Should I set `inter_op_parallelism_threads` and `intra_op_parallelism_threads` to 1 when I use ray to create a actor?How can I safely create a nested directory?Why are Python lambdas useful?“Least Astonishment” and the Mutable Default ArgumentShould I use 'has_key()' or 'in' on Python dicts?In Python, when to use a Dictionary, List or Set?Python string formatting: % vs. .formatInstalling specific package versions with pipWay to create multiline comments in Python?How to change the order of DataFrame columns?Ray: How to run many actors on one GPU?
Why use water tanks from a retired Space Shuttle?
Can a wire having a 610-670 THz (frequency of blue light) AC frequency supply, generate blue light?
Where can I find the list of all tendons in the human body?
Singlequote and backslash
The term for the person/group a political party aligns themselves with to appear concerned about the general public
Can I ask a publisher for a paper that I need for reviewing
How can I offer a test ride while selling a bike?
Is there an evolutionary advantage to having two heads?
Group manager threw me under the bus to save her friend
Does nuclear propulsion applied to military ships fall under civil or military nuclear power?
Grep complete name including dot in the word
Explain Ant-Man's "not it" scene from Avengers: Endgame
'chmod' would set file permission to 000 no matter what permission I try to set
arcpy.GetParameterAsText not passing arguments to script?
What does War Machine's "Canopy! Canopy!" line mean in "Avengers: Endgame"?
How can I stop my presentation being derailed by audience questions?
Did airlines fly their aircraft slower in response to oil prices in the 1970s?
What if you don't bring your credit card or debit for incidentals?
Strange math syntax in old basic listing
Can you use a concentration spell while using Mantle of Majesty?
Why does the UK have more political parties than the US?
Change TeXForm of ArcTan
Double integral bounds of integration polar change of coordinate
Looking for an old image of designing a cpu with plan laid out / being edited on a literal floor
Should I set `inter_op_parallelism_threads` and `intra_op_parallelism_threads` to 1 when I use ray to create a actor?
How can I safely create a nested directory?Why are Python lambdas useful?“Least Astonishment” and the Mutable Default ArgumentShould I use 'has_key()' or 'in' on Python dicts?In Python, when to use a Dictionary, List or Set?Python string formatting: % vs. .formatInstalling specific package versions with pipWay to create multiline comments in Python?How to change the order of DataFrame columns?Ray: How to run many actors on one GPU?
.everyoneloves__top-leaderboard:empty,.everyoneloves__mid-leaderboard:empty,.everyoneloves__bot-mid-leaderboard:empty height:90px;width:728px;box-sizing:border-box;
When I use ray to create the actors and learner in A2C, should I set inter_op_parallelism_threads and intra_op_parallelism_threads to 1, or leave it by default?
python tensorflow ray
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
add a comment |
When I use ray to create the actors and learner in A2C, should I set inter_op_parallelism_threads and intra_op_parallelism_threads to 1, or leave it by default?
python tensorflow ray
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
add a comment |
When I use ray to create the actors and learner in A2C, should I set inter_op_parallelism_threads and intra_op_parallelism_threads to 1, or leave it by default?
python tensorflow ray
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
When I use ray to create the actors and learner in A2C, should I set inter_op_parallelism_threads and intra_op_parallelism_threads to 1, or leave it by default?
python tensorflow ray
python tensorflow ray
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
asked Mar 24 at 11:08
Sherwin ChenSherwin Chen
410415
410415
add a comment |
add a comment |
1 Answer
1
active
oldest
votes
It depends how many resources you want to the actor to use. If there is a dedicated machine for a given actor, and it's ok for the actor to use all of the resources on that machine, then use TensorFlow's default settings. If you are creating more like one actor per core, then setting inter_op_parallelism_threads and intra_op_parallelism_threads to small values like 1 or 2 is a good idea.
In general, you can try both approaches and see which is faster.
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Yes, though it's worth trying settinginter_op_parallelism_threadsto 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.
– Robert Nishihara
Mar 25 at 22:10
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors. If I've clearly setnum_cpus=1`, would there still be any contention between the actors?
– Sherwin Chen
Mar 26 at 0:13
Yes there can still be contention, when you specifynum_cpus=1in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?
– Robert Nishihara
Mar 26 at 5:52
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors withnum_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified inray.init. But it does not say anything about how much cpu resources an actor are allowed to use.
– Sherwin Chen
Mar 26 at 7:04
|
show 1 more comment
Your Answer
StackExchange.ifUsing("editor", function ()
StackExchange.using("externalEditor", function ()
StackExchange.using("snippets", function ()
StackExchange.snippets.init();
);
);
, "code-snippets");
StackExchange.ready(function()
var channelOptions =
tags: "".split(" "),
id: "1"
;
initTagRenderer("".split(" "), "".split(" "), channelOptions);
StackExchange.using("externalEditor", function()
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled)
StackExchange.using("snippets", function()
createEditor();
);
else
createEditor();
);
function createEditor()
StackExchange.prepareEditor(
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: true,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: 10,
bindNavPrevention: true,
postfix: "",
imageUploader:
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
,
onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
);
);
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function ()
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f55323169%2fshould-i-set-inter-op-parallelism-threads-and-intra-op-parallelism-threads-t%23new-answer', 'question_page');
);
Post as a guest
Required, but never shown
1 Answer
1
active
oldest
votes
1 Answer
1
active
oldest
votes
active
oldest
votes
active
oldest
votes
It depends how many resources you want to the actor to use. If there is a dedicated machine for a given actor, and it's ok for the actor to use all of the resources on that machine, then use TensorFlow's default settings. If you are creating more like one actor per core, then setting inter_op_parallelism_threads and intra_op_parallelism_threads to small values like 1 or 2 is a good idea.
