Thanks, Bill, for your testing. This means that the GTX 680 is still not worth the premium at its current street price. It needs to be priced about $100 lower just to compete with the GTX 560 Ti 448 in terms of overall bang for the buck. And those results only confirm that any CUDA GPU with less than 192 CUDA cores on a PC with a highly overclocked CPU or an expensive high-performance CPU might as well be forced to use CPU-only: The GPU results would have been slower than the CPU-only results with such cheapo GPUs. ... View more

Yes, Bill, the GTX 680 may be a gamers' and Premiere Pro users' dream. However, it is relatively weak in some other applications that take full advantage of the compute capabilities even when compared to some of the older and lower-end GeForce GPUs. This is a GPU that for now I'd place in "great for gaming and Premiere, weak on everything else" category. ... View more

Actually, Alan, yes, there is a specific requirement for Extended ATX: The E-ATX case has eight total expansion slots. Standard ATX cases have only seven total expansion slots. E-ATX cases must accept E-ATX or any of the smaller variants of ATX. ... View more

Jim Simon wrote: The computer tech guy said a hack had the potential to crash my system He was not correct about that. I'd venture there are probably more hacked cards in use out there than officially supported cards.  I mean, there's only three non-Quardo, still in production cards on that list. Actually, only two. The GTX 470 is EOL (meaning that it's no longer in production, but may still be available at some resellers). ... View more

For now, that's your only option. At least on initial release, the Adobe list of supported cards in CS6 for Windows remains completely unchanged from CS5.5. And it beats me that Adobe has certified the Quadro 2000 - a GPU that's based on a much less powerful GPU than the GTX 560 Ti (in this case, the Quadro 2000 is based on an underclocked GTS 450) - in addition to certifying the GTX 570 and GTX 580. (And as I noted, the GTX 680 is not included because it is still too new to have been tested sufficiently by Adobe engineers.) ... View more

Search the forums for "CUDA" and "hack". The GTX 560 Ti is not on the official list of supported GPUs in Premiere Pro CS5.5; however, you can make MPE GPU acceleration work with that GPU with the CUDA hack. ... View more

While what you stated is true for the Seasonic, it is not true for the many "650W" PSUs that can barely handle 450W under realistic conditions. ... View more

eclipse_crow wrote: Swapping out a PSU is akin to getting an upgraded engine and not using a completely different car from a cheaper manufacturer of a smaller format! The OP has a Seasonic PSU and they are Mercedes class let’s say so I suggest a better analogy is this: Take a C class Mercedes and compare one with an entry level engine and one with an upgraded engine. Is the entry level engine less reliable? No. Is the entry level engine less efficient? For the same engine type it’s generally more efficient. Is the entry level engine noisier? For the same engine type generally not. On the other hand, if a given car ships with an underpowered engine as its entry level engine, then it will be significantly noisier and less efficient - and it may very well be ill-suited for highway driving because it cannot reach typical highway speeds reliably. This is the situation if an even lower-wattage PSU from the same manufacturer is used - one that turned out to be unable to handle a typical editing system configuration. (That analogy is similar to a car that's equipped with an engine that's mismatched to the weight of the vehicle.) And yes, a 1700W power supply that's outputting no more than 350W most of the time is like putting an 8.6L V12 engine in a Ford Mondeo/Fusion. ... View more

Ed, The M12II 620 is as I described in my last post: For all practical purposes, it is a 80-Plus Bronze certified PSU with a single +12V rail rated at 48A (576W). It should be enough for most Premiere Pro hardware configurations although it will consume slightly more AC wattage from the wall than PSUs with a Gold or Platinum rating at the same DC wattage output. As for SpeedGrade, there is no software-only mode. If it detects a non-CUDA GPU or a GPU that doesn't support it, SpeedGrade is simply disabled. ... View more

