Last Updated on Wednesday, 19 December 2012 21:45

This is the first part of a number of articles about a new system that needs to be built. It starts with the theory and in subsequent articles, well in a sub-menu lower on the left hand side of the page, it will show the actual progress on the build. This series may go on for months, depending on market developments, technological developments and budget restraints.

A sincere thank you to Bill Gehrke, Eric Bowen, Gary Bettan, Mitch Wood, Randall Leong and Todd Kopriva for helping me with this article and your very constructive thoughts and remarks. You have all been a big help.

From time to time we all face the question, should we build a new system now or in the next couple of months? What should we aim at, in terms of improvements? Do we need to save some more $$ and postpone our new system for a couple of months to profit from that decision for years to come? Where can we find info about option A or B to help us decide what is best for us? Is it really worth it in terms of 'Bang-for-the-buck' BFTB? Have we taken all aspects of a new system into consideration, did we not forget anything? Should we wait for those new announced products?

The ultimate purpose is to tell you how I went about building a new system, what considerations I had for certain choices, where my doubts were and how I tackled them. The limitation of the series is that it only applies to my situation and not by definition to your situation, so you have to distill anything that applies to your situation and tweak as necessary for your needs.

I mainly use my system for editing video using the Adobe Master Collection CS5.5/CS6 and well as website development with Dreamweaver for the PPBM5 Benchmark, it's successor and some other sites. The main codecs I use vary from simple ones like DV and HDV up to RED and EPIC 4K and 5K material, but the majority is HDV, AVCHD, XDCAM-EX 4:2:2 and Canon MXF 4:2:2.

My current system is getting rather dated, using only an i7-920 at 3.7 GHz with 24 GB RAM, a GTX 480 video card and a rather extensive disk I/O system. I mean it still runs without problems, is sufficiently fast for my current needs, so no urgent need to get a new system, but it is better to plan ahead than getting caught in a bind when a major component fails and you really need to get a new system the next day.

External influences

Interesting developments to say the least, but often without clear delivery dates to accompany announcements, or if they are announced with a specific date, they are often in very short supply and at premium prices. So, while interesting, keep this in mind when planning for a new system.

Having read a bit on new developments and seeing that in practice my aging system, while still a decent performer, was being overtaken by more and more modern systems, I started to look at possible components for a new system. But, I'm lucky. I can take my time to figure out exactly what I want to have, can check my choice of components, learn from the experiences of others, avoid common mistakes, and I can sleep on a difficult decision and postpone a bit if necessary. Whatever I will ultimately end up with, it has to be a big step forward from where I am now, otherwise it is a very unwise investment.

While the primary emphasis is on Premiere Pro, there are a lot of similarities for After Effects users. After Effects has very similar (but slightly divergent) hardware requirements. Considerations that are specific to After Effects will be noted in the topics below, such as memory and video card.

When Sandy Bridge was first released, it was not a feasible upgrade from an i7-920 (OC) @ 3.7. Sure, you can OverClock the Sandy Bridge further, but the 1155 platform with its limited PCIe lanes was a downgrade for me, forcing the video card to run in PCIe-8x mode, causing a 10-15% performance penalty. At best it could match my system, but not surpass it. What about Ivy Bridge then? No, still the same limitations of PCIe lanes. So, either i7-39xx or even a dual Xeon SB-EP. Hang on, that i5-2690 Xeon goes for over $ 2 K apiece and then the rest of the components are equally expensive, so let's forget about that. Initial choice of CPU: i7-39xx with the intention to overclock to 4.6 - 4.8 GHz

However, the current 3930K and 3960X have two cores disabled, as well as part of the L3 cache. I really hope that Intel will announce a 3980X with all 8 cores enabled as well as all 20 MB L3 cache. The i5-2690 has all cores enabled and the full complement of L3, so why not for the i7-39 range? To answer my own hopes as being something not expected in the short term, is that Intel is very strict about the 130W TDP limit, so do not expect that 8-core to appear until the Ivy version of these chips are available. Still time to wait and see.

I am considering either an Asus or Gigabyte X79 motherboard with 8 DIMM sockets, a FW port for older HDV/DV capture and for the rest up-to-date. I'll figure that one out later, but it looks like the choice is limited to the Asus Sabertooth X79, the Asus P9X79 WS or the Gigabyte GA-X79-UD5. I like that the Asus P9X79 WS has two 1 Gbps NIC's and more SATA-600 and USB3 connections. 2 NIC's is advantageous when using a network and/or a NAS. Initial preference of mobo: Asus P9X79 WS

Special consideration must be given to the layout and positioning of the PCIe-3.0x slots in light of the videocard chosen (see later under video card) because of the three slot width required for high-end video cards.

