You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
fyords:
...
And why is your test so bad if the execution time is 171 ms - 360 times less than on CPU? By the way, such an excess on 96 cores indicates that the graphics card was obviously released much later than the stone...Yeah, I'm happy with my test.
Yes, my graphics card was taken much later, it was joo on the wave of interest to GPU calculations, almost forced me to upgrade, I'm really not looking for power, put the first that supports OpenCL (native did not support) paid extra 50$, it was enough to GeForce GT 430,
That's the advantage of add-in cards over integrated ones (you can swap them out).
...
Here's a look. (And try it):
The result (time) is even better than average. More often less than 2 seconds.
... The main thing is to grow hands from the right place.;)
My report looks different somehow (this is from tab "Experts").
And this is from Log tab. Otherwise the time of execution is nowhere to be seen.
Regarding the hands: More details here, please. )) Judging by my report, something went wrong.
It seems to me, in the attachment is not the script.
I think it's the wrong script in the attachment.
Me too.
Here's an even better one.
// Finally got the inline function working! All I had to do was move it to the back of the kernel... :)
// and deleted extra assignment there, too).
Also renamed it to make it shorter.
There, now that's different, it's already impressive.
At the same time, the bugs on the forum are coming to light.
And you, MetaDriver, could give webinars on neuronics.
That's why the bugs are popping up on the forum.
Yeah, I'm happy with my test.
Yes, the graphics card was taken much later, it was joo on the wave of interest to GPU calculations, practically forced me to upgrade, I really did not strive for power put the first that supports OpenCL (native did not support) paid extra 50$ which was enough for GeForce GT 430,
that's the advantage of a plug-in card over an integrated card (you can swap them out).
I feel very uncomfortable/uncomfortable.
sorry.
I bought some hardware, they said I'd get it back in a week and I'll show you the results...
Actually, I'm really sorry. I'm really sorry. I'm sorry.
I feel very uncomfortable/uncomfortable.
Sorry.
I bought some hardware. They said I'd have it back in a week. I'll show you the results then.
I'm really sorry. I'm really sorry. I'm sorry.
Nope. No, you're not.
People like you don't belong on this forum. Too late to repent! Only blood will wash away your er... I don't know what you're guilty of.
...
Here's a look. (And try it):
The result (time) is even better than average. More often less than 2 seconds.
So the slow cards are also in business. The main thing is to grow hands from the right place. ;)
)) Let's grow them slowly. It's the result on my weak one:
This is at CountBars 100000. It can't take more than that, it's going down.
And my "horseless car":
Anyway, I'm sick of testing other people's work. I'm going to make my own, which is what I need. I'll post it here when I'm done.
)) We'll be growing slowly. This is the result on my weak one:
It's at CountBars 100,000. It can't survive more than that and crashes.
Try to reduce the number of parallel threads (CountPass). For example, try to halve or quadruple them at once to begin with. 1280 is a VERY large number for an algorithm. joo told me that the optimal population size is 50-60 individuals. 1280 is way too large. I put it there purely for stress testing.
You might have a speed bump precisely because of excessive thread demand (in card terms).
--
Go for 512 and see what you get. Don't be afraid to mince your program, it'll make it better. :) When you've done it, post it here.