- android - 多次调用 OnPrimaryClipChangedListener
- android - 无法更新 RecyclerView 中的 TextView 字段
- android.database.CursorIndexOutOfBoundsException : Index 0 requested, 光标大小为 0
- android - 使用 AppCompat 时,我们是否需要明确指定其 UI 组件(Spinner、EditText)颜色
我尝试使用 JOGL 和 GLSL 在 GPU 上实现 react 扩散模型。
我使用 2 个 FramebufferObject 的乒乓球技术(我也尝试过使用 1 个 FBO 和 2 个颜色附件,但没有成功)。着色器似乎是正确的,因为我已经尝试过统一(经过一些调整)并且它有效。经过一周的尝试后,我完全不知道如何让这段代码发挥作用。我真的不是 JOGL 专家,所以也许我错过了一些明显的东西。
结果是图像随着时间的推移而变白:没有反应扩散行为,我不明白为什么!
预先感谢您的帮助。这是我的代码:
package gpu2;
import java.awt.*;
import java.awt.event.*;
import javax.swing.*;
import java.nio.IntBuffer;
import java.nio.FloatBuffer;
import java.io.File;
import com.jogamp.opengl.GL2;
import com.jogamp.opengl.GLAutoDrawable;
import com.jogamp.opengl.GLEventListener;
import com.jogamp.opengl.awt.GLCanvas;
import com.jogamp.opengl.glu.GLU;
import com.jogamp.opengl.util.FPSAnimator;
import com.jogamp.opengl.GLFBODrawable;
import com.jogamp.opengl.FBObject;
import com.jogamp.opengl.FBObject.Colorbuffer;
import com.jogamp.opengl.FBObject.ColorAttachment;
import com.jogamp.opengl.FBObject.TextureAttachment;
import com.jogamp.opengl.util.glsl.ShaderCode;
import com.jogamp.opengl.util.glsl.ShaderProgram;
import com.jogamp.opengl.util.glsl.ShaderUtil;
import com.jogamp.opengl.util.GLBuffers;
import com.jogamp.opengl.util.texture.Texture;
import com.jogamp.opengl.util.texture.TextureIO;
import com.jogamp.opengl.GLCapabilities;
import com.jogamp.opengl.GLOffscreenAutoDrawable;
import com.jogamp.opengl.GLProfile;
import com.jogamp.opengl.util.awt.AWTGLReadBufferUtil;
import com.jogamp.opengl.GLDrawableFactory;
import static com.jogamp.opengl.GL.*; // GL constants
import static com.jogamp.opengl.GL2.*; // GL2 constants
import gpu2.ModelParam;
/**
* JOGL 2.0 Program Template (GLCanvas)
* This is a "Component" which can be added into a top-level "Container".
* It also handles the OpenGL events to render graphics.
*/
@SuppressWarnings("serial")
public class JOGL2Setup_GLCanvas extends GLCanvas implements GLEventListener {
// Define constants for the top-level container
private static String TITLE = "JOGL 2.0 Setup (GLCanvas)"; // window's title
private static final int CANVAS_WIDTH = 512; // width of the drawable
private static final int CANVAS_HEIGHT = 512; // height of the drawable
private static final int FPS = 30; // animator's target frames per second
private final float[] canvasVertices = {
-1f, -1f, 0.0f,
-1f, 1f, 0.0f,
1f, -1f, 0.0f,
1f, 1f, 0.0f,
};
private final float[] canvasTexCoords = {
0.0f, 0.0f,
0.0f, 1.0f,
1.0f, 0.0f,
1.0f, 1.0f,
};
/** The entry main() method to setup the top-level container and animator */
public static void main(String[] args) {
// Run the GUI codes in the event-dispatching thread for thread safety
SwingUtilities.invokeLater(new Runnable() {
@Override
public void run() {
// Create the OpenGL rendering canvas
GLCanvas canvas = new JOGL2Setup_GLCanvas();
canvas.setPreferredSize(new Dimension(CANVAS_WIDTH, CANVAS_HEIGHT));
// Create a animator that drives canvas' display() at the specified FPS.
final FPSAnimator animator = new FPSAnimator(canvas, FPS, true);
// Create the top-level container
final JFrame frame = new JFrame(); // Swing's JFrame or AWT's Frame
frame.getContentPane().add(canvas);
frame.addWindowListener(new WindowAdapter() {
@Override
public void windowClosing(WindowEvent e) {
// Use a dedicate thread to run the stop() to ensure that the
// animator stops before program exits.