At least 8 x 4 GB sticks, but possibly 8 x 8 GB sticks, depending on price. The speed is secondary for the moment, as long as it is 1600+ and the RAM is low voltage (1.35V). Much will depend on price and availability. Maybe I can use the six 1600 4 GB sticks I have and only buy two more, although less than optimal. If anything, my current 1.5V sticks may cause problems when overclocking. I'll have to figure that one out.

Why consider even 64 GB when 32 seems more than enough? The thought is that the extra price of the additional RAM is small, but the potential benefits are great. One can use somewhere between 32 and 24 GB for RAM cache to improve performance. Is it worth it? Like so often it depends on the codecs you use. If you use heavily compressed codecs like DSLR, AVCHD or RED then yes, it is worthwhile, if you use simple codecs like HDV or DV, no, it is not worthwhile.

There is another consideration to go for 64 GB memory and that is if you use After Effects quite regularly. After Effects has very similar hardware requirements, but it will happily gobble the additional RAM when going from 32 GB to 64 GB, especially with the new Global Performance Cache.

We all know that hardware MPE makes all the difference and it can only be used with certain nVidia cards. Rendering, scaling on export, blending and blurring can lead to impressive performance gains over software MPE. The number of CUDA cores is decisive in that aspect. I will wait for further news about the Kepler range, but from the leaked specs, the 680 sure looks nice and the 690 even better, albeit at a price. Is it worth it, dunno. I have to decide that later. Anyway it makes no sense in getting a 5xx card. In that case it is much cheaper to port the 480 I have to the new system. There is no need for a second Kepler card to steer a third monitor, because that capability is one of the strong points of the new Kepler range. Note that official support of Kepler video cards may be quite some time in the future, but I base my choice on the use of the 'hack', as I've done with my current 480.

Furthermore, the fact that After Effects can use multiple GPUs for CUDA computation (for the ray-traced 3D renderer) makes using some GPU setups sensible that may have been a waste for Premiere Pro.

Initial idea for the video card: Gainward GTX 680 Phantom 4 GB or a GTX 685 with 4 GB VRAM.

I want to go for a better disk I/O setup than I currently have, but just look at current day prices of HDD's. That will cost a fortune if I want to extend on my current day 16 x 1 TB disks or even replace them with faster SATA-600 disks with larger cache. One thing to note is that I suspect one of these disks in my raid30 to be starting to give trouble, causing time out errors. It has not died yet, but it may in the near future. One thing is for sure, while I can port my Areca ARC-1680iX-12 controller to a new system for the time being, I ultimately want to get the new PCIe-3.0 controller, possibly called the Areca ARC-2082iX-24 with the 24 + 4 ports. Of course with at least 4 GB of cache memory and a battery backup module (BBM).

Why a 24 port raid controller you may ask. First of all, the price difference between a 12 port and a 24 port model is very small and you do not want to be in a situation where the number of ports are a limitation. Second, if HDD prices were not so high, and they may come down in the next couple of months, then the following setup may be worth considering, provided raid support for the trim function of SSD's materializes:

Why would anybody in his right mind want so many disks? That makes no sense at all. Actually it does. Why? Because disks are still the main bottleneck in each system, the CPU, GPU and memory are way faster. If you only edit easy codecs like DV or HDV the Need for Speed is of course less than when editing RED 4K or EPIC 5K material. If you edit multiple tracks, your disk speed requirements go up, if you use multicam, your speed requirements go up, if you use 4:2:2 or even 4:4:4 material, your speed requirements go up. If the nature of your video is fast moving, lots of short clips, your speed requirements go up. If you use AE compositions exported as uncompressed, your requirements go up because of the sheer size. Redundancy costs extra disks but buys safety.

Initial preference of raid controller: Areca ARC 2082iX-24 (PCIe-3.0), expected end of Q3/2012?

Given the high prices of HDD's I have to postpone the choice and number of disks and the definite raid configurations for a later date. The controller is not out yet, so patience is the word, but realistically it may boil down to 2 x (7 R3 + 1 HS) plus 4 x R0, for a total of 20 HDD's. Well, 16 new 2 TB HDD's are still less costly than a single i5-2590 Xeon CPU, and the other disks can be ported from my existing system. That leaves me room to grow when the need arises.

Given the ambitions with the disk setup, my current Lian-Li PC-A77 with 17 disks and 2 BR burners internally will be too small. I need something that will house around 28 HDD's, 2 BR burners, 4 SSD's and a multicard reader for ingest from CF and SD cards.

With the huge number of disks in such a system, I want to have hot-swappable bays, so it is easy to exchange failed disks. Normal big towers will not do because they are simply not big enough, so this means further investigation. But it also raises another question: cooling. Air, water or even more extreme, nitrogen? I'll come back to that later.