In general, you can try both approaches and see which is faster.
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Yes, though it's worth trying settinginter_op_parallelism_threadsto 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.
– Robert Nishihara
Mar 25 at 22:10
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors. If I've clearly setnum_cpus=1`, would there still be any contention between the actors?
– Sherwin Chen
Mar 26 at 0:13
Yes there can still be contention, when you specifynum_cpus=1in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?
– Robert Nishihara
Mar 26 at 5:52
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors withnum_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified inray.init. But it does not say anything about how much cpu resources an actor are allowed to use.
– Sherwin Chen
Mar 26 at 7:04
|
show 1 more comment
It depends how many resources you want to the actor to use. If there is a dedicated machine for a given actor, and it's ok for the actor to use all of the resources on that machine, then use TensorFlow's default settings. If you are creating more like one actor per core, then setting inter_op_parallelism_threads and intra_op_parallelism_threads to small values like 1 or 2 is a good idea.
In general, you can try both approaches and see which is faster.
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Yes, though it's worth trying settinginter_op_parallelism_threadsto 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.
– Robert Nishihara
Mar 25 at 22:10
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors. If I've clearly setnum_cpus=1`, would there still be any contention between the actors?
– Sherwin Chen
Mar 26 at 0:13
Yes there can still be contention, when you specifynum_cpus=1in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?
– Robert Nishihara
Mar 26 at 5:52
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors withnum_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified inray.init. But it does not say anything about how much cpu resources an actor are allowed to use.
– Sherwin Chen
Mar 26 at 7:04
|
show 1 more comment
It depends how many resources you want to the actor to use. If there is a dedicated machine for a given actor, and it's ok for the actor to use all of the resources on that machine, then use TensorFlow's default settings. If you are creating more like one actor per core, then setting inter_op_parallelism_threads and intra_op_parallelism_threads to small values like 1 or 2 is a good idea.
In general, you can try both approaches and see which is faster.
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
It depends how many resources you want to the actor to use. If there is a dedicated machine for a given actor, and it's ok for the actor to use all of the resources on that machine, then use TensorFlow's default settings. If you are creating more like one actor per core, then setting inter_op_parallelism_threads and intra_op_parallelism_threads to small values like 1 or 2 is a good idea.
In general, you can try both approaches and see which is faster.
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
answered Mar 24 at 22:05
Robert NishiharaRobert Nishihara
85569
85569
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Yes, though it's worth trying settinginter_op_parallelism_threadsto 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.
– Robert Nishihara
Mar 25 at 22:10
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors. If I've clearly setnum_cpus=1`, would there still be any contention between the actors?
– Sherwin Chen
Mar 26 at 0:13
Yes there can still be contention, when you specifynum_cpus=1in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?
– Robert Nishihara
Mar 26 at 5:52
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors withnum_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified inray.init. But it does not say anything about how much cpu resources an actor are allowed to use.
– Sherwin Chen
Mar 26 at 7:04
|
show 1 more comment
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Yes, though it's worth trying settinginter_op_parallelism_threadsto 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.
– Robert Nishihara
Mar 25 at 22:10
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors. If I've clearly setnum_cpus=1`, would there still be any contention between the actors?
– Sherwin Chen
Mar 26 at 0:13
Yes there can still be contention, when you specifynum_cpus=1in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?
– Robert Nishihara
Mar 26 at 5:52
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors withnum_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified inray.init. But it does not say anything about how much cpu resources an actor are allowed to use.
– Sherwin Chen
Mar 26 at 7:04
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Thanks, so if I create an actor per core, then both should be set to 1?
– Sherwin Chen
Mar 25 at 10:04
Yes, though it's worth trying setting
inter_op_parallelism_threads to 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.– Robert Nishihara
Mar 25 at 22:10
Yes, though it's worth trying setting
inter_op_parallelism_threads to 2 as well to see how that performs. Depending on what the actors are actually doing (e.g., are some of them idle some of the time), that may perform better. However, if they're all busy the whole time, then 1 will prevent too much contention between the actors.– Robert Nishihara
Mar 25 at 22:10
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors
. If I've clearly set num_cpus=1`, would there still be any contention between the actors?– Sherwin Chen
Mar 26 at 0:13
Thanks for your patience. I now understand most of your answer, except that "1 will prevent too much contention between the actors
. If I've clearly set num_cpus=1`, would there still be any contention between the actors?– Sherwin Chen
Mar 26 at 0:13
Yes there can still be contention, when you specify
num_cpus=1 in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?– Robert Nishihara
Mar 26 at 5:52
Yes there can still be contention, when you specify
num_cpus=1 in the remote decorator, the Ray scheduler will make sure not to create more than one such actor per core. However, the actor could still be heavily multithreaded and could still use all of the available CPU cores. Ray does not actually prevent the actor from using all of the resources on the machine. Does that make sense?– Robert Nishihara
Mar 26 at 5:52
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors with
num_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified in ray.init. But it does not say anything about how much cpu resources an actor are allowed to use.– Sherwin Chen
Mar 26 at 7:04
I see. Please allow me to repeat it and see if I understand it right. If I decorate all actors with
num_cpus=1, the maximum number of actor allowed to be created is the number of cpus specified in ray.init. But it does not say anything about how much cpu resources an actor are allowed to use.– Sherwin Chen
Mar 26 at 7:04
|
show 1 more comment
Thanks for contributing an answer to Stack Overflow!
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
To learn more, see our tips on writing great answers.
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function ()
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstackoverflow.com%2fquestions%2f55323169%2fshould-i-set-inter-op-parallelism-threads-and-intra-op-parallelism-threads-t%23new-answer', 'question_page');
);
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