new Thread() {
@Override
public void run() {
if (animator.isStarted()) animator.stop();
System.exit(0);
}
}.start();
}
});
frame.setTitle(TITLE);
frame.pack();
frame.setVisible(true);
animator.start(); // start the animation loop
}
});
}
// Setup OpenGL Graphics Renderer
private GLU glu; // for the GL Utility
private GL2 gl;
//OpenGl data
private int vboVertices;
private int vboTextCoord;
private Texture textureFile;
private FBObject fbo[];
private ShaderProgram shaderCompute;
private ShaderProgram shaderVisu;
private ShaderProgram shaderComputeInit;
private int currentSourceBuffer = 0;
private int currentDestBuffer = 1;
private int currentFrame = 0;
private int maxFrameCount = 5000000;
private float clearUniform = 0;
ModelParam params = new ModelParam();
public JOGL2Setup_GLCanvas() {
this.addGLEventListener(this);
}
@Override
public void init(GLAutoDrawable drawable) {
gl = drawable.getGL().getGL2(); // get the OpenGL graphics context
glu = new GLU(); // get GL Utilities
gl.glClearColor(0.0f, 0.0f, 0.0f, 0.0f); // set background (clear) color
gl.glEnable(GL_TEXTURE_2D);
gl.glEnable( GL_COLOR_MATERIAL );
gl.glEnable( GL_FRAMEBUFFER );
gl.glHint(GL_PERSPECTIVE_CORRECTION_HINT, GL_NEAREST); // best perspective correction
viewOrtho(gl);
gl.glViewport(0,0,CANVAS_WIDTH,CANVAS_HEIGHT);
int[] buffers = new int[2];
gl.glGenBuffers(2, buffers, 0);
vboVertices = buffers[0];
vboTextCoord = buffers[1];
gl.glBindBuffer(GL_ARRAY_BUFFER, vboVertices);
gl.glBufferData(GL_ARRAY_BUFFER, canvasVertices.length*(Float.SIZE/Byte.SIZE)*3, FloatBuffer.wrap(canvasVertices), GL_STATIC_DRAW);
gl.glBindBuffer(GL_ARRAY_BUFFER, vboTextCoord);
gl.glBufferData(GL_ARRAY_BUFFER, canvasTexCoords.length*(Float.SIZE/Byte.SIZE)*2, FloatBuffer.wrap(canvasTexCoords), GL_STATIC_DRAW);
gl.glBindBuffer( GL_ARRAY_BUFFER, 0 );
// ------------ create Texture Source------------------------
textureFile = initializeTexture();
if (textureFile==null) {
System.out.println("cannot load texture from disk");
}
// ------------ load shaders ------------------------
shaderCompute = loadShader(gl, "compute.vsh", "compute.fsh");
shaderComputeInit = loadShader(gl, "compute.vsh", "computeInit.fsh");
shaderVisu = loadShader(gl, "visu.vsh", "visu.fsh");
// ------------ create FBO ------------------------
initFBO();
}
/**
* Called back by the animator to perform rendering.
*/
@Override
public void display(GLAutoDrawable drawable) {
if (currentFrame < maxFrameCount) {
prepareNextStep();
renderToFBO();
currentFrame++;
}
renderFBOToScreen();
}
private void prepareNextStep() {
currentSourceBuffer = 1 - currentSourceBuffer;
currentDestBuffer = 1 - currentDestBuffer;
}
private void renderToFBO()
{
fbo[currentDestBuffer].bind(gl);
//gl.glClear(GL_COLOR_BUFFER_BIT);
viewOrtho(gl);
shaderCompute.useProgram(gl, true);
setShaderUniformFloat(gl, shaderCompute.program(), "diffuseU", 0.211f);
setShaderUniformFloat(gl, shaderCompute.program(), "diffuseV", 0.088f);
setShaderUniformFloat(gl, shaderCompute.program(), "feed", 0.007f);
setShaderUniformFloat(gl, shaderCompute.program(), "kill", 0.08f);
setShaderUniformFloat(gl, shaderCompute.program(), "Tech", 1f);
setShaderUniformFloat(gl, shaderCompute.program(), "currentFrame", currentFrame);
setShaderUniformFloat2(gl, shaderCompute.program(), "resolution", CANVAS_WIDTH, CANVAS_HEIGHT);
drawDataBuffer(shaderCompute, true);
shaderCompute.useProgram(gl, false);
fbo[currentDestBuffer].unbind(gl);
}
