Calendar Latest Topics
 
 
 


Reply
  Author   Comment  
Edward

Administrator
Registered:
Posts: 260
Reply with quote  #1 
Hi,

If anyone is looking to build themselves a high quality, but relatively low cost computer for running ML, you might want to take a look at the following options.

Case: http://www.newegg.com/Product/Product.aspx?Item=N82E16811352011

PSU: http://www.newegg.com/Product/Product.aspx?Item=N82E16817139048

Mainboard: http://www.newegg.com/Product/Product.aspx?Item=N82E16813157512

CPU:
Option1: http://www.newegg.com/Product/Product.aspx?Item=N82E16819117372
Option2: http://www.newegg.com/Product/Product.aspx?Item=N82E16819117369

RAM: http://www.newegg.com/Product/Product.aspx?Item=N82E16820148545

GFX: http://www.newegg.com/Product/Product.aspx?Item=N82E16814121916

SSD: http://www.newegg.com/Product/Product.aspx?Item=9SIA2W02DV8166
or
http://www.newegg.com/Product/Product.aspx?Item=N82E16820147467 (may require a better motherboard)
 
CPU Cooler: http://www.newegg.com/Product/Product.aspx?Item=N82E16835103099
0
stremblay

Member
Registered:
Posts: 65
Reply with quote  #2 
Hey Ryklin. Nice build!

I'm guessing that with this configuration, you would hook up your GFX to the subjects' monitor and the experimenter's monitor to the motherboard GPU ?

Thanks!
0
Edward

Administrator
Registered:
Posts: 260
Reply with quote  #3 
I actually had a recent conversation with a techy on this very subject. As it turns out, if you use the on-board graphics (ie intel graphics), even though it may be hardware accelerated (DirectX/OpenGL), it will nonetheless draw some CPU resources. Therefore, by getting a completely independent GPU, you will get better overall computing performance. I'm not saying don't do it, in fact I did not get a GPU either, but I'm planning to eventually so that I can experience the full power.
0
Edward

Administrator
Registered:
Posts: 260
Reply with quote  #4 
Just want to follow up on this question of where to connect the subject's monitor.

There are basically three types of GPU's, nVidia, ATI, and Intel. These dominate the market.  Typically, all mother boards have an Intel GPU built-in to the motherboard. Usually, there are two monitor connectors on a mother board as well. So, you might be wondering if you can make due with just the on-board graphics. Is there any need to invest into a separate expensive graphics card, when ostensibly the mother board's graphics alone should be sufficient.

The answer depends on the stimulus presentation software. If your software generates graphics using the CPU, then having the additional graphics card won't make any difference. However if your software generates graphics entirely on the GPU (like most modern video games), then you probably must have a separate graphics card because Intel GPU's may not entirely support hardware accelerated software. All stimulus presentation software that I know of, such as Psychopi, Psychtoolbox, and our very own MonkeyLogic use some combination of hardware and software graphics rendering. None are completely CPU independent.
 
0
Jaewon

Administrator
Registered:
Posts: 971
Reply with quote  #5 
While the specifications of graphic libraries can be updated anytime, hardware cannot be modified once sold. Because of that reason, applications are supposed to detect capabilities of the GPU on runtime and request hardware acceleration only for the part that the GPU can do. Therefore, the hardware acceleration is just an option that you can turn on and off and there is no graphic that can be generated without the CPU. Intel began to support DX 10.1 from the first generation of the Core processor, so it is unlikely that your graphic is generated with no hardware acceleration. Just Intel's GPU is not as fast as the other manufacturers'.

MonkeyLogic does not do any rendering or whatsoever, as long as it depends on XGL. XGL supports only bitmap blitting, which is a 20 year old technology. The advances that the latest GPUs made are mostly in 3D graphics and effects, none of which XGL uses. For MonkeyLogic, it is better to buy a faster CPU with more system memory or a cheapest GPU that has large on-board memory.
0
Jedema

Junior Member
Registered:
Posts: 2
Reply with quote  #6 

Hi All,
Thanks for the helpful information on this site. I am setting up 2 new rigs and I am looking for some advice regarding the components to purchase for use with ML. I have used ML a couple of times before, but primarily used E-prime and I have no experience setting up the solid configuration for ML. At this point I am planning to use stationary images as stimuli (potentially some pre-generated fractals) on a Windows 7 64 machine. 

I have read the recommendations for a single NI card (PCI-6229 or PCIe-6323 X-Series) with the NIMH DAQ toolbox. Is there a reason to spend a little more money on the cards with higher sampling rate capacity that could be a benefit for future use (i.e. PCIe-6343, PCIe-6353 or PCIe-6363) ? 

NI recommends (obviously) that you purchase a full copy of Labview but I have not seen much mention of Labview on this ML site. Do you need a full license for Labview (~$3K) or will Labview base ($1K) suffice or is it not needed at all if using MonkeyLogic with Matlab ?

Because "fast" continually changes with CPUs, is a current i5 or i7 with 8-16GB of RAM sufficient or is there a need to add more RAM?

What is the current recommendation for graphics cards ? It seems that it might not be necessary for me to get a graphics card for the simple stationary stimuli that I will use but what would be considered a cheap GPU with large memory at this point if I should get one ?

0
Jaewon

Administrator
Registered:
Posts: 971
Reply with quote  #7 
Buy an X-series NI. The older series are usually more expensive for the same performance. If your computer has PCIe slots, there is no need to buy a board with an old interface. Think carefully how many digital IO you will need, so that you can do all IO with one board. Since NI's accessories are expensive, you can save a lot, if one board can handle all your IO need. Buy the cheapest board among the ones that meet your need. You don't need "high" sampling rates.

Labview is not needed. Never.

Assuming that your subjects do not care for full HD or 4K, I would think any GPU that costs more than $50 is a waste of money. Intel Iris graphics is perfectly fine.
0
Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.