Analyzing Your Desired Output(s) To Determine Your Required Input(s) In A New Age
Welcome LBattis.
The title that you've chosen, "The Balancing Act," for your post is most appropriate. In everything that I do, I've found that the perfect balance starts with my imagining my end point; then I work backwards all the way to the present. I always put my desired outcome at the forefront; that is to first imagine the feeling that I want to have at the conclusion (and record it); then I ask myself what in particular generated that feeling (and record it); then what generated/produced that or those things that generated that feeling (and record it), and so forth (and record it) - working my way all the way back from that goal - the feeling of success/job well done - to the present. Then I have a map showing exactly how I ought to travel to get to exactly where I want to end up. Some might say that I approach things ass-backward; but I'd respond that although I do approach things backward from my final vision, it depends on what they mean by "backward." In nature, the hind quarters are generally on the back. Moreover, it's easier for me to accomplish my goal if it's been well defined and if I know in the minutest of detail exactly how to get there. So with that in mind, lets start with what you want to accomplish.
All this is to support a 2K video production environment from camera thru upload of mastered video to delivery system, and to publish a web magazine up to its servers; all this in support of the family ‘cottage industry’.
If, e.g., you desired to write code/ formulae etc. where memory errors were critical, then a Xeon processor supporting ECC memory would be very important. However, since it looks like that is not your purpose, then an i7 would be completely satisfactory. I have systems with i7 Gigabyte (G)UD3, (G)UD5, and (G)UP4 motherboards (using, e.g., i7 975s, i7 980X and i7 3930Ks CPUs). I use them for graphics and video production. The lack of ECC memory hasn't been an issue at all. I do code. So for that purpose, I use systems with Xeon Nehalems/Westmeres and Sandy/Ivy Bridge CPUs, like, e.g., the X5680 or E5-4650 QBED (which is just the E5-2680 with sufficient QPI pathways for 4 CPU system communication).
Tutor, I’ve read your threads pretty thoroughly. I’ve profited from that experience. I’m getting ready to purchase hardware and have one conundrum I thought I’d share.
You provided the following at the beginning of your thread on,
“All We Know About Maximizing CPU Related Performance”:
1) Single CPU system -
(a) Gigabyte motherboards recommended for a good combination of price/performance/flexibility. Consider an X79 Sandy Bridge (socket 2011) motherboard and a Xeon 1650 [3.2 -> 3.8 GHz] - has error correction support - $583 for best price/performance if you can use 6 cores . . .
That recommendation was for a general purpose build in that it covers more bases, but there's no need for a Xeon if your use pattern is precisely defined and doesn't trigger the need for ECC.
With that in mind I am considering the relative merits of E5 vs i7 in 6-cores. My thinking is to use the: GIGABYTE GA-X79-UP4 LGA 2011 Intel X79 SATA 6Gb/s USB 3.0 ATX, as foundation with one of the following:
i7-3930K Sandy Bridge-E 6-Core 3.2GHz (3.8GHz Turbo) LGA 2011 130W Desktop Processor BX80619i73930K
E5-1650 v2 (12M Cache, 3.50 GHz) FC-LGA12A
Is a good choice. I've found that the GIGABYTE GA-X79-UP4 LGA 2011 Intel X79 SATA 6Gb/s USB 3.0 ATX is an excellent foundation for a self-build. I now have eight of them. The vast majority of mine have the E5-4650 (QBED) C1 processors [see, e.g.,
http://www.ebay.com/itm/ES-Intel-Xe...CPU-/111412526373?pt=CPUs&hash=item19f0b43d25 Yes, their selling price has started to creep up] because at the time that I purchased components for those systems, I could purchase those 8-cores for
< $500 each and they have the exact performance characteristics of the E5-2680 V1, including having a max turbo of 3.5 GHz (w/o any clock tweaking). On the UP4 motherboard, every one of those CPUs can be over clocked by, at least, another 5%. The i7 3930K has a lot more over clocking headroom and still holds its own, even against the Ivy Bridge equivalents.