Ed.Macke wrote: I'm glad to hear somebody else likes Seasonic. When I was building my rig, Seasonic was not a brand that came up a lot in the various builder forums. But I did a lot of homework and it seemed like a good choice so I went for it. To be certain I suggest checking what the 12V rail is for your PSU. It has (2) +12V rails, each at 24A. Doing the math real quick seems to mean 576W DC. Actually, with most "multi-rail" PSUs, never add up the two rails together. However, in the case of both the S12-II and M12-II 620W PSUs, the two "24A" rails do add up. In this case, both of the "rails" are connected to a common +12V output transformer whose maximum capacity just happens to be 48A (576W). And each of the two "rails" have an OCP (Over-Current Protection) circuit set at 24A. However, someone else dissected that Seasonic PSU, and actually found it to have just a single 48A +12V rail. No separate rails, no separate OCP circuit. Thus, for all practical purposes, the Seasonic 620W PSU is a single-rail PSU with 48A on the +12V. ... View more

eclipse_crow wrote: Power supplies are rated for DC output whereas most reviews measure the AC input. There used to be a big difference between the two but with the advent of Gold and Platinum rated PSUs the difference can be 10% or less. Since you have a SeaSonic 620W PSU who are one of the top manufactures (retail and ODM) I suggest that 620W is more than adequate to run a GTX 570. To be certain I suggest checking what the 12V rail is for your PSU. You would be unlikely to pull more than 400W DC at peak and typically under 350W DC at load so a high quality 620W such as you have gives a lot of headroom. There are a lot of people that are seemingly ‘scared’ of power supplies and think you need a 100% headroom to be safe. A quality PSU such as you have is designed to run at its rated output safely. If you had a cheapo unit that would be different. You have a very large margin anyway so even the scared crowd would need to be particularly paranoid to be scared here. 350W load versus 620W rating equals 56% of the maximum rating for the PSU. To make a car analogy: If you buy a $30k car (analogous to a 620W PSU) and drive it to 56% of the 7,000 rpm red line (~3,900 rpm) is it any less reliable than a $60k car (analogous to a 1.2kW PSU) driven only to say 28% (half the rating) of the same red line (~1,950 rpm)? If both cars are reliable brands then it’s really not an issue surely. Bravo. The power supply calculators have to account for the fact that there are many mediocre-quality PSUs still sold today that cannot handle more than about 60 to 70 percent of their rated wattage (for example, a few cheap "700W" PSUs can't really handle more than 500W). In fact, a few PSU calculators take into account really bad-quality PSUs that could barely handle even half of their ratings (for example, a "650W" PSU that barely handles 300 to 350W under realistic conditions). ... View more

Thanks, Dave, for the update. ... View more

Thanks. It makes sense to not enable OpenCL MPE GPU acceleration support for the Windows edition of Premiere since OpenCL for Windows still hasn't matured enough. And with the current state of Windows' OpenCL drivers, the fastest AMD GPU in the OpenCL-enabled MPE GPU-acceleration mode would still have performed slower than an NVIDIA CUDA GPU that costs half as much money in the CUDA-specific MPE GPU acceleration mode. Although I have now managed to (semi-)permanently lock my particular CS5.5 installation to the MPE software-only mode regardless of the GPU by completely deleting the cuda_supported_cards.txt file. This occurred on my auxiliary editing rig. ... View more

Why?  Wil Rencez has already suggested it may well be possible for Mac users to 'hack' additional Radeons. Thanks for the clarification. Jim. Unfortunately, there will be no such OpenCL hack for the Windows version of Premiere Pro CS6 because such support will not be initially included with this edition. However, OpenCL support in the Windows edition of Premiere Pro CS6 may be added in a future update. And what Todd stated about OpenCL support in the OSX edition of Premiere reflects the fact that hacks to enable GPU acceleration support are (or will not be) officially supported by Adobe itself. This means that if any problems arise from such a hack, technical support may be denied until the hack is completely reversed. ... View more

John T Smith wrote: For what you will do, do NOT buy Quadro! As Harm would say... over priced and under specced Actually, this may change with CS6: It may make excellent use of certain capabilities of the Quadro that are seriously lacking in the GeForce. But we will not know for sure until the first few batches of results appear on the PPBM6 results list. If what I stated proves true, then the performance ranking may very well go strictly by price, where a Quadro 2000 performs roughly equal to a current-gen GeForce of equal price. ... View more

