CUDA based programming GPU selection

[T.Vengeance]

What's the recommended GPU for running CUDA based programming? I'm planning to use CUDA within my code but a GTX 660TI isn't gonna cut it for running a neural network. Should I go the newest NVIDIA gpu on the block? 1080TI? Or should I look into their quadro series. I have no idea what the quadro is used other than in workstation environments.

[ARikozuM]

You should be looking at Nvidia's purpose-built deep learning line-up. 

 

What program are you going to use? 

[T.Vengeance]

5 minutes ago, ARikozuM said:

You should be looking at Nvidia's purpose-built deep learning line-up. 

 

What program are you going to use? 

I'm developing a neural network from the ground up in C# implementing the NEAT algorithm. Going to be using the CUDAfy.NET libraries unless you would recommend a different library for CUDA programming in C#.

[2FA]

Honestly the Tesla cards would be best but if you can't afford them, Geforce or Quadro.

[ARikozuM]

2 minutes ago, T.Vengeance said:

I'm developing a neural network from the ground up in C# implementing the NEAT algorithm. Going to be using the CUDAfy.NET libraries unless you would recommend a different library for CUDA programming in C#.

...

 

1 minute ago, DeadEyePsycho said:

Honestly the Tesla cards would be best but if you can't afford them, Geforce or Quadro.

What this guy said. 

[T.Vengeance]

Just now, DeadEyePsycho said:

Honestly the Tesla cards would be best but if you can't afford them, Geforce or Quadro.

Between Geforce or Quadro, why does NVIDIA differentiate between the two? Is it just like ECC vs non-ECC ram, being that Quadros are workstation-typed so they are more reliable?

[TheGhzGuy]

4 minutes ago, T.Vengeance said:

I'm developing a neural network from the ground up in C# implementing the NEAT algorithm. Going to be using the CUDAfy.NET libraries unless you would recommend a different library for CUDA programming in C#.

I have very little idea of what you just said but it sounds really cool.

[T.Vengeance]

2 minutes ago, TheGhzGuy said:

I have very little idea of what you just said but it sounds really cool.

Haha, it's cropped up recently, a neural network with node/connection designs that are implemented through evolution by natural selection principles, but nothing revolutionary. Many have done it and, yeah it is really cool so I'm hopping on the hype train.

[matrix07012]

3 minutes ago, T.Vengeance said:

Between Geforce or Quadro, why does NVIDIA differentiate between the two? Is it just like ECC vs non-ECC ram, being that Quadros are workstation-typed so they are more reliable?

Here:

https://stackoverflow.com/questions/10532978/difference-between-nvidia-quadro-and-geforce-cards

[ARikozuM]

2 minutes ago, T.Vengeance said:

Between Geforce or Quadro, why does NVIDIA differentiate between the two? Is it just like ECC vs non-ECC ram, being that Quadros are workstation-typed so they are more reliable?

Quadros' drivers are generally worked on for much longer to ensure stability with any program that requires it. 

Geforce is usually done on a game by game basis. 

[2FA]

3 minutes ago, T.Vengeance said:

Between Geforce or Quadro, why does NVIDIA differentiate between the two? Is it just like ECC vs non-ECC ram, being that Quadros are workstation-typed so they are more reliable?

Quadro is all about reliability, GeForce is all about raw power.

[TheGhzGuy]

1 minute ago, T.Vengeance said:

Haha, it's cropped up recently, a neural network with node/connection designs that are implemented through evolution by natural selection principles, but nothing revolutionary. Many have done it and, yeah it is really cool so I'm hopping on the hype train.

That makes much more sense. What are you going to use it for specifically? 

[T.Vengeance]

5 minutes ago, TheGhzGuy said:

That makes much more sense. What are you going to use it for specifically? 

End goal....have it stock trade for me (rolls eyes yes I know everyone's tried and failed). But for the meantime, since its a neural network, been meaning to test out my code on learning how to talk through analyzing the FFT data of speech with a stream of incoming IPA symbols, essentially a text to speech that doesn't sound like crappy google's TTS or SIRI.

[matrix07012]

I don't how much memory your neural network will need. Quadro P6000 features 24GB of VRAM.

http://www.velocitymicro.com/blog/geforce-vs-quadro-whats-the-difference/

[T.Vengeance]

13 minutes ago, matrix07012 said:

Here:

https://stackoverflow.com/questions/10532978/difference-between-nvidia-quadro-and-geforce-cards

Thanks man, read the post. Since I'll need double floating point calculations, I guess quadro is the way to go. Is there a roundabout minimum price that I should be spending? Like with geforce cards, you wouldn't get a GTX 720, it's just dumb for the money you pay for, you would expect to pay around at least 250 to 300 bucks for a Geforce card that makes it worth over just going integrated graphics or AMD's APU line. Is there something equivalent for Quadro?

[TheGhzGuy]

13 minutes ago, T.Vengeance said:

End goal....have it stock trade for me (rolls eyes yes I know everyone's tried and failed). But for the meantime, since its a neural network, been meaning to test out my code on learning how to talk through analyzing the FFT data of speech with a stream of incoming IPA symbols, essentially a text to speech that doesn't sound like crappy google's TTS or SIRI.

Still pretty cool. Never been into trading anything, so good luck!