I don't understand this. When your program does the stitching of many frames from time-lapsed video recording, it is a CPU-intensive operation. The output would be some video files on your computer. This is basically a batch process, done without you looking at any of the video. How is a fast GPU on your box going to help you?
Hi Boli: Thanks for asking a very relavent question. We use a time lapse camera in still mode to take 1 frame per minute at 1920x1080 resolution. Then we take those frames and make a video out of them. There are 1,440 frames per day x 10 days = 14,400 frames x 2mb per frame = 28.8gb of files we're working with. We also do video editing but that's just video clips being sliced and joined with audio track overlays. This process is only slightly hardware demanding compared to the above. That being said, let's get back to the original question: Am I better off with an on-gpu 3.0gHz cpu and 64gb ram than I would be with a dedicated gpu that had 2gb of ram? I can get much more ram at the gpu than I can in a dedicated gpu.
Hi mgookin: To understand your process further, my key question to ask is: is this a batch process - you don't see any display on the screen. The computer just take the input files, does that the program tells it, and produces some output files? Or is it an interactive process - the computer displays a frame, you look at it, adjust with some parameters, hit a button, then it goes on displaying another frame, and you look at it, etc...? If that is a batch process, I don't think any GPU will help you get it done faster.
Typically here's how it goes: Take a 32gb SD card out of a camera and copy those files onto the hard drive. See the files on the hard drive, erase the SD card, remove SD card. Open the video program (either Windows movie maker or Adobe Premier Pro) Import the files into the video program. This takes time. Tell the video program to make it into a video. Depending on what's going on (in the pictures) we may tell it 1fps or we may tell it 30fps.
Based on your descriptions, I don't think a GPU (whether on-chip GPU or on a separate video card) will help your process. What I envisioned is that this is basically a batch process. Yes you input the parameter of 1 fps or 2 fps or whatever based on what you see in the still pictures. After that, the computer loads each still picture (JPG format?) and stitch every one of them and create the motion picture video format (MPEG-something). It is CPU intensive to go through the algorithm to produce the MPEG output. But meanwhile, nothing will be displayed on the monitor (which is the only time that the GPU will come in). I think the best for you to optimize your investment is to get the fastest CPU you can afford, load it with as much RAM as you can afford, and equip yourself with as large a SSD as you can afford.
Interestingly Intel has a configurator which is linked from an Adobe site. You can enter your hardware specs, software specs, and tell it what your typical graphics work is. It told me that if I build a new system with Xenon cpu and whatever graphics card, my rendering time goes from 92 seconds to 88 seconds. Certainly not worth it. Seems the cpu does the work and the gpu is only for what goes on the screen. At least that's what I take from it.
The whole idea behind integrated graphics is that they can be spread across multiple products, no need to deal with multiple sets of drivers for multiple architectures, no royalty payments to third parties. Intel can also make it pretty power efficient delivering a complete solution to a mobile device maker
So how then do you do multiple monitors? The mobos made for these cpu's with integrated gpu only have one dvi output.
Mine has 2 outputs (1 VGA 1 DVI), I have 2 monitors there . I plug a 3rd monitor using a Display link USB to VGA adapter