I have an Intel 3930K and AMD 6234 running Linux and both CPU types support AVX. All the completed Multi-directed GW search tasks appear to be waiting for validation. I have not seen any tasks marked as invalid yet for this application.
Don't get fooled by (essentially arbitrary) plan-class names. All our 64 Bit (GW) apps use AVX where possible and fall back to SSE when not, regardless of whether the plan class name includes "AVX" or not.
Sometime in the last very few hours the server status page numbers for validations on the CV run came off their previous relentless 0,0,0 values. Current numbers show 3050 valid tasks, 558 invalid, and only 4 inconclusive.
[edit: while I was still composing this post those numbers shifted to 4570 valid, with zeroes showing for invalid and inconclusive]
G still shows 0,0,0, so perhaps the CV validator got a trial run, maybe not yet the G validator.
Looking around a little on the result pages of for some frequent posters, I spotted some validations.
So does this mean that the Mac OS X version, which doesn't include an optimization in the name, is actually optimized for AVX or SSE2?
Every one of my Ubuntu 14.04 Linux machines have dumped the CV work units, but my Macs are cranking through them apace. I converted one of my Ubuntu machines (i7-5820k) to a Hackintosh running Mac OS Sierra this weekend, and it began devouring them nicely. However, I have no completed Linux WU to compare performance, unless GO/NO-GO counts.
I too miss the ability to look at tasks by type as well as status. That was pretty handy, especially with high volumes of tasks.
We have recently been reassured that something of this sort is on a todo list. But an application transition like this is one of the times we most regret the current lack.
just wanted to let the project / admin know that today various WU got cancelled by the server with the error message "no response". The time assumed as sufficient by the server was estimated by 3 days (14.10. WU assigned with 17.10. considered as deadline).
So far my observations based on some WUs are that my machine needs more than 20 hours per WU. Three days to complete a WU doesn't seem to be sufficient from my perspective. Unfortunately the CPU time spend on the cancelled WUs is now lost.
In one of the forum posts I read that 8 hours of crunching time are considered in average. At least for my machine it took much longer and I run a i7-6700k, which should provide a decent performance.
Example: Workunit 258374930. Interestingly it took my wing man more than 25 hours to complete the WU.
Hope this helps ? Happy to assist and support in case further information is needed or considered useful
Not sure why your i7 would be taking so long unless it is running a lot of other stuff at the same time, such as GPU work and the computer is pushed for resources.
My old AMD Phenoms, both Windows 32 bit and Linux 64 bit, get through a work unit in around 11 hours.
My old AMD Phenoms, both Windows 32 bit and Linux 64 bit, get through a work unit in around 11 hours.Conan
Caution: there is more than one size WU being issued, whether measured by estimated flops as distributed, credit as awarded, or elapsed time on the same machine. I've seen greater than 30% ET difference between two WUs with the same flops estimate on the same machine, which generally gives very reproducible timings on the applications which distribute identical work content WUs. Not true of this latest GW work. I've seen really huge variability (greater than 10X) for CV work. I've not spotted nearly so much for G work, but that may be an artifact of limited observation or early distribution.
I have an Intel 3930K and AMD
)
I have an Intel 3930K and AMD 6234 running Linux and both CPU types support AVX. All the completed Multi-directed GW search tasks appear to be waiting for validation. I have not seen any tasks marked as invalid yet for this application.
Don't get fooled by
)
Don't get fooled by (essentially arbitrary) plan-class names. All our 64 Bit (GW) apps use AVX where possible and fall back to SSE when not, regardless of whether the plan class name includes "AVX" or not.
BM
Sometime in the last very few
)
Sometime in the last very few hours the server status page numbers for validations on the CV run came off their previous relentless 0,0,0 values. Current numbers show 3050 valid tasks, 558 invalid, and only 4 inconclusive.
[edit: while I was still composing this post those numbers shifted to 4570 valid, with zeroes showing for invalid and inconclusive]
G still shows 0,0,0, so perhaps the CV validator got a trial run, maybe not yet the G validator.
Looking around a little on the result pages of for some frequent posters, I spotted some validations.
example validated CV WU 1
example validated CV WU 2
So does this mean that the
)
So does this mean that the Mac OS X version, which doesn't include an optimization in the name, is actually optimized for AVX or SSE2?
Every one of my Ubuntu 14.04 Linux machines have dumped the CV work units, but my Macs are cranking through them apace. I converted one of my Ubuntu machines (i7-5820k) to a Hackintosh running Mac OS Sierra this weekend, and it began devouring them nicely. However, I have no completed Linux WU to compare performance, unless GO/NO-GO counts.
Quite recently the server
)
Quite recently the server status page indicates some G validation activity, also the CV validation totals continue to climb.
I, personally, got a validation on a G WU.
I may have gotten more, but without filters the result pages on my GPU hosts are very, very laborious to search.
I too miss the ability to
)
I too miss the ability to look at tasks by type as well as status. That was pretty handy, especially with high volumes of tasks.
Jonathan Jeckell wrote:I too
)
We have recently been reassured that something of this sort is on a todo list. But an application transition like this is one of the times we most regret the current lack.
Hello,just wanted to let
)
Hello,
just wanted to let the project / admin know that today various WU got cancelled by the server with the error message "no response". The time assumed as sufficient by the server was estimated by 3 days (14.10. WU assigned with 17.10. considered as deadline).
So far my observations based on some WUs are that my machine needs more than 20 hours per WU. Three days to complete a WU doesn't seem to be sufficient from my perspective. Unfortunately the CPU time spend on the cancelled WUs is now lost.
In one of the forum posts I read that 8 hours of crunching time are considered in average. At least for my machine it took much longer and I run a i7-6700k, which should provide a decent performance.
Example: Workunit 258374930. Interestingly it took my wing man more than 25 hours to complete the WU.
Hope this helps ? Happy to assist and support in case further information is needed or considered useful
br Michael
Not sure why your i7 would be
)
Not sure why your i7 would be taking so long unless it is running a lot of other stuff at the same time, such as GPU work and the computer is pushed for resources.
My old AMD Phenoms, both Windows 32 bit and Linux 64 bit, get through a work unit in around 11 hours.
Conan
Conan_4 wrote:My old AMD
)
Caution: there is more than one size WU being issued, whether measured by estimated flops as distributed, credit as awarded, or elapsed time on the same machine. I've seen greater than 30% ET difference between two WUs with the same flops estimate on the same machine, which generally gives very reproducible timings on the applications which distribute identical work content WUs. Not true of this latest GW work. I've seen really huge variability (greater than 10X) for CV work. I've not spotted nearly so much for G work, but that may be an artifact of limited observation or early distribution.