void drawDataBuffer(ShaderProgram currentShader, boolean sencondImage) {
// --- draw vbo
gl.glEnableClientState(GL_VERTEX_ARRAY);
gl.glEnableClientState(GL_TEXTURE_COORD_ARRAY);
//textcoords
gl.glBindBuffer(GL_ARRAY_BUFFER,vboTextCoord);
gl.glTexCoordPointer(2, GL_FLOAT, 0, 0);
//vertices
gl.glBindBuffer( GL_ARRAY_BUFFER, vboVertices );
gl.glVertexPointer(3, GL_FLOAT, 0, 0);
//activate texture data from last fbo
final FBObject.Colorbuffer texSource = (FBObject.Colorbuffer) fbo[currentSourceBuffer].getColorbuffer(0);
gl.glActiveTexture(GL_TEXTURE0);
gl.glBindTexture(GL_TEXTURE_2D, texSource.getName());
setShaderUniform1i(gl, currentShader.program(), "textureData", 0);
if (sencondImage) {
//activate texture with image from file
gl.glActiveTexture(GL_TEXTURE1);
gl.glBindTexture(GL_TEXTURE_2D, textureFile.getTextureObject());
textureFile.bind(gl);
setShaderUniform1i(gl, currentShader.program(), "textureImage", 1);
}
//draw buffer on screens
gl.glDrawArrays(GL_TRIANGLE_STRIP, 0, canvasVertices.length / 3);
//disable texture image
if (sencondImage) {
gl.glActiveTexture(GL_TEXTURE1);
textureFile.disable(gl);
}
//disable texture data
gl.glActiveTexture(GL_TEXTURE0);
gl.glDisable(texSource.getName());
gl.glDisableClientState(GL_TEXTURE_COORD_ARRAY);
gl.glDisableClientState(GL_VERTEX_ARRAY);
}
public void renderFBOToScreen()
{
gl.glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT); // clear color and depth buffers
gl.glLoadIdentity(); // reset the model-view matrix
viewOrtho(gl);
gl.glEnable(GL_TEXTURE_2D);
final FBObject.Colorbuffer tex0 = (FBObject.Colorbuffer) fbo[currentDestBuffer].getColorbuffer(0);
gl.glActiveTexture(GL_TEXTURE0);
gl.glBindTexture(GL_TEXTURE_2D, tex0.getName());
//activate shader
shaderVisu.useProgram(gl, true);
// --- draw vbo
gl.glEnableClientState(GL_VERTEX_ARRAY);
gl.glEnableClientState(GL_TEXTURE_COORD_ARRAY);
//textcoords
gl.glBindBuffer(GL_ARRAY_BUFFER, vboTextCoord);
gl.glTexCoordPointer(2, GL_FLOAT, 0, 0);
//vertices
gl.glBindBuffer( GL_ARRAY_BUFFER, vboVertices );
gl.glVertexPointer(3, GL_FLOAT, 0, 0);
//draw buffer on screens
gl.glDrawArrays(GL_TRIANGLE_STRIP, 0, canvasVertices.length / 3);
gl.glDisableClientState(GL_TEXTURE_COORD_ARRAY);
gl.glDisableClientState(GL_VERTEX_ARRAY);
gl.glBindBuffer( GL_ARRAY_BUFFER, 0 );
//desactivate shader
shaderVisu.useProgram(gl, false);
}
private void initFBO()
{
try {
gl.glEnable(GL_TEXTURE_2D);
fbo = new FBObject[2];
//first fbo
fbo[currentSourceBuffer] = new FBObject(); // Create FrameBuffer
fbo[currentSourceBuffer].init(gl, CANVAS_WIDTH, CANVAS_HEIGHT, 0);
fbo[currentSourceBuffer].reset(gl, CANVAS_WIDTH, CANVAS_HEIGHT, 0); // int width, height - size of FBO, can be resized with the same call
fbo[currentSourceBuffer].bind(gl);
int tex = genTexture(gl);
gl.glBindTexture(GL_TEXTURE_2D, tex);
gl.glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA32F, CANVAS_WIDTH, CANVAS_HEIGHT, 0, GL_RGBA, GL_FLOAT, null);
fbo[currentSourceBuffer].attachTexture2D(gl, 0, GL_RGBA32F, GL_RGBA, GL_FLOAT, GL_NEAREST, GL_NEAREST, GL_CLAMP_TO_EDGE, GL_CLAMP_TO_EDGE);
//gl.glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, tex, 0);
int DrawBuffers[] = {GL_COLOR_ATTACHMENT0};
gl.glDrawBuffers(1, DrawBuffers, 0); // "1" is the size of DrawBuffers
fbo[currentSourceBuffer].unbind(gl);
//second fbo
fbo[currentDestBuffer] = new FBObject(); // Create FrameBuffer
fbo[currentDestBuffer].init(gl, CANVAS_WIDTH, CANVAS_HEIGHT, 0);
