[phenixbb] Phenix 1.18 RSR runtime

[Pavel Afonine]

Thanks Hannah,
thanks for quick reply and sharing files! It really is expected to be N 
times faster with nproc.. So I'm really keen to check this!
I will investigate right away once I have files.
Best,
Pavel


On 4/30/20 00:19, Hannah Bridges wrote:
> Hi Pavel,
>
> Thank you for the rapid reply. We have tried increasing nproc, but it 
> is still pretty slow. I will send my files privately so that you can 
> check if something is awry.
>
> Best wishes
>
> Hannah
>
> On 30/04/2020 08:15, Pavel Afonine wrote:
>> Hi Hannah,
>>
>> it can be N times faster (N = number of available CPUs) if you use 
>> nproc to specify number of CPUs to use.
>>
>> Alternatively you can disable the rotamer fitting step, which will 
>> bring the runtime back to previous "~1hrs" (but expect rotamer 
>> statistics go off).
>>
>> If you send me files I am happy to check all is in order.
>>
>> Pavel
>>
>> On 4/30/20 00:02, Hannah Bridges wrote:
>>> Dear Phenix developers,
>>>
>>> We have updated to phenix version 1.18-3845-000 and found that 
>>> realspace refinement (5 cycles) with minimization global, local grid 
>>> search and ADP is taking about 36 hours, compared to ~ 1hr in 
>>> versions 1.17 and 1.16. The slow step is the local grid search which 
>>> is taking >6 hours each cycle.
>>>
>>> We are very glad for the option to run the job detached, but I 
>>> wondered if this sort of timing was expected for this version?
>>>
>>> Best wishes
>>>
>>> Hannah
>>>
>>