Today's graphics offers "on-CPU" graphics or separate card. This is different from back when the mobo had a graphics chip which I think we can all agree was substandard. So how well do these on-CPU graphics perform? Seems to me if my graphics processor is an Intel Core I-7 (http://www.newegg.com/Product/Product.aspx?Item=N82E16819116501) that's quite a bit of horsepower. And wouldn't it be using the ram from the mobo? Mobo ram is effectively unliminted today. Reason I'm asking is I need a file server. I'm going to donate my workstation to be the file server and build myself a new box. The new box is not for trading. It's my e-mail, internet, etc. box which will also be used for broadcast video production hence the graphics question. We do time lapse HD recording over long periods and when we stitch those frames (10k of them) it will lock up a regular pc. Seems to me that if I buy that Core I-7 I'll have a kick ass CPU and a kick ass GPU. Make sense?
No matter what the marketing geeks say, on-CPU GPU processing performance will lag dedicated GPU processing. If you're seeking performance, your best bet is to get a Nvidia or ATI card. Beyond that, you need to ask yourself the following questions... Which app and machine will you be doing the HD recording? Which app and machine will you use to stitch the frames together? Will both run on the new machine? At the same time? Are either/both apps able to offload their processing onto the GPU? If so, do they use CUDA or OpenCL protocols? If CUDA, you'll want an Nvidia card. If OpenCL, you'll want a AMD/ATI. Do you have a Gigabit switch and gigabit cables to connect the 2 machines (or, at the least, a gigbit crossover cable)? And do they both run at gigabit speeds (a lot of the cheap stuff won't) Does your fileserver have SATA-3 ports and SATA-3 drives? Is the data important, and you need redundancy? If so, you need Raid 5. If performance is more important, and the data is expendable, then you'll want RAID-0. If they're mix-n-match, then you'll need different RAID volumes with different schemes (RAID types). Each volume should have dedicated drives. So many questions, so little info given...
There is still limit how much RAM is used by on-chip GPU and if I am not mistaken data transfer would use same data bus. I was using i7 on chip GPU and performance was impressive. However, if you planning heavy HD editing getting good separate card would help because of data transfer bandwidth and more GPU cores and separate memory on GPU. Also look for newer i7 (not socket 1155) because this type of motherboard has limited number of PCI express lanes if it happens that you would decide later to use more than one GPU (more sharing of bandwidth)
The client does not need to talk to the server in order to produce the video. Seems I get more processors from a stand alone gpu and more ram from the on-cpu arrangement. Agree? So what's more efficient for video production? Having n x 100 processors or having tons of ram?
No. There is still limit on RAM usage by on-chip GPU. It is just shared with CPU with some addititional features that makes its addressing and data transfer more efficient for GPU unit. For heavy video editing separate card is more suitable because of more cores but also better shader units, dedicated memory etc.
Improved as the new CPU graphics are, they are still no match for performance GPUs. With a high performance GPU, it's my understanding the GPU takes over the majority of the workload with CPU supplement.. ?? If so and if video graphics performance is important to your use, best to go with an "adequate" CPU and a strong GPU. I'm still a bit amazed about the efforts CPU makers are going to... to eliminate the need for a discrete graphics card and save the user $20 or so in not having to buy a separate video card. Fine for basic use of web surfing, email, etc... but not for higher performance requirements. The entire concept of "video graphics chip on the mobo", and "video graphics built into the CPU" are all about cheap, cheap, cheap... not quality, capability, or performance.
The GPU has more processors and uses its own fast RAM. The GPU operations are relatively simple operations, and the RAM is to hold the data for pre & post computing. More system RAM is better to enable system, apps, and data to run from memory instead of swapping back and forth to disk. Overall, the more system RAM, the better. For your purposes, splurge on the RAM - get 16-64 GB if you can afford it. Then configure Windows to not use a pagefile unless your applications absolutely need it. Test system and application performance with and without a pagefile first. If all is fine w/o, then you're set. If not, then turn it back on. From what I've heard - offloading the processing to the GPU would speed up the job(s) by almost a magnitude or better (10x) while leaving your system completely usable. Take that with a grain of salt. Check the specific app(s) first to make sure you can off-load the processing onto the GPU. And you may want to Google to see which GPU gets the best performance for the price for that particular app (i.e. the app might be optimized for a certain GPU/chipset). Running the job(s) only on your CPU could leave your system slower/unresponsive for other apps while the video program is running. No matter what CPU or how much RAM you have installed. Be sure to build your system around the intended use (and apps), and match the individual components to that use, not the other way around.
This is not true today. This is new direction in computing technology of providing relatively high end graphics with impressive performance on the same chip as CPU. The reason is that in the future GPU will become co-processor++ where CPU will deal with basic input output and compute intensive tasks will be offloaded to GPU. It is perfectly fine to use on chip GPU because it is not the old good on board low end graphics. Separate GPU would be recommended for power users for HD video, CAD/CAM, graphic design and General purpose GPU computing. The rest of us can safely use built in GPU.