The motherboard that you've chosen is, of course, my favorite. It's the best value for a 4 double wide PCIe Sandy/Ivy Bridge single CPU system (Gigabyte makes a dual CPU system - GA-7PESH3 [
http://b2b.gigabyte.com/products/product-page.aspx?pid=4468#ov ] if your applications justify use of a dual CPU system; and, of course, if your applications would benefit more from a 4 CPU system, then consider Supermicro [
http://www.superbiiz.com/detail.php?name=SY-847R7FP ] . JUST MAKE SURE TO STAY AWAY FROM THE VERSION 1.0 GIGABYTE UP4 MOTHERBOARD (GET V1.1 INSTEAD) AND TO FIRST UPGRADE TO THE LATEST BIOS. The initial bios releases (pre-4.0) had many issues. I've found the later bioses to be friendly/stable.
My suspicion is that the time to future-proof the new build is upon me. On the downside, as I understand it I will be losing ECC Memory, while the plus is 4-channel DDR3. Am I right to be suspicious? Or is the exercise premature? In the final analysis should I watch for the right Mobo to use with an 8-core model E5-2690 v3 for the best balance? (Haven’t seen any pricing yet on the ‘low-end’ of this product line and its motherboard requirements.)
Absolute future-proofing is, of course, impossible. But wisely spending your money for what you need when you need it is nothing but applaudable. While the choice of CPU will determine whether you have ECC support (Xeon vs. i7), the UP4 handles them both. 4-channel DDR3 is a very important consideration. Depending on whether and how you chose to over clock a CPU, not getting memory ay least one or more steps higher than the system's standard memory speed may become a limitation. As I've pointed out earlier in this thread, a Sandy or Ivy Bridge Xeon can be over clocked by only a little (about up to ~7.55 % at most), but you'd still need to get memory at least one step higher to do so safely because overlooking the CPU also overclocks the memory by the same percentage. With those Xeons, you cannot change the turbo boost range, only the clock speeds (within low limits) at those ranges. Increasing the turbo boost range/steps (which you can do with an i7 CPU) doesn't necessitate faster memory. But in any event, I'd recommend that you get memory as fast, at a minimum, as the motherboard's max - here currently 2133 MHz [
http://www.gigabyte.com/products/product-page.aspx?pid=4766#sp ]. A couple of the benefits of a self-build is that you can over clock the CPU and memory and start with faster memory than that specified by Intel for the CPU (e.g., Intel spec's 1600 MHz memory for Sandy and 1866 memory for Ivy and with your GUP4 selfbuild you can begin with 2133 MHz memory). From what I've read, some of the V3 Xeons may have up to four more cores each and, at the high end (MHz and dollariwise), support more over clocking. But if you wait for the V4s or a while longer for the V5s, then I'm sure you'll get much better performance. My point is that while there is no way to future-proof absolutely, we should make system additions when we need them and do so most economically. When the family ‘cottage industry’ needs additional resources, don't be afraid to arm yourself with knowledge and make the wisest purchasing decision that you can make under the circumstances. Don't let the unachievable future-proof make you like a deer blinded by a tractor trailer's headlights just before impact. This is particularly so when the acquisition is needed (and will be used to produce additional income) and in what may be becoming the Age Of GPU Computing.
What particular video applications will you be using and will you rely on GPUs for added processing power (and will it require OpenCL and/or CUDA)? For GPU processing support, I currently use ATI GPU based systems (e.g., 3xPowerColor 5970s (circa 2009) in GUP3s [and am looking at getting either quad R9 280Xs or quad 7970s for another Gigabyte UP4 system] ) for Final Cut Pro and other OpenCL computing software, but since there are currently so few of my applications than use OpenCL, the vast majority of my systems use Nvidia CUDA GPUs. I use GPU computing to keep my 2007 MacPro2,1s still relevant and highly useful and yet they're seven years old. Aggregation too has its place, particularly when GPUs are relatively so cheap and yet so powerful. For me the key is having ample PCIe slots. I hate saying what I'll never do because things change. So I say, "I don't currently ever see myself getting/building a new system with less than 4 PCie slots." And I know that just by saying that, my negative karma will trigger the soon release of a faster and more powerful interconnect than PCIe or Nvidia to produce and release a 100 GPU $200 Titan CUDA card with an oTitan equivalency of 200 and that occupies just 1 PCIe slot. But I can say confidently, "I'll never be a stag with truck tire marks on my ass."
P.S. Or, might I become a similarly situated doe?
