Thanks for reporting, Alex! That Celeron uses a Silvermont Atom with Ivy Bridge GPU, so the results should at least apply to HD2500 / HD4000 as well. I'm posting this information in the BRP4 Intel GPU app feedback thread, so let's get the current thread back to topic.
By now it has been shown that the other Maxwell 2 cards are also affected: GTX960, GTX980Ti and very probably also Titan X. Could you edit the title to reflect this?
"low memory clock on Maxwell2 cards (960/970/980, probably Titan X)"
As a compromise between being informative, not too detailed, and not too likely to be outdated by events soon.
By the way, I'm grateful to those who brought up this matter and techniques for revising the behavior. My GTX 970 has continued to run for over six months with initial settings coming out of this discussion with stable consistent success (including the transition to a CUDA55 Parkes application) with a very considerable productivity increase over that which it would have had otherwise.
By now it has been shown that the other Maxwell 2 cards are also affected: GTX960, GTX980Ti and very probably also Titan X. Could you edit the title to reflect this?
Thanks,
MrS
@ExtraTerrestrial Apes
As the OP of this thread you should be able to change the title by making a new post, click "post to thread" and not quote or reply, and then when editing that post change the title.
What Holmis was describing is actually what you did in creating your latest message. You used the 'Post to thread' button instead of either 'Quote' or 'Reply', since at the top of your message there was no reference ('in response to ...') to any other specific message. Having made that post, you then had one hour in which you could have edited it. When making any such edit, you can change the thread title since, as the thread starter, you 'own' the thread :-).
I've seen this technique described before but since I can always see a nice 'Edit thread title' button, I've never needed to use it :-).
I also noticed that on both my 970's I had to dial down to 3.7ghz mem clock with cuda55 from the 3.8ghz mem clock with cuda 32. Even still, i'm getting some errant invalids, which means I may take it down further still to 3.6 to see if that completely eliminates the occurrences of invalids.
Thanks for reporting, Alex!
)
Thanks for reporting, Alex! That Celeron uses a Silvermont Atom with Ivy Bridge GPU, so the results should at least apply to HD2500 / HD4000 as well. I'm posting this information in the BRP4 Intel GPU app feedback thread, so let's get the current thread back to topic.
MrS
Scanning for our furry friends since Jan 2002
*bump* It might be a good
)
*bump*
It might be a good idea to make this a sticky thread.
MrS
Scanning for our furry friends since Jan 2002
RE: *bump* It might be a
)
+1
I just installed a 980Ti and I've been trying to figure out why it was running in P2 and limiting memory to 3.3GHz. It's fixed now. Thank you.
Thanks @ Mods! By now it
)
Thanks @ Mods!
By now it has been shown that the other Maxwell 2 cards are also affected: GTX960, GTX980Ti and very probably also Titan X. Could you edit the title to reflect this?
Thanks,
MrS
Scanning for our furry friends since Jan 2002
RE: TCould you edit the
)
I suggest for a revised title:
"low memory clock on Maxwell2 cards (960/970/980, probably Titan X)"
As a compromise between being informative, not too detailed, and not too likely to be outdated by events soon.
By the way, I'm grateful to those who brought up this matter and techniques for revising the behavior. My GTX 970 has continued to run for over six months with initial settings coming out of this discussion with stable consistent success (including the transition to a CUDA55 Parkes application) with a very considerable productivity increase over that which it would have had otherwise.
RE: Thanks @ Mods! By now
)
@ExtraTerrestrial Apes
As the OP of this thread you should be able to change the title by making a new post, click "post to thread" and not quote or reply, and then when editing that post change the title.
@Holmis: thanks.. although I
)
@Holmis: thanks.. although I can't find what you described. But nevermind, the updated title seems quite good :)
@Topic: I had to dial down my memory OC on a GTX970 a bit from my initial values. But 1.85 / 3.7 / 7.4 GHz works without visible errors now.
MrS
Scanning for our furry friends since Jan 2002
RE: ... I can't find what
)
I fixed the title as per archae86's suggestion.
What Holmis was describing is actually what you did in creating your latest message. You used the 'Post to thread' button instead of either 'Quote' or 'Reply', since at the top of your message there was no reference ('in response to ...') to any other specific message. Having made that post, you then had one hour in which you could have edited it. When making any such edit, you can change the thread title since, as the thread starter, you 'own' the thread :-).
I've seen this technique described before but since I can always see a nice 'Edit thread title' button, I've never needed to use it :-).
Cheers,
Gary.
I also noticed that on both
)
I also noticed that on both my 970's I had to dial down to 3.7ghz mem clock with cuda55 from the 3.8ghz mem clock with cuda 32. Even still, i'm getting some errant invalids, which means I may take it down further still to 3.6 to see if that completely eliminates the occurrences of invalids.
@Manuel: don't forget that
)
@Manuel: don't forget that instability and invalids can also be caused by the core clock being too high (for the current summer temperature).
@Gary: thanks, now I got it. I thought there was supposed to be some new button in the "reply to thread" window. It's clear now.
MrS
Scanning for our furry friends since Jan 2002