fbo[currentDestBuffer].reset(gl, CANVAS_WIDTH, CANVAS_HEIGHT, 0); // int width, height - size of FBO, can be resized with the same call
fbo[currentDestBuffer].bind(gl);
tex = genTexture(gl);
gl.glBindTexture(GL_TEXTURE_2D, tex);
gl.glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA32F, CANVAS_WIDTH, CANVAS_HEIGHT, 0, GL_RGBA, GL_FLOAT, null);
fbo[currentDestBuffer].attachTexture2D(gl, 0, GL_RGBA32F, GL_RGBA, GL_FLOAT, GL_NEAREST, GL_NEAREST, GL_CLAMP_TO_EDGE, GL_CLAMP_TO_EDGE);
//ogl.glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, tex, 0);
gl.glDrawBuffers(1, DrawBuffers, 1); // "1" is the size of DrawBuffers
fbo[currentDestBuffer].unbind(gl);
} catch (Exception e) {
System.out.println("Problem with fbo init " + e);
e.printStackTrace();
}
}
private Texture initializeTexture() {
Texture t = null;
try {
t = TextureIO.newTexture(new File("e:/shaders/wiki.jpg"), false);
t.setTexParameteri(gl, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
t.setTexParameteri(gl, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
t.setTexParameteri(gl, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
t.setTexParameteri(gl, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
} catch (Exception e) {
System.out.println("Unable to read texture file: " + e);
e.printStackTrace();
}
return t;
}
private ShaderProgram loadShader(GL2 gl, String vertexShader, String fragmentShader)
{
ShaderCode vertShader = ShaderCode.create(gl, GL2.GL_VERTEX_SHADER, 1, getClass(), new String[]{"e:/shaders/"+vertexShader},false);
vertShader.compile(gl);
ShaderCode fragShader = ShaderCode.create(gl, GL2.GL_FRAGMENT_SHADER, 1, getClass(), new String[]{"e:/shaders/"+fragmentShader},false);
fragShader.compile(gl);
ShaderProgram newShader = new ShaderProgram();
newShader.init(gl);
newShader.add(vertShader);
newShader.add(fragShader);
newShader.link(gl, System.out);
vertShader.destroy(gl);
fragShader.destroy(gl);
return newShader;
}
public static void setShaderUniform1i(GL2 inGL,int inProgramID,String inName,int inValue) {
int tUniformLocation = inGL.glGetUniformLocation(inProgramID,inName);
if (tUniformLocation != -1) {
inGL.glUniform1i(tUniformLocation, inValue);
} else {
System.out.println("UNIFORM COULD NOT BE FOUND! NAME="+inName);
}
}
public static void setShaderUniformFloat(GL2 inGL,int inProgramID,String inName,float inValue) {
int tUniformLocation = inGL.glGetUniformLocation(inProgramID,inName);
if (tUniformLocation != -1) {
inGL.glUniform1f(tUniformLocation, inValue);
} else {
System.out.println("UNIFORM COULD NOT BE FOUND! NAME="+inName);
}
}
public static void setShaderUniformFloat2(GL2 inGL,int inProgramID,String inName,float inValue1 ,float inValue2) {
int tUniformLocation = inGL.glGetUniformLocation(inProgramID,inName);
if (tUniformLocation != -1) {
inGL.glUniform2f(tUniformLocation, inValue1, inValue2);
} else {
System.out.println("UNIFORM COULD NOT BE FOUND! NAME="+inName);
}
}
private void viewOrtho(GL2 gl) // Set Up An Ortho View
{
gl.glMatrixMode(GL_PROJECTION); // Select Projection
gl.glPushMatrix(); // Push The Matrix
gl.glLoadIdentity(); // Reset The Matrix
gl.glOrtho(-1.0, 1.0, -1.0, 1.0, -1.0, 1.0);
gl.glMatrixMode(GL_MODELVIEW); // Select Modelview Matrix
gl.glPushMatrix(); // Push The Matrix
gl.glLoadIdentity(); // Reset The Matrix
}
private int genTexture(GL2 gl) {
final int[] tmp = new int[1];
gl.glGenTextures(1, tmp, 0);
return tmp[0];
}
/**
* Called back before the OpenGL context is destroyed. Release resource such as buffers.
