I could say welcome back, but I'd feel silly. The difference in credits between us can be expressed in, like, AU's...
Welcoming anyone is NEVER silly.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I used to come to this forum to mainly chat when seti was down. I'm just starting to crunch some einstein. I'll never been a big contributor as I just have one old machine, but I still like being a part of "it". Thanks for the welcome and the cake.
Some of you may know me from the seti@home. I did some optimizing of the code there.
Now I'm trying to do the same (at least something similar) here.
One of the first things I noticed in the code was that the memory access pattern at one point of the code was not efficient. There was a table of constants declared __constant. Nothing bad at that. But the constant buffer was read from a different address by each thread. That hurt NVIDIA GPU performance a lot.
A quick hack was to use a binary editor to change
"__constant" to "__global "
Thanks to Bernd M. for implementing that to the official code (1.28 NVIDIA Power Up).
Since then I have learnt more of the code and done some minor improvements now and then. There are some users testing my recent version(s) and the optimization (hobby) continues.
Some of you may know me from the seti@home. I did some optimizing of the code there.
Now I'm trying to do the same (at least something similar) here.
One of the first things I noticed in the code was that the memory access pattern at one point of the code was not efficient. There was a table of constants declared __constant. Nothing bad at that. But the constant buffer was read from a different address by each thread. That hurt NVIDIA GPU performance a lot.
A quick hack was to use a binary editor to change
"__constant" to "__global "
Thanks to Bernd M. for implementing that to the official code (1.28 NVIDIA Power Up).
Since then I have learnt more of the code and done some minor improvements now and then. There are some users testing my recent version(s) and the optimization (hobby) continues.
--
petri33
Hi Petri !!
You keep on doing what you're doing, we love it! I'm just sayin' for the all of us... Thank you !!
And here I brought chocolate
)
And here I brought chocolate cake to appease the leopard....But she's nowhere to be found :(
Guess I'll leave it in the refrigerator just in case....
I could say welcome back, but
)
I could say welcome back, but I'd feel silly. The difference in credits between us can be expressed in, like, AU's...
E pluribus unum
Jinkei wrote: I could say
)
Welcoming anyone is NEVER silly.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Haha, noted!
)
Haha, noted!
E pluribus unum
*sneaks to the fridge,
)
*sneaks to the fridge, looking to steal some chocolate cake*
E pluribus unum
haha. I'll sneak a slice of
)
haha. I'll sneak a slice of chocolate cake too.
I used to come to this forum to mainly chat when seti was down. I'm just starting to crunch some einstein. I'll never been a big contributor as I just have one old machine, but I still like being a part of "it". Thanks for the welcome and the cake.
Is this the smaller radio
)
Is this the smaller radio telescope or backlog?
#RIPArecibo
E pluribus unum
Seeing that the "record date"
)
Seeing that the "record date" on the BRP tasks are from 2016 they probably are older / backlog tasks.
Hi! Some of you may know
)
Hi!
Some of you may know me from the seti@home. I did some optimizing of the code there.
Now I'm trying to do the same (at least something similar) here.
One of the first things I noticed in the code was that the memory access pattern at one point of the code was not efficient. There was a table of constants declared __constant. Nothing bad at that. But the constant buffer was read from a different address by each thread. That hurt NVIDIA GPU performance a lot.
A quick hack was to use a binary editor to change
Thanks to Bernd M. for implementing that to the official code (1.28 NVIDIA Power Up).
Since then I have learnt more of the code and done some minor improvements now and then. There are some users testing my recent version(s) and the optimization (hobby) continues.
--
petri33
petri33 wrote: Hi! Some of
)
Hi Petri !!
You keep on doing what you're doing, we love it! I'm just sayin' for the all of us... Thank you !!
Proud member of the Old Farts Association