SvendBendt wrote: Adobe could easily make everybody happy by simply allowing OpenCL on PC (and not just a few MacBooks) That will not happen (at least on the initial release of CS6). It may happen with a future update. ... View more

True (at least on the initial release of CS6). But also starting with CS6, I am expecting that NVIDIA CUDA GPUs that are not officially supported for GPU acceleration to be semi-permanently locked into the software-only mode (at least until a future update is released): The .txt file may be eliminated altogether and the CUDA GPU support may be hard-coded into the program's executable itself. However, the list of officially supported NVIDIA CUDA cards remains unchanged in CS6 from CS5.5. This means that the new GTX 680 will not be officially supported on release of CS6 since that GPU was introduced after the development of CS6 was largely completed. And Adobe made a wise move in limiting initial OpenCL GPU acceleration support to those MacBook Pros running OSX Lion equipped with the HD 6750M and HD 6770M chips with 1GB or more VRAM: That particular move was made largely because Apple no longer sells any of its systems with NVIDIA GPUs. The HD 6750M and HD 6770M GPUs in iMacs will also not be supported because those systems with said GPUs have only 512MB of VRAM - too little VRAM to even enable GPU acceleration at all. ... View more

Here is the biggest reason why you need multiple drives in an editing system: 1) If you have only a single hard drive for absolutely everything including the OS, think of the SATA connection as a single-lane road with reversible traffic controls. And because of the single traffic lane, traffic can only move in one direction at a time. And a certain block of traffic in that direction must pass completely through before any traffic is allowed to move in the opposite direction. That's what's called "half-duplex". An analogy to this would be the reversible-direction express lanes in the middle of Chicago's Kennedy Expressway (I-90/94). 2) Even as separate drives, hard disks are limited in maximum sequential transfer bandwidth. The fastest 7200 RPM mechanical hard drives can hit just under 190 MB/s on the outer tracks, and drop off to around 80 MB/s on the inner tracks. That's one reason why a hard drive should not be filled to its maximum capacity. It's also why multiple drives in a RAID array can improve sequential performance, especially if you're working with RED 4K and uncompressed HD material. ... View more

Actually, that also depends on the PSU itself. There are numerous astronomically expensive, ultra-high-powered PSUs out there that simply waste electricity no matter what. At less than 20 percent of maximum capacity, a PSU is much less efficient than a PSU that's at 50 percent of maximum load. This is because a PSU's circuitry, like a car's engine, needs to be run at a certain minimum just to not waste power. And what good is a 1700W PSU if it's going to be outputting less than 350W most of the time that it's powered on? ... View more

One more addition to that procedure: The "clean install" procedure will corrupt any sound card or onboard sound drivers which may have been installed! This will make the NVIDIA HDMI out the only audio device installed. Therefore, you must reinstall any onboard audio or sound card drivers after this procedure. ... View more

