[Help!] Recommend me a RIG for Deep Learning / Machine Learnin

[pewpewpew]

Hi,


I am a beginner for ML and i realize whilst running my Tensorflow on ASUS STRIX-GTX960-DC2OC-2GD5 PCI-E 2GB GRAPHICS CARD, the program kept crashing

Tensorflows requires high bandwidth GPU and based on numerous resources ive encountered it seems to me the optimal bang-for-buck GFX is GTX 1080Ti, 1080 or 1070

im quite sold on this https://www.amazon.com/EVGA-GeForce...=UTF8&qid=1503299851&sr=1-2&keywords=gtx+1070 for this purpose

I know nothing about PCs generally, but if i were to use this 1070 GTX, what are the other hardware you would recommend?

my budget is low and most of it is blown on this GFX, so appreciate cheaper/best bang for buck hardware as much as possible

much appreciated

 

[Psycovirus]

I quickily read up on Machine Learning Subreddit's Wiki and they mentioned

Generally speaking, you should utilize whatever GPU has the most memory, highest clock speed, and most CUDA cores available to you. This was the NVIDIA Titan X from the previous generation.

1070 is a great budget solution for your work flow. However, due to cryptocurrency mining boom, 1070s has been in short supply thus most of the 1070s are overpriced.

The EVGA card will cost you about SGD$675 to import to Singapore. The cheapest GTX 1070 i saw is GTX 1070 Galax EXOC, $619 at PCThemes.. Must check if they still have stock.

If you are to considered GTX 1080, EVGA GTX 1080 FTW is going to cost $810 to import. There is also a GTX 1080 G1 Gaming from Gigabyte which can be purchased in SLS for $815.

If you think 40% more Tflops cacluation power is worth extra $135, go and get a GTX 1080! I don't know how much difference the performance will be in Tensorflow.

 

[pewpewpew]

thanks for the kind reply!

may i ask, could yo kindly suggest the rest?

mobo, processor, etc.. im really extremely new to this!

has anyone tried ordering most of the parts from amazon to singapore?

there is no local warranty (though im aware amazon entertain returns) is that a problem?

any of these parts (esp PSU) would have issues with our local electrical supply (240 compared to 110 in the states)

 

[Psycovirus]

May I know what is your budget?
Do you need high core count/speed CPU?
How much RAM required for Machine Learning?

Lastly, need Keyboard/Mouse/Monitor too?

 

[pewpewpew]

May I know what is your budget?
Do you need high core count/speed CPU?
How much RAM required for Machine Learning?

Lastly, need Keyboard/Mouse/Monitor too?

Thank you.

I'm hoping for about 2-2.5k

Looking at i7, RAM maybe 16 is enough

Need keyboard and monitor

No need mouse

Again, thanks!

 

[Psycovirus]

Thank you.

I'm hoping for about 2-2.5k

Looking at i7, RAM maybe 16 is enough

Need keyboard and monitor

No need mouse

Again, thanks!

This is a difficult task for me. I do not know much about machine learning or Tensorflow.

I am unsure if AMD's Ryzen CPU will work better for this system since they are better value for money.

For GPU too, AMD just released RX Vega 64 which has as much or even faster FP32 compared to 1080 Ti, < 12 Tflops. It even has 25 Tflops of FP16 compute, which GTX 1080 Ti do not have. However, AMD has promised a working driver for Machine Learning applications like Tensorflow in near future. While GTX 1080 Ti is working with Tensorflow.

Also, for the monitor, just a simple monitor, 24" FHD will do? because that will be much cheaper than gaming monitors...

So, I am not sure what kind of build to recommend for Machine Learning. Hope an expert in ML can help you with it.

 

[ondooy]

Titan X or Tesla.
don't waste your money on consumer GPU.

 

[Psycovirus]

Reddit Thread amd_has_superior_compute_performance_what_exactly

It's discussing about hardware support. nVidia GTX 1080 Ti is recommended.

Sent from OnePlus ONEPLUS A3000 using GAGT

 

[pauley mccoy]

Hi, sorry to off topic abit but where can I start learning about machine learning?

 

[CKtalon]

Hi, sorry to off topic abit but where can I start learning about machine learning?

Try Coursera

https://www.coursera.org/specializations/deep-learning

To OP, get a 1080 Ti 11GB

 

[gld998]

Get the 1070 if your funds are limited(still faster than a regular GTX Titan X (Maxwell)).....but looking at your budget you can spring for a 1080ti if your planning on doing larger networks say a LSTM with 1024 hidden units (and batch size > 64). Small LSTM (128 hidden units; batch size > 64) on a GTX 1080 Ti will not be that much faster than running it on a GTX 1070.

For CPU just about any multi threaded single core cpu can cope with a single GPU setup...clock speeds don't matter much as long as it is more than 2ghz which most cpus are nowadays anyway. Saying that though multi core cpus does help in other tasks like preprocessing or researching/browsing via google etc.
If your planning to parallelize neural networks or more for research on multiple GPUs in the future then try to get get a motherboard with more PCIE slots.

For ram or memory you at least need as much as what or GPU has but as always getting more would be better.

Lastly if you have the space buy 3 cheap monitors maybe 24 inches as it helps tremendously in your workflow and setup e.g. documents/research/browser on the left monitor....coding in the middle and to do list/system monitor/small apps on the right.

Hopefully that will help out someone to do a build lists for you as I'm out of touch with pricing....good luck.

Why u all recommend gtx1070? Now its so inflated, might as topup abit and get gtx1080 non ti

Sent from Samsung SM-T719Y using GAGT