What causes GPU global memory latency?
I was asking myself what causes GPU global memory latency. I mean the global memory works e.g. on a frequency of 1700MHz (e.g.) and has a 352bit lane, this means it can move around 600GB/s. But then everyone states, this is the slowest memory on a GPU and you have around 200 - 800 cycles access time. If your core works with around 1000MHz this means you have around 2us per access, which is much slower than 600GB/s.
Where does this difference come from?
memory gpu latency global
add a comment |
I was asking myself what causes GPU global memory latency. I mean the global memory works e.g. on a frequency of 1700MHz (e.g.) and has a 352bit lane, this means it can move around 600GB/s. But then everyone states, this is the slowest memory on a GPU and you have around 200 - 800 cycles access time. If your core works with around 1000MHz this means you have around 2us per access, which is much slower than 600GB/s.
Where does this difference come from?
memory gpu latency global
Bandwidth is different from latency. Because GPUs prioritize bandwidth their memory controllers buffer requests and schedule them to reduce idle channel time. Multiple requests can be in-process concurrently; while one request is opening a row in a bank of one DRAM another can be transferring data and many others can be waiting. (Also 200 cycles at 1 GHz is only 200ns not 2µs.) The question deserve a real answer, but this is a quick hint.
– Paul A. Clayton
Jan 27 at 20:13
add a comment |
I was asking myself what causes GPU global memory latency. I mean the global memory works e.g. on a frequency of 1700MHz (e.g.) and has a 352bit lane, this means it can move around 600GB/s. But then everyone states, this is the slowest memory on a GPU and you have around 200 - 800 cycles access time. If your core works with around 1000MHz this means you have around 2us per access, which is much slower than 600GB/s.
Where does this difference come from?
memory gpu latency global
I was asking myself what causes GPU global memory latency. I mean the global memory works e.g. on a frequency of 1700MHz (e.g.) and has a 352bit lane, this means it can move around 600GB/s. But then everyone states, this is the slowest memory on a GPU and you have around 200 - 800 cycles access time. If your core works with around 1000MHz this means you have around 2us per access, which is much slower than 600GB/s.
Where does this difference come from?
memory gpu latency global
memory gpu latency global
asked Jan 25 at 6:02
Emerson FittipaldiEmerson Fittipaldi
1
1
Bandwidth is different from latency. Because GPUs prioritize bandwidth their memory controllers buffer requests and schedule them to reduce idle channel time. Multiple requests can be in-process concurrently; while one request is opening a row in a bank of one DRAM another can be transferring data and many others can be waiting. (Also 200 cycles at 1 GHz is only 200ns not 2µs.) The question deserve a real answer, but this is a quick hint.
– Paul A. Clayton
Jan 27 at 20:13
add a comment |
Bandwidth is different from latency. Because GPUs prioritize bandwidth their memory controllers buffer requests and schedule them to reduce idle channel time. Multiple requests can be in-process concurrently; while one request is opening a row in a bank of one DRAM another can be transferring data and many others can be waiting. (Also 200 cycles at 1 GHz is only 200ns not 2µs.) The question deserve a real answer, but this is a quick hint.
– Paul A. Clayton
Jan 27 at 20:13
Bandwidth is different from latency. Because GPUs prioritize bandwidth their memory controllers buffer requests and schedule them to reduce idle channel time. Multiple requests can be in-process concurrently; while one request is opening a row in a bank of one DRAM another can be transferring data and many others can be waiting. (Also 200 cycles at 1 GHz is only 200ns not 2µs.) The question deserve a real answer, but this is a quick hint.
– Paul A. Clayton
Jan 27 at 20:13
Bandwidth is different from latency. Because GPUs prioritize bandwidth their memory controllers buffer requests and schedule them to reduce idle channel time. Multiple requests can be in-process concurrently; while one request is opening a row in a bank of one DRAM another can be transferring data and many others can be waiting. (Also 200 cycles at 1 GHz is only 200ns not 2µs.) The question deserve a real answer, but this is a quick hint.
– Paul A. Clayton
Jan 27 at 20:13
add a comment |
0
active
oldest
votes
Your Answer
StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "3"
};
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%2fsuperuser.com%2fquestions%2f1398212%2fwhat-causes-gpu-global-memory-latency%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
0
active
oldest
votes
0
active
oldest
votes
active
oldest
votes
active
oldest
votes
Thanks for contributing an answer to Super User!
- 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%2fsuperuser.com%2fquestions%2f1398212%2fwhat-causes-gpu-global-memory-latency%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
Bandwidth is different from latency. Because GPUs prioritize bandwidth their memory controllers buffer requests and schedule them to reduce idle channel time. Multiple requests can be in-process concurrently; while one request is opening a row in a bank of one DRAM another can be transferring data and many others can be waiting. (Also 200 cycles at 1 GHz is only 200ns not 2µs.) The question deserve a real answer, but this is a quick hint.
– Paul A. Clayton
Jan 27 at 20:13