NVIDIA CUDA for strategy testing - page 2

To add comments, please log in or register
Janis Smits
731
Janis Smits  

Dude, I really didn't want to upset you. You just remained me in 2009.

Then I used to have E6300@3.5ghz, 8GB ram and then I just bought new and shiny GF8800 Ultra. I got same ideas (and excitement) that you have now, then after 6months one great programmer contacted me and explained me alot about gpu computing. He actually did write some kind of CUDA code to accelerate his EA optimization (He had GF9600GT). Basically what he sad to me is this:" GPU computing is so extremely parallel that It's just impossible to make GPU calculate whole thing (like cpu do it), only some input parts can be calculated by GPU.

Just wanted to say that then I would be extremely happy if cloud existed. Back then you couldn't do alot if your means were to build expert that could work great as long term investment. Now you can!



BTW: What card do you have? I know it's not Ares 2 because you are talking about CUDA. :D

Strategy tester advanced computing. (Using multi core cpu or GPU) - MQL4 forum
  • www.mql5.com
Strategy tester advanced computing. (Using multi core cpu or GPU) - MQL4 forum
MetaQuotes
Moderator
216360
MetaQuotes  
OpenCL and GPU computing in MetaTrader 5: https://www.mql5.com/en/search#!keyword=opencl
Bruce Lesko
484
Bruce Lesko  
CeleronS:

Dude, I really didn't want to upset you. You just remained me in 2009.

Then I used to have E6300@3.5ghz, 8GB ram and then I just bought new and shiny GF8800 Ultra. I got same ideas (and excitement) that you have now, then after 6months one great programmer contacted me and explained me alot about gpu computing. He actually did write some kind of CUDA code to accelerate his EA optimization (He had GF9600GT). Basically what he sad to me is this:" GPU computing is so extremely parallel that It's just impossible to make GPU calculate whole thing (like cpu do it), only some input parts can be calculated by GPU.

Just wanted to say that then I would be extremely happy if cloud existed. Back then you couldn't do alot if your means were to build expert that could work great as long term investment. Now you can!



BTW: What card do you have? I know it's not Ares 2 because you are talking about CUDA. :D



Well I was being very subjective when I said it was expensive.  It was when I bought it at least.  GTX 580 3GB water cooled.  Not setup in SLI but has 512 CUDA cores.  I am currently looking at the GTX Titan.  You can probably already guess I am a gamer first and trader second as I use the largest single mainstream GPUs so I do not have to use SLI.  
GeForce GTX TITAN | GeForce
GeForce GTX TITAN | GeForce
  • www.geforce.com
With the DNA of the world’s fastest supercomputer and the soul of NVIDIA® Kepler™ architecture, GeForce® GTX TITAN GPU is a revolution in PC gaming performance. GeForce GTX TITAN...
Janis Smits
731
Janis Smits  
Be careful! Titan is ripoff, they charge alot for something that in games it give only 30% performance increase on over GTX680 on average.

As for MQL5 and Open CL you will be better of with Radeon. If I compare with bitcoin mining hardware(also huge open cl guys on gpu computing).



GTX680 gives ~120 Mhash/s

HD7970 gives ~680 Mhash/s


So it's about  5.6x times better for AMD gpu.



Btw: your gtx580 have faster calculation than gtx680. 140 Mhash/s

Mining hardware comparison - Bitcoin
  • en.bitcoin.it
Below are some statistics about the mining performance of various hardware used in a mining rig. Mhash/s = millions hashes per second (raw speed performance; may not be very energy efficient with some models) Mhash/J = millions hashes per joule (energy efficiency; 1 joule of energy is 1 watt during 1 second: 1 J = 1 W*s) W = watt (maximum...
Heinz Traub
308
Heinz Traub  

Btw: your gtx580 have faster calculation than gtx680. 140 Mhash/s

Yeah, Fermi based gpus are great hardware for gpgpu computing. Kepler has more cores but Fermi has a locked frequency of 2X core clock. This is good for software that needs fewer cores.
12
To add comments, please log in or register