- Last Updated on Friday, 25 April 2014 10:57
The CPU, Mobo, Memory, Video and Raid Controller
It may be in vain, but while the build progresses and I decide on more and more components, I am hoping that Intel will come out with an Ivy Bridge-E CPU, based on the 22nm production process, that will have all 8 cores and the full complement of 20 MB L3 cache enabled and still remain within the Intel TDP limits.
Rumors are that Intel will introduce the i7-3970X in Q4, still based on the 32 nm (Sandy Bridge-E) technology, but with a base clock of 3.5 GHz and a turbo boost at 4.0 GHz for the same price as the current i7-3960X, whose price will be reduced at that time. It will still be a hexa core chip with 15 MB L3 cache, but will run 200 MHz faster than the i7-3960X. The TDP of the new chip will be 150W.
Leaked information about the Intel roadmap indicates that the Ivy Bridge-E series (i7-49xx) will not come out before Q3-2013 and will still have two cores and 5 MB L3 cache turned off. See Intel Desktop Roadmap. Disappointing news and it removes any reason to wait any longer, so the order for CPU, memory, video card and some SSD's will go out shortly. You can see the items ordered on the Price Tag page.
Areca has the PCIe-3.0 card available now, so that will be the next order, but first is the installation of Windows and tuning the OS, before installing the raid card.
Motherboard and CPU Cooler
Luckily motherboards do not change often, so it is a safe bet to get the Asus P9X79 WS, even if Ivy Bridge-E CPU's are announced at a later moment. But that also means if I want to install that motherboard, I will have to decide on the CPU cooler now, because I need to install the back-plate together with the mobo. I can install the CPU and the cooler at a later date. So I ordered the Noctua NH-D14 SE2011 together with the mobo.
The main reason for ordering these now is that prices of motherboards and CPU coolers are pretty stable, but not so for CPU, memory and video card. By installing the mobo now it gives me more information to decide on the way cable management should be handled.
I have recently read this statement from Samsung:
“Samsung will also aggressively move to establish the premium memory market for advanced applications including enterprise server systems and maintain the competitive edge for Samsung Green Memory products, while working on providing 20 nanometer (nm) class* based DDR4 DRAM in the future.” Also see Samsung DDR4 Memory Technology.
I don't know how far in the future, but with registered modules in 8 GB and 16 GB sizes being delivered to major CPU and controller makers, it makes sense to wait a bit and hope for the future not being too far off.
Latest info is that the waiting can go on indefinitely, so I've decided on the G.Skill 64 GB DDR3-2133 Octo-Kit, and Asus confirmed compatibility with the Asus P9X79 WS motherboard today, so I have ordered these sticks. Why? Here is some general advise on how to go about choosing the right memory:
- Make up your mind about the number of memory sticks you want to install. On the X79 platform, do you want to use 4 slots, or do you want to fully populate all 8 slots? This is very important especially with i7-3930K / 4930K CPU's, which have a rather finicky memory controller.
- Make a shortlist of quad or octo packs of memory sticks in the size you want, limiting yourself to ONLY sticks using 1.5V or less.
- Check the motherboard manufacturer's site QVL list to see which of the sticks on your shortlist are approved. Pay attention to the number of slots for which they are approved. Some modules are approved for 4 slots only, others for fully populated use.
- Check the memory manufacturer's site QVL list to see which motherboards are approved.
- Try to establish if there are users that have been successful in using the sticks on your shortlist and delete any that were giving trouble to others. NewEgg can be a good source for that.
- Now that you have a very short shortlist, contact the Technical Support departments of both the motherboard manufacturer and the memory manufacturer and ask them to confirm or deny the compatibility of the memory modules in the quantity you need.
- If both confirm compatibility, check the physical height of the modules including heatsink in relation to the intended CPU cooler. Do they fit without modding?
In general, only buy matched sets of memory sticks. There are too many different sticks with the same type number, that will cause grief when bought as four single sticks, but you limit the risk by only buying quad sets or octo sets for fully populated X79 platforms.
CAS latency is no longer important, in fact, with the finicky nature of the memory controller of the Sandy Bridge-E, it can be advantageous to choose a higher CAS latency for improved compatibility. What can be important - relatively speaking, only in single digit % performance differences - is the rated speed of the memory, 1333 or 1600 or 2133 or even 2400.
Even though the X79 platform only supports - at least officially - up to DDR3-1600, one can use XMP profiles in the BIOS to have the memory run at its rated speed of say 2133. For my new 'Monster' I followed these steps and ended up with a fully populated 8 x 8GB octo set of G.Skill RipjawsZ F3 DDR3-2133 sticks that are running without any problems at 2180 MHz.
The much touted Maximus solution, at least by Adobe, is an utter waste of money, because it requires a very expensive Quadro card plus a Tesla C2075 card, that is even slower than a simple two generations old GTX 470, for the simple reason that it lacks CUDA cores and memory bandwidth to make it faster. Mind you, this does not include the as yet unsuppoted K20 Tesla card, which may alter the perspective altogether, being based on the Kepler architecture.
We have several Maximus solutions in the current PPBM5 benchmark and despite prices up to € 6000 for the Quadro 6000 plus a Tesla C2075, these solutions are easily outperformed by many GTX 470/480/570/580/670 and 680 cards, that cost only a fraction. The only thing impressive about a Maximus solution is the price. The only reason to opt for a Maximus solution, despite the cost, is if you absolutely need 10 bit output to your very expensive 10 bit monitors, and then you only get quality, not performance.
What determines video card performance for PR?
Based on information currently available, it is not so much the number of shaders or CUDA cores. If that were the case, all Kepler cards would easily leave Fermi and older cards in the dust with about three times the number of CUDA cores, but that is not the case. The determining factor is memory bandwidth and that explains why the Kepler range is only slightly faster than Fermi and why Kepler is significantly faster than all Quadro cards.
Currently there are only 5 models of the nVidia GeForce GTX 680/4 GB available in the Netherlands. My choice is the EVGA Geforce GTX 680 Classified with 4 GB VRAM because of the nicely overclocked speed without costing much more than the non-overclocked models. There still are no rumors about a new version with the much desired 384 bit wide memory bus, what I used to call the GTX 685.
The Areca ARC-1882iX-24/28 Gen3 supports PCIe-3.0 and is now available, so it has been ordered. However, the standard configuration comes with a 1 GB memory stick for cache. In contrast to NewEgg, there are no 4 GB sticks installed on some models over here, so I had to order a separate 4 GB stick from Areca. I have not been able to find a more affordable brand of memory stick that will work on this controller. The other problem with the standard configuration is that the Dutch distributor only supplies the card with 6 SFF-8087 to 8087 cables and I need 6 SFF-8087 to SATA forward breakout cables (50-60 cm) so those had to be ordered separately. Additionally, I have not yet found a SFF-8088 forward breakout cable (1 m) for the last 4 ports, but that is a later worry. I can now start with 24 ports.