*/
@Override
public void dispose(GLAutoDrawable drawable) { }
}
以及对应的GLSL Shader:
#version 120
uniform sampler2D textureData;
uniform sampler2D textureImage;
uniform vec2 resolution;
uniform float diffuseU;
uniform float diffuseV;
uniform float feed;
uniform float kill;
uniform float Tech = 1.0;
uniform float currentFrame = 0.0;
void main() {
//coords
vec2 position = ( gl_FragCoord.xy / resolution.xy );
vec2 pixel = 1./resolution;
//get data from texture
vec4 imgSource = texture2D(textureImage, gl_TexCoord[0].st);
vec2 oldUV = texture2D(textureData, gl_TexCoord[0].st).rg;
if(currentFrame<10){
if (distance(position,vec2(0.5,0.5 - currentFrame * 0.01f)) < 0.2f)
oldUV = vec2(0.0,0.2);
else if (distance(position,vec2(0.5,0.5 - currentFrame * 0.01f)) < 0.3f)
oldUV = vec2(0.5,0.1);
else
oldUV = vec2(0.1,0.0);
gl_FragColor = vec4(oldUV.rg, 0.0, 1.0);
return;
}
//get neightboors
vec2 dataUp = texture2D(textureData, position + pixel * vec2(0., 1.)).rg;
vec2 dataDown = texture2D(textureData, position + pixel * vec2(0., -1.)).rg;
vec2 dataLeft = texture2D(textureData, position + pixel * vec2(-1., 0.)).rg;
vec2 dataRight = texture2D(textureData, position + pixel * vec2(1., 0.)).rg;
//adapt parameters
vec2 imgParam = imgSource.rg;
float dU = diffuseU ;//+ 0.01 * (imgParam - 0.5);
float dV = diffuseV ;//+ 0.01 * (imgParam - 0.5);
float F = feed ;//+ 0.01 * (imgParam - 0.5);
float K = kill ;//+ 0.01 * (imgParam - 0.5);
//compute new values
vec2 laplace = (dataUp+dataDown+dataLeft+dataRight) - 4.0 * oldUV;
float uvv = oldUV.r * oldUV.g * oldUV.g;
// calculate delta quantities
float du = dU * laplace.r - uvv + F*(1.0 - oldUV.r);
float dv = dV * laplace.g + uvv - (F+K)*oldUV.g;
vec2 newUV = oldUV + vec2(du, dv)* Tech;
gl_FragColor = vec4(newUV.rg, 0.0, 1.0);
}
最佳答案
一些注意事项:
避免使用已弃用的 OpenGL(和 GLU),使用 GL4(或 GL3)
除非你需要awt/swt/swing,更喜欢newt,更多here
更喜欢 Animator
而不是 FPSAnimator
,更多 here
更喜欢直接缓冲区而不是数组,因为否则 jogl 必须在下面创建它们,并且您无法跟踪(=完成后释放)这些 native 分配
借助显式位置,GL4 还可以让您避免必须处理的所有统一开销(以及潜在的错误)(尤其是在运行时)
目前更喜欢直接缓冲区管理而不是 FBObject,除非您真的知道 FBObject 在做什么。一旦你让它工作起来,你就可以继续使用该类。这可能是(其中之一)原因是您的代码无法正常工作,因为某些内容没有按照您的需要进行设置。而且,替换FBObject所需的代码行本质上是相同的
(如果您因任何原因无法使用显式位置)更喜欢某种字面方式来定义纹理统一位置,这通常是另一个导致错误的棘手部分,例如 this
还喜欢纹理采样器,为您提供更大的灵 active