I found a case that easily meets my requirements, albeit at a price. And it can be tailored to my specific needs. And most importantly, it will fit under my desk, even on casters and casters are a necessity for such a large case. It beats working in the cramped space under the desk for maintenance, upgrades or disk replacement. Enlarging the case with a pedestal or going for a larger model would not fit, so they were out.

When I tell you the details, your initial reaction will be: 'He's crazy to even consider such a case'. Well, that may be the case in this case, but OTOH a case like this is like a tripod system. It can easliy outlive several generations of systems, giving an expected life span of more than 10 years. You never need to get another case. The same with a good tripod, it can easily run in $ 3K to $ 5K+ figures, but it is a 'life-time' investment. Almost certain case choice: CaseLabs MAGNUM TH10

I want to have the capability to use up to 4 SSD's, so this Addonics_SSD_bay seems to fit the bill, taking only a single 5.25" bay. For the hot-swappable HDD cages I will opt for 3 Chenbro_Hot_Swappable_bay units, since the TH10 does not fit any more.

Why the Chenbro drive cage and not the Supermicro CSE-M35TQ? They are similar, 5 hot-swappable disks in a 3 bay housing and have the same functionality. The things that steered me towards the Chenbro is that it includes two USB connections on the front of the cage, the drive cages on the PSU side are compatible with Chenbro backplanes, so I have only one kind of backplanes in this system. Last, there are rumors that the Supermicro included fan is pretty loud.

With the case and drive cages almost a certainty, the question remains the cooling, since the case does not include any cooling, but has huge amounts of space for mounting a number of radiators, although the drive cages will limit my options.

Initial thoughts are that a system with 28 HDD's will make so much noise anyway (each internal drive cage has a 120 mm fan and the hot-swappable Chenbro's have a 80 mm fan at the back) with 7 fans for the drives, the options for Kepler water cooling are currently non-existent, so the benefit of water cooling must, at least for the time being, come from the CPU cooling. It may result in a very slight advantage over air cooling, almost negligent, but will require a tedious installation, is much more expensive and requires more maintenance, so it does not look appealing to me. My initial feeling is KISS, just a Noctua NH-D14 in push-pull configuration. But that means also that for all those possible fans I have to budget a large number of Scythe Slip Stream SY or Noctua NF fans. A separate article is planned on thermal design, where we will examine the pro's and con's of positive and negative pressure and the consequences. Initial preference CPU cooler: Noctua NH-D14 CPU cooler

Initial preference case fans: Scythe Slip Stream SY1225SL 12L (very quiet) or Noctua NF-P12 (because of a high CF/db ratio).

I have checked this possible system, making several assumptions and ended up with a required PSU of around 1500W with all those disks, overclocking, video card, fans etc. This is based on the advertised 195W use of the 680 Kepler card, so this is the best I can find at the moment on power requirements.

Keep in mind that 1500W may seem huge and be a serious burden for your electricity bill, but the PSU will only use what it needs. A 1500W PSU will not use more energy than a 500W PSU in a small system if the power is not needed. It will run cooler and more stable however. It is like a Volkswagen Beetle (500W) trying to keep up behind the safety car in a Formula 1 race, it has to run at full power to even keep up, while the Formula 1 car (1500W) has to worry that his tires will not get too cold.

My inital thoughts are to install two PSU's in this case, one a Corsair or Seasonic 80 Plus Gold PSU of around 500W to power the video card and the BR burners, plus some of the case fans, and another PSU of around 1000W to power the system and hard disks. Whichever way you turn it, staggered spin-up of the disks is a necessity.

Initial thoughts: Corsair or Seasonic 500W plus 1000W PSU's

On second thought, serious doubt coming up now about the two PSU's. Some PSU's refuse to start if there is no motherboard attached. This may be a serious issue and limit my choices. In the past I have noticed this issue with a CoolerMaster PSU, that refused to start if it was not attached to a mobo. This may mean I have to short two connections to circumvent this. I will investigate further and keep you appraised.

Conclusion of Part 1

The major components have been identified, some preferences indicated, but it is still very preliminary. Next step: Order the case with all required case components and start working on part 2.

A rather novice at video editing may think, after reading all this, it is way over my head and the writer is only focussed on disks and the case. "I have been editing with a two disk configuration for years without major problems, so this is utterly exorbitant and from another planet."

That is (partly) correct, but you have to consider where I am coming from and where I want to go. I have a decent system with a PPBM5 score of 157 seconds. The best i7-39xx system currently holds a score of 133 seconds. That difference in performance does not justify the extra cost of a new CPU, motherboard and extra RAM for such a relatively small performance increase, especially if one compares that 133 score at an OverClock of 4.8 GHz versus my 157 score at an OverClock of 3.7 GHz.

In my case, the upgrade to a 2011 platform and an i7-3930K will make a difference, but not enough to justify the cost, unless other components are upgraded at the same time.