded in the CPU. Disregard that if you're using DDR533 or 550 sticks. Memtest, running by itself only checks for errors under basically idle condition, valid but incomplete picture - running it while crunching Einstein will give it a true test of your memory under stress conditions, where errors are more likely to appear. While ultrafast FSB/memory speeds give better scores on some benchmarks and memory-hammering apps (Photoshop or video-editing come to mind), they will do absolutely nothing to improve performance in Einstein - it just doesn't need much bandwidth. What Einstein does respond to is raw CPU speed, in a direct ratio. Can't you back off a good deal more on the FSB, bump the multiplier back up to 10x (or more), and return to 2400-2500 speeds?
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Gary,
It has been a week since I lowered the FSB. I have had no errors and seven workunits have been completed sucessfully. Must have overclocked just a hair too much for E@H. Thanks for you help.
Hey, those results look pretty good. You've lost a little bit of speed (18,900 instead of 18,100 seconds) but that's sure better than invalid results. My overclocking experience is with Athlon XPs which are now multiplier locked. If the multiplier on A64 isn't locked, why don't you do as Michael suggests and run at something like 10x240 or perhaps 10x250 and see if the CPU is still relatively cool? If the memory runs OK at 265 it should have no problem at 250 and you might beat your previous best at 10x250 :).
Gary,
It has been a week since I lowered the FSB. I have had no errors and seven workunits have been completed sucessfully. Must have overclocked just a hair too much for E@H. Thanks for you help.
Just wanted to say that I am impressed that you figured this out!
The error message was reporting that a certain quantity which must be finite was infinite. As you surmised, the best explanation is that either the memory or CPU was not working correctly with the overclocked speed.
Two of the last three results
)
Two of the last three results returned were invalid. I would back off the overclock a bit more (say 10MHz on FSB) to see if the errors stop. If they do, I think you have your answer. If they don't .... Well, let's just hope they do!! :).
Cheers,
Gary.
OK - Have reduce the FSB to
)
OK - Have reduce the FSB to 265 and rebooted. Hopefully this will resolve the issue.
Gary, It has been a week
)
Gary,
It has been a week since I lowered the FSB. I have had no errors and seven workunits have been completed sucessfully. Must have overclocked just a hair too much for E@H. Thanks for you help.
RE: Yes, I am overclocking
)
AW,
I think it more likely that your setting are overstressing the memory. 265-275 FSBis really pushing it, because you don't have the option to uncouple the FSB and memory bus on the A64, memory controller is embedded in the CPU. Disregard that if you're using DDR533 or 550 sticks. Memtest, running by itself only checks for errors under basically idle condition, valid but incomplete picture - running it while crunching Einstein will give it a true test of your memory under stress conditions, where errors are more likely to appear. While ultrafast FSB/memory speeds give better scores on some benchmarks and memory-hammering apps (Photoshop or video-editing come to mind), they will do absolutely nothing to improve performance in Einstein - it just doesn't need much bandwidth. What Einstein does respond to is raw CPU speed, in a direct ratio. Can't you back off a good deal more on the FSB, bump the multiplier back up to 10x (or more), and return to 2400-2500 speeds?
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: Gary, It has been a
)
Hey, those results look pretty good. You've lost a little bit of speed (18,900 instead of 18,100 seconds) but that's sure better than invalid results. My overclocking experience is with Athlon XPs which are now multiplier locked. If the multiplier on A64 isn't locked, why don't you do as Michael suggests and run at something like 10x240 or perhaps 10x250 and see if the CPU is still relatively cool? If the memory runs OK at 265 it should have no problem at 250 and you might beat your previous best at 10x250 :).
Cheers,
Gary.
RE: Gary, It has been a
)
Just wanted to say that I am impressed that you figured this out!
The error message was reporting that a certain quantity which must be finite was infinite. As you surmised, the best explanation is that either the memory or CPU was not working correctly with the overclocked speed.
Well done!
Director, Einstein@Home