下次不要等一个星期,早点告诉我们! :) 沮丧可能是一件令人讨厌的事情,很容易让你沮丧。我们可以一起帮助您让它发挥作用;)
关于java - GP/GPU : ping pong technique with JOGL,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/36686895/
谁能解释或指出在多 GPU/多显示器设置中渲染如何工作的解释(或至少一些线索)? 例如,我安装了 5 个 NVIDIA Quadro 4000 视频卡并连接了 9 个显示器。显示不进行任何分组。刚刚在
以下代码报错: import spacy spacy.require_gpu() Traceback (most recent call last): File "/home/user/Pycha
正如问题已经暗示的那样,我是深度学习的新手。我知道模型的学习过程在没有 GPU 的情况下会很慢。如果我愿意等待,如果我只使用CPU可以吗? 最佳答案 在计算深度学习(以及一般的神经网络)中执行的许多操
我知道 Renderscript 的设计是为了掩盖我正在运行的处理器的事实,但是有没有办法编写代码,以便在支持 GPU 计算的设备(目前是 Nexus 10)上运行显卡?有什么方法可以判断脚本的功能正
关闭。这个问题是opinion-based 。目前不接受答案。 想要改进这个问题吗?更新问题,以便 editing this post 可以用事实和引文来回答它。 . 已关闭 8 年前。 Improv
我想以编程方式找出可用的 GPU 及其当前内存使用情况,并根据内存可用性使用其中一个 GPU。我想在 PyTorch 中执行此操作。 我在这个 post 中看到了以下解决方案: import torc
我喜欢 GPU Gems 的结构化技术摘要。但是自上次发布以来已经过去了很长时间,应该开发新算法来处理新型硬件。 我可以阅读有关最近 GPU 技术成就的哪些信息? 潜伏在编程板上是唯一的方法吗? 最佳
我一直在做一些关于测量数据传输延迟的实验 CPU->GPU 和 GPU->CPU。我发现对于特定消息大小,CPU->GPU 数据传输速率几乎是 GPU->CPU 传输速率的两倍。谁能解释我为什么会这样
当我使用选项 --gres=gpu:1 向具有两个 GPU 的节点提交 SLURM 作业时,如何获取为该作业分配的 GPU ID?是否有用于此目的的环境变量?我使用的 GPU 都是 nvidia GP
我用 gpu、cuda 7.0 和 cudnn 6.5 安装了 tensorflow。当我导入 tensorflow 时,它运行良好。 我正在尝试在 Tensorflow 上运行一个简单的矩阵乘法,但
我们正在寻找有关 slurm salloc gpu 分配的一些建议。目前,给定: % salloc -n 4 -c 2 -gres=gpu:1 % srun env | grep CUDA CUD
我是否必须自定义为非 GPU Tensorflow 库编写的代码以适应tensorflow-gpu 库? 我有一个 GPU,想运行仅为非 GPU tensorflow 库编写的 Python 代码。我
我是否必须自定义为非 GPU Tensorflow 库编写的代码以适应tensorflow-gpu 库? 我有一个 GPU,想运行仅为非 GPU tensorflow 库编写的 Python 代码。我
我正在使用 pytorch 框架训练网络。我的电脑里有 K40 GPU。上周,我在同一台计算机上添加了 1080。 在我的第一个实验中,我在两个 GPU 上观察到相同的结果。然后,我在两个 GPU 上
有没有办法在 Slurm 上超额订阅 GPU,即运行共享一个 GPU 的多个作业/作业步骤?我们只找到了超额订阅 CPU 和内存的方法,但没有找到 GPU。 我们希望在同一 GPU 上并行运行多个作业
我可以访问 4 个 GPU(不是 root 用户)。其中一个 GPU(2 号)表现怪异,它们的一些内存被阻塞但功耗和温度非常低(好像没有任何东西在上面运行)。请参阅下图中 nvidia-smi 的详细
我正在尝试通过 Tensorflow 运行示例 seq2seq,但它不会使用 GPU。以下是我在带有 Tesla K20x 的 Linux 系统上安装 Tensorflow 所采取的步骤 git cl
一位电气工程师最近提醒我不要使用 GPU 进行科学计算(例如,在精度非常重要的地方),因为没有像 CPU 那样的硬件保护措施。这是真的吗?如果是的话,典型硬件中的问题有多普遍/严重? 最佳答案 实际上
关闭。这个问题不满足Stack Overflow guidelines .它目前不接受答案。 想改善这个问题吗?更新问题,使其成为 on-topic对于堆栈溢出。 7年前关闭。 Improve thi
最近我研究了强化学习,有一个问题困扰着我,我找不到答案:如何使用 GPU 有效地完成训练?据我所知,需要与环境持续交互,这对我来说似乎是一个巨大的瓶颈,因为这项任务通常是非数学的/不可并行化的。然而,
我是一名优秀的程序员,十分优秀!