- Due to the experiences with the 1.17 "Beta" app versions the general GPU RAM requirements for FGRPB1G has been lowered to ~766 MB
- There is a new "Beta Test" app version 1.18 with a few improvements Christophe developed over the holidays. It should be significantly faster than 1.17, in particular on GPUs that support double precision.
Thanks for this new information, Bernd. I look forward to getting some 1.18 Units in the next couple days on both systems. It will take about 2-3 days to crunch through my remaining 1.17 Units on both machines. As usual, I will monitor, and report. I will also continue monitoring the MAC NVIDIA OpenCL Bug.
This improvement came from an idea I had about Fast Fourier Transform symmetries. You know ... ideas that suddenly came when you are taking your shower.
It's working well for the GPU where maths cost little but memory a lot, but it's not very efficient for the CPU.
So this idea won't help the CPU version of this app so far.
I have a machine with an i3 4130 (dual core, 4 threads) which runs 2 CPU tasks and 2 GPU tasks concurrently. The GPU is a HD7850. It uses the same OS and video driver as listed in my previous post. It has been crunching a pair of the 1.17 GPU tasks in the range of 2890 - 3086 secs and averaging around 2960 secs. It also has now completed a couple of the new 1.18 tasks. The times were 2078 secs and 2170 secs.
So, yes, the older Pitcairn series card shows around a 25% improvement as well.
@Holmis - It looks like the improvements may at last be helping NVIDIA cards. Your GTX 970 was giving the same sort of times as my old HD7850s and now it's giving the much better times you might expect from a more powerful GPU. With the AMD cards I'm running, I don't need 1 CPU core per GPU task. The crunch times are not really affected by sharing a single CPU core (or two threads for i3s) between the pair of GPU tasks.
I can see a significantly reduced CPU usage on AMD FuryX and Radeon HD7950 against v1.17 - about 1/3.
HD7950 running 2xWU: v1.17: 2000, v1.18: ~1200 seconds !
Fury X doesn't benefit so much due to low DPFP - almost no improvement (~1600 s running 2xWU), but 1/3 CPU usage.
GTX1050 Ti running 1xWU: v1.17: 2550, v1.18: ~1500 seconds !
- Due to the experiences with the 1.17 "Beta" app versions the general GPU RAM requirements for FGRPB1G has been lowered to ~766 MB
- There is a new "Beta Test" app version 1.18 with a few improvements Christophe developed over the holidays. It should be significantly faster than 1.17, in particular on GPUs that support double precision.
Thanks.
Unfortunately, I just tried a 1.18 WU on a GTX 260 with 896 MB memory and it failed. https://einsteinathome.org/task/603230102. Other tasks on this host, except maybe the before last one were done on a quadro 4000 with 2000 MB.
Unfortunately, I just tried a 1.18 WU on a GTX 260 with 896 MB memory and it failed. https://einsteinathome.org/task/603230102. Other tasks on this host, except maybe the before last one were done on a quadro 4000 with 2000 MB.
The 1.17 app has a minimum RAM requirement of ~750MB, 1.18 requires ~1GB. You probably got the task assigned for the memory of the Quadro, but the Client actually ran it on the smaller one. I don't know anything else to advise you than to turn off "Beta Test work" at least for that host (i.e. venue) in your preferences to avoid that.
was really looking forward to the new 1.18 tasks, but on one of my hosts these are all failing; see https://einsteinathome.org/host/12464084/tasks/error
Anyone know what I can do to fix it? Especially as my other host still has some 1.17s and happily crunching, would be good to know what I need to know before the upgrade happens there - or it too will have to wait...
TIA,
Kailee.
EDIT I: My other host now also got some 1.18 WUs. All of these also failed; https://einsteinathome.org/host/12464453/tasks/error. Both of these hosts are running OSX (El Capitan), and have R9 280x GPUs. For now I have switched off getting Beta WUs so at least they can do some 1.17 work (though one will have to wait a few hours as it ran out of quota...).
EDIT II: Ok so with all the earlier errors and this host (12464084) having run out of quota the next comms has been deffered by 24:09:30... Any way of fixing this and having host 12464084 do some crunching over the next day instead of waiting (luckily my other one is still getting the non-beta 1.17s) ? Many thanks for any help!!
RAC is 500 Points away from January 6th's level of 66K RAC. So, RAC is still climbing on the 1.17 Units.
[EDIT:]
Well, in finishing this Post, it seems that I picked up the 500 Points... Now at Jan 6th RAC Level. WOOHOO!!! Things are looking up, again.
[1.18 Units:]
I have two 1.18 Units in queue on Win XP Pro x64 with GTX-760 card. They came in today, and are due for completion on January 30th. The GTX-760, (still, (obviously), crunching 1.17 Units), will be a few days behind in getting to the 1.18 Units.
The MAC/Hackintosh with two GTX-750TI SC cards is still picking up 1.17 Units. So, in reading Kai Lee's Post that his MAC has received some is encouraging to me that I too will be capable of receiving them on my MAC. It is sad, though, to read that the 1.18's are failing on his two El Capitan MACs... I will continue monitoring, and report when I receive 1.18's on my MAC. Hopefully, (in the next few days), when I'm finished crunching 1.17's on the MAC Bernd and crew will have Kai Lee's issues sorted, and I too will be crunching 1.18's on the MAC.
Bernd Machenschalk wrote:-
)
Thanks for this new information, Bernd.
