I´m curious, will the Polaris run more than 1 WU@time?
By memory alone yes, the ones on Dave's pc linked below are only taking 111mb for each workunit. It's a 4gb gpu so there's plenty of memory overhead, but I don't know how to tell by looking at his stats how many units he is running at one time.
I´m curious, will the Polaris run more than 1 WU@time?
By memory alone yes
I think memory is not the base issue of concern here. My understanding is that the GCN 1.2 parts (Volcanic islands, including Tonga and Fiji) won't produce correct results for Einstein GRP6 work using the current code. I'm not aware that anyone has diagnosed the reason. It could be a hardware architectural flaw, a driver bug, or an application error.
Polaris is to be GCN 1.3, but whether that change will affect things remains to be seen, I think. I'm eager to see Gamboleer's result, and unless I hear of a disastrous failure to meet hopes, am quite likely to order early myself.
The event log didn't show the Titans at all, only the 1080.
For the BOINC part of this matter, did you have a cc_config file with: 1 ?
I'm not clear on just when this is or is not required, but suspect it may required for dissimilar GPU cards where identical ones don't need it.
But that would not seem to explain why your system did not recognize the cards (I suppose this means they were not listed in system|Device manager|Display adapters).
Still doesn't explain why the system doesn't see the Titans.
I understand I would need the cc_config.xml for it to recognize all the cards but why would the computer itself not see the Titans?
Precision X and SIV64 both didn't see the cards either, only the 1080.
So I have to wonder if there is something about installing the 1080 that prevents the other cards from being seen/used/installed/recognized (what is the appropriate word?)
For me, since I have multiple GPU it would require replacing all of my cards.
For a single GPU or dual machine it would make sense.
Or possibly something in you mobo is causing problems when you try to use those cards in those slots.
SIV64 certainly does (or soon will) recognise a GTX 1080, if the motherboard allows it - Ray Hinchliffe sent me an email this morning, with a screenshot of SIV V5.10 Beta-25 showing GTX 1080 details. He also showed me a GTX 980 display, though that may have been from a different machine - can't be sure whether he has overcome the 'both at the same time' problem.
The question he asked me - and I may as well ask it here - is whether the formula which both BOINC and SIV use is giving the correct 'GFLOPs peak' value for the Pascal geometry (compute capability 6.01, GFLOPs 8,878 in his example). Anyone able to confirm that independently from sources other than BOINC or SIV?
I didn't run it here but over
)
I didn't run it here but over on Seti.
Installed it into a system with Titan Xs.
After installing it with the appropriate driver (the only 1 for a 1080), the system failed to show any Titans, only the 1080.
Removal of the driver and installing a lower driver version continued to show the 1080 but failed to show any Titans.
Removal of the 1080 and reinstalling of Nvidia Drivers brought the Titans back.
All of this was without even starting up BOINC, the computer would not show those other cards on the same board as the 1080.
Before removing it, I did launch BOINC and Seti.
The event log didn't show the Titans at all, only the 1080.
So I don't know if there is an issue wherein only a 10*0 is allowed to function on a board or what.
I´m curious, will the
)
I´m curious, will the Polaris run more than 1 WU@time?
RE: I´m curious, will the
)
To be determined. I plan to buy one as soon as the RX 480 is released, and my first test will be to run paired with a 7970 at 2 units each.
RE: I´m curious, will the
)
By memory alone yes, the ones on Dave's pc linked below are only taking 111mb for each workunit. It's a 4gb gpu so there's plenty of memory overhead, but I don't know how to tell by looking at his stats how many units he is running at one time.
RE: RE: I´m curious,
)
I think memory is not the base issue of concern here. My understanding is that the GCN 1.2 parts (Volcanic islands, including Tonga and Fiji) won't produce correct results for Einstein GRP6 work using the current code. I'm not aware that anyone has diagnosed the reason. It could be a hardware architectural flaw, a driver bug, or an application error.
Polaris is to be GCN 1.3, but whether that change will affect things remains to be seen, I think. I'm eager to see Gamboleer's result, and unless I hear of a disastrous failure to meet hopes, am quite likely to order early myself.
For those who don´t for some
)
For those who don´t for some reason the 390X, Fury X etc (Hawai´s GPU´s) when run 2 or more WU at a time produces a lot of erros.
That happening at least in E@H and S@H. Something driver related, not sure if was ficed on the newer driver.
The issue is not memory related since all top end AMD GPU´s has a lot of memory.
RE: The event log didn't
)
For the BOINC part of this matter, did you have a cc_config file with: 1 ?
I'm not clear on just when this is or is not required, but suspect it may required for dissimilar GPU cards where identical ones don't need it.
But that would not seem to explain why your system did not recognize the cards (I suppose this means they were not listed in system|Device manager|Display adapters).
archae86 wrote: I'm not clear
)
In Zalster's case it is required i would think, see Boinc Forum: Multibrand GPUs and use_all_gpus
Still doesn't explain why the
)
Still doesn't explain why the system doesn't see the Titans.
I understand I would need the cc_config.xml for it to recognize all the cards but why would the computer itself not see the Titans?
Precision X and SIV64 both didn't see the cards either, only the 1080.
So I have to wonder if there is something about installing the 1080 that prevents the other cards from being seen/used/installed/recognized (what is the appropriate word?)
For me, since I have multiple GPU it would require replacing all of my cards.
For a single GPU or dual machine it would make sense.
So I leave this conversation for others.
Or possibly something in you
)
Or possibly something in you mobo is causing problems when you try to use those cards in those slots.
SIV64 certainly does (or soon will) recognise a GTX 1080, if the motherboard allows it - Ray Hinchliffe sent me an email this morning, with a screenshot of SIV V5.10 Beta-25 showing GTX 1080 details. He also showed me a GTX 980 display, though that may have been from a different machine - can't be sure whether he has overcome the 'both at the same time' problem.
The question he asked me - and I may as well ask it here - is whether the formula which both BOINC and SIV use is giving the correct 'GFLOPs peak' value for the Pascal geometry (compute capability 6.01, GFLOPs 8,878 in his example). Anyone able to confirm that independently from sources other than BOINC or SIV?