Linda, There have been two different Z68-UD5 boards from Gigabyte: The original GA-Z68X-UD5-B3 (which lacks a video out and most of the goodies that come with the Z68 chipset, and therefore is little more than a P67 motherboard with Intel's SRT added on), and the newer GA-Z68XP-UD5 (which has an HDMI port for video-out from the CPU's integrated IGP, and therefore supports QuickSync and Virtu - and as such is a newer board that's more likely to fully support the new Ivy Bridge (e.g. i7-37##) CPUs (which are tentatively scheduled to come out at the end of next month) with their PCI-e 3.0 hub integrated). When the Z68XP came out, the original Z68X was discontinued. On the other hand, the GA-Z68X-UD7-B3 has never been updated since it was first introduced: As a premium board that's (over)loaded with many premium features, there is no room at all whatsoever to even include a video out for the IGP. As such, my comment on the UD7 is the same as the original GA-Z68X-UD5-B3. And both of the boards that do not have video out will restrict the PCI-e 3.0 operation of IB CPUs to only x8 mode because those boards have only a Gen2 x8/x8 switch. And Gen2 switches cannot do anything at all whatsoever with eight of the PCI-e 3.0 lanes because the switch cannot read them at all. Newer revisions of the Z68XP-UD5 board do have a Gen3 x8/x8 switch, so that you would be able to take full advantage of future PCI-e 3.0 graphics cards (however, the newest GeForce 600 series GPUs do not yet take anywhere near full advantage of even PCI-e 2.0, let alone PCI-e 3.0). As for the choice of the GPU, stick with the 560 Ti even if it has only 384 CUDA cores and a 256-bit VRAM bus: The GTX 470's technically superior specs (448 CUDA cores and a 320-bit VRAM bus) is more than offset by that GPU's tendency to run relatively hot even at idle. This, in turn, results in the GTX 470 frequently speeding up its fan to maximum speed in the middle of an encode, and the GPU itself might throttle back as the encoding progresses, creating a new bottleneck. In fact, in my own testing the GTX 470 with its reference cooler is only slightly faster than a reference non-Ti GTX 560 (with only 336 CUDA cores and a 256-bit VRAM bus). I am now running my system with a "GTX 560 Ti 448", which has 448 CUDA cores and a 320-bit VRAM bus but is a newer GPU with a higher clock speed (as it is actually based on a GTX 570 with one of its processing units disabled). The "GTX 560 Ti 448" should have really been named the "GTX 565" or "GTX 568" because its performance is much closer to a GTX 570 than to a regular GTX 560 Ti. Also, search the forums for "CUDA hack". Those posts and threads will give you an idea on how to enable MPE GPU acceleration on a CUDA card that's not officially supported by Adobe in that mode. Randall ... View more

Thanks, Bill, for the mention of the forthcoming GTX 6## series. I am suspecting that my particular GTX 470 that I have had for almost two years had been underperforming. As it turned out, I had purchased an early-revision reference GTX 470 card with a lousy reference cooler. That card had been running significantly hotter than it should have, which may have been the explanation for the MPEG-2 scores within PPBM5 that are just barely faster than a GTX 285. I confirmed this after I tested my old GT 240 DDR5 card, which as expected delivered a MPEG-2 DVD score of 249 seconds (on my main i7-2600K system overclocked to 4.3GHz with 16GB of RAM) versus 111 seconds with the GTX 470. Though that score is a bit faster than I had previously thought, it was still slow enough to drag my system's total time from 239 seconds (with the GTX 470 at the same CPU overclock setting) to 404 seconds. I will be testing two other GeForce cards with 1GB or more RAM (a GTX 560 non-Ti (336-core) that's currently in my i5-2400 auxiliary rig and a newly-purchased GTX 560 Ti 448 just to check that your 68-second MPEG-2 DVD score isn't a fluke). UPDATE: As it turned out in my own testing, it was my particular GTX 470 that was, indeed, underperforming. It was only 10 seconds faster than the plain, non-Ti GTX 560 in the MPEG-2 DVD test -- 111 seconds for the GTX 470 versus 121 seconds for the GTX 560. Furthermore, that GTX 560 Ti 448 is no fluke: I achieved nearly the same score as you did (any slight differences were due to the differences in the tuning of the systems). I will be putting up an updated score of my main i7-2600K system (with the GTX 560 Ti 448) on the PPBM5 site. If that system were counted with the 66 i7-2600, 2600K and 2700K systems running CS5.5 that are currently on the list, my system's new 190-second score (at a higher 4.6GHz) would rank fourth. By the way, I would have been perfectly satisfied with the standard 384-core GTX 560 Ti on that main system of mine. However, most of the 384-core GTX 560 Ti cards from eVGA cost almost as much money as the eVGA GTX 560 Ti 448 FTW that I have (or put it this way, the cheapest eVGA-branded 384-core GTX 560 Ti costs only $45 less than the $290 GTX 560 Ti 448). That's too close for comfort, IMHO. Had you included both of the "standard" GTX 560 series GPUs in your chart, the GTX 560 non-TI would have ranked about equal to the GTX 285 while the GTX 560 Ti (384-core) would have scored about mid-way vetween the GTX 560 Ti 448 and the GTX 285. The GT 440 with DDR5 RAM would have scored about 100 seconds slower overall than the GTX 550 Ti (this means that a cheapo nVidia GPU with only 96 CUDA cores and 128-bit DDR5 RAM would have made that overclocked i7-2600K as slow as or slower than a stock-speed i5 with a fast GPU). As it stands, the GTX 560 Ti 448 did so well because it uses the exact same GPU core as the GTX 570 and GTX 580. Message was edited by: RjL190365 ... View more