I look forward to getting some 1.18 Units in the next couple days on both systems. It will take about 2-3 days to crunch through my remaining 1.17 Units on both machines. As usual, I will monitor, and report. I will also continue monitoring the MAC NVIDIA OpenCL Bug.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
This improvement came from an
)
This improvement came from an idea I had about Fast Fourier Transform symmetries. You know ... ideas that suddenly came when you are taking your shower.
It's working well for the GPU where maths cost little but memory a lot, but it's not very efficient for the CPU.
So this idea won't help the CPU version of this app so far.
Cheers,
Christophe
I have a machine with an i3
)
I have a machine with an i3 4130 (dual core, 4 threads) which runs 2 CPU tasks and 2 GPU tasks concurrently. The GPU is a HD7850. It uses the same OS and video driver as listed in my previous post. It has been crunching a pair of the 1.17 GPU tasks in the range of 2890 - 3086 secs and averaging around 2960 secs. It also has now completed a couple of the new 1.18 tasks. The times were 2078 secs and 2170 secs.
So, yes, the older Pitcairn series card shows around a 25% improvement as well.
@Holmis - It looks like the improvements may at last be helping NVIDIA cards. Your GTX 970 was giving the same sort of times as my old HD7850s and now it's giving the much better times you might expect from a more powerful GPU. With the AMD cards I'm running, I don't need 1 CPU core per GPU task. The crunch times are not really affected by sharing a single CPU core (or two threads for i3s) between the pair of GPU tasks.
Cheers,
Gary.
I can see a significantly
)
I can see a significantly reduced CPU usage on AMD FuryX and Radeon HD7950 against v1.17 - about 1/3.
HD7950 running 2xWU: v1.17: 2000, v1.18: ~1200 seconds !
Fury X doesn't benefit so much due to low DPFP - almost no improvement (~1600 s running 2xWU), but 1/3 CPU usage.
GTX1050 Ti running 1xWU: v1.17: 2550, v1.18: ~1500 seconds !
-----
My trusty old HD7970 GHz
)
My trusty old HD7970 GHz edition went from 1100secs on v1.17 to 640secs on v1.18 (single task). Very nice
Edit: Unfortunately the reduced runtime for the GPU tasks made boinc think my CPU got faster as well. Now filling my task list with LOTS of CPU jobs,
Bernd Machenschalk wrote:-
)
Thanks.
Unfortunately, I just tried a 1.18 WU on a GTX 260 with 896 MB memory and it failed. https://einsteinathome.org/task/603230102. Other tasks on this host, except maybe the before last one were done on a quadro 4000 with 2000 MB.
Unfortunately, I just tried a
)
The 1.17 app has a minimum RAM requirement of ~750MB, 1.18 requires ~1GB. You probably got the task assigned for the memory of the Quadro, but the Client actually ran it on the smaller one. I don't know anything else to advise you than to turn off "Beta Test work" at least for that host (i.e. venue) in your preferences to avoid that.
BM
1.18 is a large improvement
)
1.18 is a large improvement on my GTX 1080. Running 3 tasks at a time, completion times went from around 35 minutes to around 23 minutes.
Hi all, was really
)
Hi all,
was really looking forward to the new 1.18 tasks, but on one of my hosts these are all failing; see https://einsteinathome.org/host/12464084/tasks/error
Anyone know what I can do to fix it? Especially as my other host still has some 1.17s and happily crunching, would be good to know what I need to know before the upgrade happens there - or it too will have to wait...
TIA,
Kailee.
EDIT I: My other host now also got some 1.18 WUs. All of these also failed; https://einsteinathome.org/host/12464453/tasks/error. Both of these hosts are running OSX (El Capitan), and have R9 280x GPUs. For now I have switched off getting Beta WUs so at least they can do some 1.17 work (though one will have to wait a few hours as it ran out of quota...).
EDIT II: Ok so with all the earlier errors and this host (12464084) having run out of quota the next comms has been deffered by 24:09:30... Any way of fixing this and having host 12464084 do some crunching over the next day instead of waiting (luckily my other one is still getting the non-beta 1.17s) ? Many thanks for any help!!
[RAC Update - 1-16-2017 at
)
[RAC Update - 1-16-2017 at 4:00 PM - PST:]
RAC is 500 Points away from January 6th's level of 66K RAC. So, RAC is still climbing on the 1.17 Units.
[EDIT:]
Well, in finishing this Post, it seems that I picked up the 500 Points... Now at Jan 6th RAC Level. WOOHOO!!! Things are looking up, again.
[1.18 Units:]
I have two 1.18 Units in queue on Win XP Pro x64 with GTX-760 card. They came in today, and are due for completion on January 30th. The GTX-760, (still, (obviously), crunching 1.17 Units), will be a few days behind in getting to the 1.18 Units.
The MAC/Hackintosh with two GTX-750TI SC cards is still picking up 1.17 Units. So, in reading Kai Lee's Post that his MAC has received some is encouraging to me that I too will be capable of receiving them on my MAC. It is sad, though, to read that the 1.18's are failing on his two El Capitan MACs... I will continue monitoring, and report when I receive 1.18's on my MAC. Hopefully, (in the next few days), when I'm finished crunching 1.17's on the MAC Bernd and crew will have Kai Lee's issues sorted, and I too will be crunching 1.18's on the MAC.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees