I think 6xx nvidias downclock when they run "too" hot? Try to adjust manual Fanspeed to 100% and see if it goes down under 70 degress and look if the frequency goes up back again then and report us.
I stil believe this is more a problem of nvidia-settings not displaying the right clock rates under Linux. Anyway, since MAGIC has demonstrated with GPU-Z that no underclocking happens under Windows, I guess it comes down to comparing crunching performance under Windows and Linux. If the Linux performance is close to the Windows performance, we can be pretty sure they are running at full specs under Linux as well, no matter what nvidia-settings displays, right??
yes, of course.
but things get stanger.
if believe to magic's stats, then his 660 ti runs 2wu at a time for a ~2650 sec. i.e. ~1300 sec per 1 wu.
but my 660 ti (underclocked i think) runs 2wu for a 2100 sec. i.e. 1050 sec per wu.
or about a 24% faster.
Anyway, since MAGIC has demonstrated with GPU-Z that no underclocking happens under Windows
Simply not true.
Load
No load
The 680 will get stuck at 705, and there are fixes for it, but the OP never responded to my post and it doesn't look like he's using it now so I didn't bother to go into details with him.
Quote:
but things get stanger.
if believe to magic's stats, then his 660 ti runs 2wu at a time for a ~2650 sec. i.e. ~1300 sec per 1 wu.
but my 660 ti (underclocked i think) runs 2wu for a 2100 sec. i.e. 1050 sec per wu.
or about a 24% faster.
Yes because there are many factors that go into performance. What other applications are running? How long does BOINC run without using the system? What are the other system specs?
For example, if I run my minecraft server (this doesn't use the video card) under load on my primary card I go from about .071%/sec to .045%/sec. That's just with that server running that has nothing to do with the video card.
I also shaved almost a full 20 mins off of CPU work units just by putting BOINC on an SSD. There are many factors.
Processor plays a big role btw. Your 660 specs were on an i5 core. His are on an Athlon II.
I also shaved almost a full 20 mins off of CPU work units just by putting BOINC on an SSD. There are many factors.
in which project? i see no difference between one of my pc with hdd and another with ssd, both had (now it's upgraded, so no direct comparison shoud be performed) same performance in einstein.
Anyway, since MAGIC has demonstrated with GPU-Z that no underclocking happens under Windows
Simply not true.
Load
No load
Well, of course we are talking "under load" here. Who cares about idle, that never happens w/ Einstein@Home ;-). The question is whether the clock frequencies displayed by nvidia-settings under Load in Linux are for real (that would be underclocking under load), or not.
As for SSD vs HDD: I have no explanation for a speedup in this situation unless the account is (mis)configured to checkpoint excessively often like every few seconds, which wasn't the case for the machine in question.
As for SSD vs HDD: I have no explanation for a speedup in this situation unless the account is (mis)configured to checkpoint excessively often like every few seconds, which wasn't the case for the machine in question.
Hmmm, perhaps some other system aspect, that we aren't directly tracking here, say swap file performance is being improved and thus indirectly assisting. I know of a few systems where SSD's are explicitly used for swap file use only ... precisely to minimise the cost of cache misses.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
That is my 660Ti
)
That is my 660Ti
I think 6xx nvidias downclock
)
I think 6xx nvidias downclock when they run "too" hot? Try to adjust manual Fanspeed to 100% and see if it goes down under 70 degress and look if the frequency goes up back again then and report us.
DSKAG Austria Research Team: [LINK]http://www.research.dskag.at[/LINK]
no, it's not hot, it's just a
)
no, it's not hot, it's just a little warm
I stil believe this is more a
)
I stil believe this is more a problem of nvidia-settings not displaying the right clock rates under Linux. Anyway, since MAGIC has demonstrated with GPU-Z that no underclocking happens under Windows, I guess it comes down to comparing crunching performance under Windows and Linux. If the Linux performance is close to the Windows performance, we can be pretty sure they are running at full specs under Linux as well, no matter what nvidia-settings displays, right??
Cheers
HBE
yes, of course. but things
)
yes, of course.
but things get stanger.
if believe to magic's stats, then his 660 ti runs 2wu at a time for a ~2650 sec. i.e. ~1300 sec per 1 wu.
but my 660 ti (underclocked i think) runs 2wu for a 2100 sec. i.e. 1050 sec per wu.
or about a 24% faster.
RE: Anyway, since MAGIC has
)
Simply not true.
Load
No load
The 680 will get stuck at 705, and there are fixes for it, but the OP never responded to my post and it doesn't look like he's using it now so I didn't bother to go into details with him.
Yes because there are many factors that go into performance. What other applications are running? How long does BOINC run without using the system? What are the other system specs?
For example, if I run my minecraft server (this doesn't use the video card) under load on my primary card I go from about .071%/sec to .045%/sec. That's just with that server running that has nothing to do with the video card.
I also shaved almost a full 20 mins off of CPU work units just by putting BOINC on an SSD. There are many factors.
Processor plays a big role btw. Your 660 specs were on an i5 core. His are on an Athlon II.
RE: I also shaved almost a
)
in which project? i see no difference between one of my pc with hdd and another with ssd, both had (now it's upgraded, so no direct comparison shoud be performed) same performance in einstein.
This project right here, on
)
This project right here, on my i7. Bikeman confirmed this earlier.
RE: RE: Anyway, since
)
Well, of course we are talking "under load" here. Who cares about idle, that never happens w/ Einstein@Home ;-). The question is whether the clock frequencies displayed by nvidia-settings under Load in Linux are for real (that would be underclocking under load), or not.
As for SSD vs HDD: I have no explanation for a speedup in this situation unless the account is (mis)configured to checkpoint excessively often like every few seconds, which wasn't the case for the machine in question.
Cheers
HB
RE: As for SSD vs HDD: I
)
Hmmm, perhaps some other system aspect, that we aren't directly tracking here, say swap file performance is being improved and thus indirectly assisting. I know of a few systems where SSD's are explicitly used for swap file use only ... precisely to minimise the cost of cache misses.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal