| Pages: [1] :: one page |
|
|
| Author |
Thread Statistics | Show CCP posts - 1 post(s) |
Zendoren
Aktaeon Industries
   |
Posted - 2010.11.18 03:05:00 -
[1]
Edited by: Zendoren on 18/11/2010 03:07:51
TL;DR: Can NividaÆs Video Cards Help reduce server lag???
I was reading over one of the DevBlogs about lag and the Destiny physics engine starving out other processes within EVE and a question came to mind. Why not build in support for NvidiaÆs new CUDA processing architecture? From what I have read so far, this is a perfect example of what CUDA was meant for. Don't get me wrong, I love ATI and am an avid user of their cards but, IÆve been paying attention and notice recently that CCP has been working closely with Nvidia on their Carbon architecture for Incarna and WoD. So i figured why not work with what you know! Theoretically this will add more processing power to the servers without having to add more nodes and large hardware commitments.
Leave a comment if you know what im talking about! 
|
Nova Fox
Gallente
Novafox Shipyards
|
Posted - 2010.11.18 03:09:00 -
[2]
because there are people who are proud owners of AMD gpus.
Pre-order your Sisters of ≡v≡ Exploration ship today, Updated 1SEPT10
|
Zendoren
Aktaeon Industries
|
Posted - 2010.11.18 03:11:00 -
[3]
 Originally by: Nova Fox
because there are people who are proud owners of AMD gpus.
This is a server side implementation not on the clients!
|
Treiq
|
Posted - 2010.11.18 04:36:00 -
[4]
Edited by: Treiq on 18/11/2010 04:36:46
ATI got direct compute, same thing really...
GPU and CPU does diffrent jobs, you cant use one to replace the other.
+doubt they want to rewrite the whole server code.
|
Linda Duane
Gallente
Bluestar Enterprises
|
Posted - 2010.11.18 04:51:00 -
[5]
Edited by: Linda Duane on 18/11/2010 04:54:01
Originally by: Zendoren
Originally by: Nova Fox
because there are people who are proud owners of AMD gpus.
This is a server side implementation not on the clients!
You shouldn't only read the dev blogs, but also the comments of all the players in the accompanying thread.
Destiny runs on the server AND on the client to bring you a smooth experience, although you aren't connected all the time (this concept let's gap 30secs and more without you noticing). If your client get's off track the servers info gets taken over and your client can keep going on from there.
CCP won't program destiny for CUDA and x68_64 .. not enough manpower.
Also by sticking with one codebase the results of the server and client will match.
And on another page.. Team Gridlock already is looking into getting more processes off from the nodes and onto other nodes.. like they did the last time with the Charackter Nodes.
The other thing they are working on is to break solar systems off, so that there can be a destiny engine for every grid (afaiu)
|
Viktor Fyretracker
Caldari
Fyretracker Heavy Industries
|
Posted - 2010.11.18 05:00:00 -
[6]
But can CCP control when Destiny goes into FTL. =)
|
Khalia Nestune
Honorless Internet Jerks
|
Posted - 2010.11.18 05:10:00 -
[7]
Originally by: Viktor Fyretracker
But can CCP control when Destiny goes into FTL. =)
Yes, but they can only stop to explore the new system for 8 hours...
|
|
CCP Jericho
|
Posted - 2010.11.18 05:44:00 -
[8]
Moved from EVE General Discussion.
|
|
Rasz Lin
Caldari
Racketeers
|
Posted - 2010.11.18 07:49:00 -
[9]
Ur funny man, CCP cant even write multithreated code and you want CUDA? :))))
|
Blasphemour
|
Posted - 2010.11.18 07:54:00 -
[10]
Originally by: Rasz Lin
Ur funny man, CCP cant even write multithreated code and you want CUDA? :))))
They probably can but it would mean a complete re-write of the server code... that would take a lot of time, in which they can not fix bugs, add stuff, fix lag, etc.
|
|
|
Green Shift
Mad Bombers
HYDRA RELOADED
|
Posted - 2010.11.18 08:12:00 -
[11]
It isn't that simple. Having x more computing power does not mean faster in this age.
I'll try to explain it.
CPUs have an architecture that is designed to receive an instruction and process it as fast as possible.
GPUs have an architecture that is designed around getting a huge amount of identical instruction (but most likely not identical data) and running them in parallel.
For instance a CPU runs our programs because we do 1 instruction after the other in order they are received.
GPU does our graphic processing because we often need to do 1(or more) identical operation per pixel that we want to display on our screen but you don't care at all what order it is done in. (SIMD if you want to read further)
Something like a physic engine can be implemented in gpu computing because we know at the start of our operation what our data is eg. All the ships on grid, and we do this a set amount per seconds so it is entirely server based (eg people have no control about when the server does its physics simulation you only get told which direction your moving in and what speed) and the calculations are identical (a formula). However this still does not mean eve would benefit from it.
Something like damage registering is not good for gpu computing because the order damage comes in matters. We do not know how much damage comes into 1 ship at any given point in time as it depends on how many people are firing and it is often better to work on this dynamically. Eg when someone fires then you generate damage done followed by reduction in health of the victim.
Also another factor is what your current code is like. It is quite easy to implement these things in from a scratch build but eve is built on existing code (quite old and patched code) so improvements have to be designed very carefully.
There is platform independent version of this kind of technology and it's called openCL.
|
AterraX
Caldari
|
Posted - 2010.11.18 09:02:00 -
[12]
Originally by: Green Shift
There is platform independent version of this kind of technology and it's called openCL.
Comparing CUDA too OpenCL in their current state is like comparing motorbikes too chickens...world of difference.
And don't forget PhysX/APEX run via CUDA.
____________________________________________________________________________________________________
Fact of EVE forums: They will always come an anounomys alt-toon and question someones character... |
Kolatha
|
Posted - 2010.11.18 11:06:00 -
[13]
Edited by: Kolatha on 18/11/2010 11:06:42
Blade servers typically look something like this
Sourced from this blog.
As you can see, no space for a video card, that's why not.
|
Grez
Neo Spartans
Laconian Syndicate
|
Posted - 2010.11.18 21:12:00 -
[14]
Direct Compute > CUDA.
---
|
Vogue
Skynet Nexus
|
Posted - 2010.11.18 21:32:00 -
[15]
The world's fastest supercomputer mostly uses NVIDIA GPU's as well as Intel Xeon's - The Chinese Tianhe-1A. Mostly it has been USA supercomputers at #1. Japan has been the only other country to take #1 with a NEC system some years ago.
If this is one of those 'national prestige' areas then USA will try and build a far more powerful supercomputer than the Tianhe-1A. USA by far dominates the top500 list.
There are export restrictions on 'high tech' to China. What is this? As generic use CPU\GPU's I assume are available at any quantity all over the world. This restrictions are linked to weapon systems.
..................................................
|
Rasz Lin
Caldari
Caldari Provisions
|
Posted - 2010.11.20 04:19:00 -
[16]
Originally by: AterraX
Comparing CUDA too OpenCL in their current state is like comparing motorbikes too chickens...world of difference.
more like C vs Java
Originally by: AterraX
And don't forget PhysX/APEX run via CUDA.
dont use "run" and PhysX/APEX in same sentence, its more of a walk with one broken leg, almost crawl, for reference try Mafia 2 :/ (<30fps with APEX, >80 without, on FASTEST possible Nvidia card). I fully expect "walking in stations" (if we ever see it) to also have same performance problems due to PhysX/APEX used for clothing.
|
Zendoren
Aktaeon Industries
|
Posted - 2010.11.20 05:06:00 -
[17]
Originally by: Green Shift
It isn't that simple. Having x more computing power does not mean faster in this age.
I'll try to explain it.
CPUs have an architecture that is designed to receive an instruction and process it as fast as possible.
GPUs have an architecture that is designed around getting a huge amount of identical instruction (but most likely not identical data) and running them in parallel.
For instance a CPU runs our programs because we do 1 instruction after the other in order they are received.
GPU does our graphic processing because we often need to do 1(or more) identical operation per pixel that we want to display on our screen but you don't care at all what order it is done in. (SIMD if you want to read further)
Something like a physic engine can be implemented in gpu computing because we know at the start of our operation what our data is eg. All the ships on grid, and we do this a set amount per seconds so it is entirely server based (eg people have no control about when the server does its physics simulation you only get told which direction your moving in and what speed) and the calculations are identical (a formula). However this still does not mean eve would benefit from it.
Something like damage registering is not good for gpu computing because the order damage comes in matters. We do not know how much damage comes into 1 ship at any given point in time as it depends on how many people are firing and it is often better to work on this dynamically. Eg when someone fires then you generate damage done followed by reduction in health of the victim.
Also another factor is what your current code is like. It is quite easy to implement these things in from a scratch build but eve is built on existing code (quite old and patched code) so improvements have to be designed very carefully.
There is platform independent version of this kind of technology and it's called openCL.
This wont be such a big task if the code is OOP.
I know that python code is not normally OOP but sequential; however, you have to admit that OOP makes things easier at the end of the day!
Anywho, they have to re-factor the code some day. Might as well be sooner than latter.
|
alittlebirdy
|
Posted - 2010.11.20 23:15:00 -
[18]
Originally by: Blasphemour
Originally by: Rasz Lin
Ur funny man, CCP cant even write multithreated code and you want CUDA? :))))
They probably can but it would mean a complete re-write of the server code... that would take a lot of time, in which they can not fix bugs, add stuff, fix lag, etc.
LOL cause they do that now right?
|
|
|
| |
|
| Pages: [1] :: one page |
| First page | Previous page | Next page | Last page |