But this is just a kind suggestion. Came into my mind when I thought about nature of the term 'technical news'. I understand that term so that it's pretty much related to announcements provided by project stuff. Then the actual experiences of crunching those new things, especially on a longer timeline, might better fit elsewhere on this forum. But again this was just a thought.
Okay... Please a deep breath and step back. I don't care who you are or who stared it or who is or isn't at fault. This forum is not a place for arguments, flame wars, ecetera.
But this is just a kind suggestion. Came into my mind when I thought about nature of the term 'technical news'. I understand that term so that it's pretty much related to announcements provided by project stuff. Then the actual experiences of crunching those new things, especially on a longer timeline, might better fit elsewhere on this forum. But again this was just a thought.
Thank you for that suggestion; and, if this was just about RAC I'd agree. However; RAC is just ONE area affected by these newer Units. The MAC OpenCL Bug is another. Both RAC and the OpenCL Bug were reported here in this Thread as a DIRECT result of these new Units.
As RAC is now Increasing, (therefore, NOT complaining here), I'm continuing my chronicling what I see as a result of Christian and Bernd adjusting the Credits Awarded per Unit - which has a DIRECT effect on RAC. Since the floor dropped out from under us at 693 Credits per Unit, and continued descending even with an increase to 1365 Credits per Unit, it was necessary to continue chronicling even AFTER Bernd made the adjustment to 3465 Credits per Unit. RAC chronicling here helped Christian and Bernd to make those adjustments.
Okay... Please a deep breath and step back. I don't care who you are or who stared it or who is or isn't at fault. This forum is not a place for arguments, flame wars, ecetera.
- 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.
- 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.
My first 1.18 tasks just finished on my GTX970 and I have to say WOW!
It took 24.5 minutes while 1.17 tasks usually take about 47-51 minutes, that's a 50% speedup! I'm running x2 and the 1.18 task was paired with a 1.17 task.
I also observed that it still uses a full CPU core as support.
If this speedup holds for all FGRP-GPU tasks then I suspect that some adjustments to the estimates needs to be done again or my DCF will start swinging more wildly. My DCF right now is at 0.84 and the FGRP-GPU tasks, both 1.17 and 1.18, are estimated at about 54 minutes.
I have a Haswell Pentium dual core (G3258) with a brand new MSI R7 370 GPU 2GDST OC that MSI call "Armor 2x". It runs 1 CPU task and 2 GPU tasks concurrently. The OS is Linux and the video driver is fglrx (AMD Catalyst version 13.12 - 2014 vintage). The 1.17 version tasks have been taking approximately 2600 secs for the pair on this card. I've just looked at a full page of validated tasks which were all in the range 2584 - 2601 secs elapsed time. I promoted a pair of the 1.18 version and they have now both completed. The times were 1916 secs and 2006 secs respectively. That's getting close to a 25% reduction in elapsed time!!
Of course it's far too early to be definitive about this but it's tremendously exciting to see such an improvement for a card that is essentially just an old Pitcairn HD7850/7870 series GPU. These new cards are quite a bit faster than my big group of 7850s so I'm now interested to see if all those old cards show a similar reduction in elapsed time :-).
Maybe a better place for all
)
Maybe a better place for all these observations would be somewhere at 'Cruncher's corner', https://einsteinathome.org/community/forum/17
But this is just a kind suggestion. Came into my mind when I thought about nature of the term 'technical news'. I understand that term so that it's pretty much related to announcements provided by project stuff. Then the actual experiences of crunching those new things, especially on a longer timeline, might better fit elsewhere on this forum. But again this was just a thought.
TimeLord04 wrote:FREE
)
Here is not the US of A.
Discuss? Realize that you're talking to yourself. Ever thought about a diary?
Add me please.
Okay... Please a deep breath
)
Okay... Please a deep breath and step back. I don't care who you are or who stared it or who is or isn't at fault. This forum is not a place for arguments, flame wars, ecetera.
I recommend the Dalek Relaxation Tape for humans.
Einstein@Home Project
Richie_9 wrote:Maybe a better
)
Thank you for that suggestion; and, if this was just about RAC I'd agree. However; RAC is just ONE area affected by these newer Units. The MAC OpenCL Bug is another. Both RAC and the OpenCL Bug were reported here in this Thread as a DIRECT result of these new Units.
As RAC is now Increasing, (therefore, NOT complaining here), I'm continuing my chronicling what I see as a result of Christian and Bernd adjusting the Credits Awarded per Unit - which has a DIRECT effect on RAC. Since the floor dropped out from under us at 693 Credits per Unit, and continued descending even with an increase to 1365 Credits per Unit, it was necessary to continue chronicling even AFTER Bernd made the adjustment to 3465 Credits per Unit. RAC chronicling here helped Christian and Bernd to make those adjustments.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
Shawn Kwang wrote:Okay...
)
Just watched the video. That's AWESOME!!!
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
H__ctor VII wrote:Petec888
)
+1. I have a few 280x's though only 2 running E@H. Keep 'em coming like this!
Kailee.
- Due to the experiences with
)
- 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.
BM
Wow, that's great news,
)
Wow, that's great news, Scotty ! Will report as soon as I run a v1.18
-----
Bernd Machenschalk wrote:-
)
My first 1.18 tasks just finished on my GTX970 and I have to say WOW!
It took 24.5 minutes while 1.17 tasks usually take about 47-51 minutes, that's a 50% speedup! I'm running x2 and the 1.18 task was paired with a 1.17 task.
I also observed that it still uses a full CPU core as support.
If this speedup holds for all FGRP-GPU tasks then I suspect that some adjustments to the estimates needs to be done again or my DCF will start swinging more wildly. My DCF right now is at 0.84 and the FGRP-GPU tasks, both 1.17 and 1.18, are estimated at about 54 minutes.
I have a Haswell Pentium dual
)
I have a Haswell Pentium dual core (G3258) with a brand new MSI R7 370 GPU 2GDST OC that MSI call "Armor 2x". It runs 1 CPU task and 2 GPU tasks concurrently. The OS is Linux and the video driver is fglrx (AMD Catalyst version 13.12 - 2014 vintage). The 1.17 version tasks have been taking approximately 2600 secs for the pair on this card. I've just looked at a full page of validated tasks which were all in the range 2584 - 2601 secs elapsed time. I promoted a pair of the 1.18 version and they have now both completed. The times were 1916 secs and 2006 secs respectively. That's getting close to a 25% reduction in elapsed time!!
Of course it's far too early to be definitive about this but it's tremendously exciting to see such an improvement for a card that is essentially just an old Pitcairn HD7850/7870 series GPU. These new cards are quite a bit faster than my big group of 7850s so I'm now interested to see if all those old cards show a similar reduction in elapsed time :-).
Brilliantly done Christophe!!!
Cheers,
Gary.