So eh... yeah... today this happened. Guess I'll see some of you there in May then. Who's going to GTC this year?
Also, general GPU programming thread I guess. Watcha working on?
Anyone have their hands on a Pascal Quadro and gotten to work with the new GPUDirect API?
nvidia but MAYBE amd xd
dank digits xd
>>59000003
Close but no cigar
>>59000000
Can't wait for GTC 2017, hope to see Volta for Tesla announcement just like last year's Tesla P100 announcement
Do I really need a GPU for quick stuff in tensorflow?
Just started learning Cuda for class. Is it really normal to fire up one kernel for every little computation? There are only a few thousand cores on my gpu and it just feels silly to make so many kernels instead of just looping.
Also anybody know where the compiled Cuda code is in the elf? Does it get its own section?
>>59002706
>Just started learning Cuda for class. Is it really normal to fire up one kernel for every little computation? There are only a few thousand cores on my gpu and it just feels silly to make so many kernels instead of just looping.
GPU threads are very weak compared to what the CPU can do.
>Also anybody know where the compiled Cuda code is in the elf? Does it get its own section?
nvcc even splits it into separate files
>>59001542
Yes
Alright OP, you got me interested. How do I start getting into this as well. OpenGL?
>>59003414
TensorFlow uses cuDNN.