First off, SSDs still aren't durable enough for long-term use for repeated writes and rewrites. You see, SSDs still have a limited number of rewrite cycles. At the rate the scratch disk is rewriting, you would have used up the SSD's entire life cycle in as little as just a few days! And the factory warranties do not cover such abusive wear and tear. Thus, in such a circumstance, you might have to pay as much as tens of thousands of dollars every week just to even replace those SSDs. Second, NEVER RAID any SSD unless it's specifically designed for RAID! Raiding a consumer SSD will completely defeat the TRIM function and will not only degrade performance, but will also cause even more frequent failures. In the worst cases, a RAIDed SSD array will fail after only a few minutes of operation. These are the reasons why hard drives are still better suited for those situations that require multiple repeated rewrites in such a short period of time despite the higher hard disk prices. But now, hard drive prices are now slowly beginning to fall: A month ago, the 1TB WD Black cost all of $250 plus shipping at Newegg. Now, it costs $140 plus shipping - still about $50 higher than the amount that same drive cost just six months ago. ... View more

Claire, I'd recommend going with mostly new hard drives in that system if you're going to be doing anything at all whatsoever with DSLR video. You see, the Maxtor drives that are in your possession are now extremely old - so old that two of them in RAID 0 ("aid0") are still slower than just one of last year's 7200 RPM hard drives. Those old drives could not achieve even 70 MB/s in sequential transfer speed on the outer tracks (because those old Maxtors used a platter layout of only 100GB per platter, resulting in the three-platter design of your old 250GB and 300GB disks) while current hard drives can reach nearly 190 MB/s on the outer tracks (and even the fastest of the two-year-old drives can reach 140 MB/s). In your inventory the fastest drive in your system is currently the 1TB Samsung F1 (note the model number HD103UJ), which uses a three-platter design to achieve its 1TB capacity. That drive can reach about 120 MB/s on the outer tracks. The 2TB WD Green is actually slower than the 1TB Samsung F1 in sequential transfer speed despite the denser platters of the Green because that Green spins at only 5400 RPM (as I noted in my previous post in this thread). Thus, I am amending "No green drives" to also include "No hard drives more than three years old regardless of spindle speed". In other words, drives that are suitable in an editing system must not only spin at 7200 RPM or higher, but also have a platter layout density of at least 300GB per double-sided platter. Those old Maxtors fall far short of that 300 GB per platter minimum (they employ 100GB per platter at most). ... View more

Just when you think it's safe to recommend an i7-3930K, bad news hits the DRAM market: One of the DRAM IC manufacturers, Elpida, filed for bankruptcy. This is partly due to the continued weak overall demand for DRAM despite falling prices. As a result of the Elpida bankruptcy, expect RAM prices to increase by the end of this year - and very significantly throughout next year. What that does is that the 8GB DIMMs will increase slightly in price from the current $70 per DIMM (in the case of the Corsair XMS3 CMX16GX3M2A1600C11 kit I recently purchased) to roughly $90 per DIMM by the end of this year, but the prices for smaller-sized (capacity) 4GB DIMMs will skyrocket from the current $18 per DIMM to about $40 to $50 per DIMM. This will have a direct impact on the total cost of building your own PC. By the way, I did notice in my i7-2600K system that whenever I set Maximum Render Quality (MRQ) to On, my system's MPEG-2 DVD time in the PPBM5 benchmark test increased exponentially, from 115 seconds with MRQ Off to nearly 800 seconds with MRQ On. There is a bottleneck somewhere within the LGA 1155 platform (as Shawn discovered in his testing) - mainly not enough threads available for MRQ to do its work efficiently. Also, I expect that even an i7-3820 will suffer from the exact same problem as the i7-2600K in terms of not allowing anywhere close to 100% CPU utilization in software-only MPE mode with MRQ turned on sinply because MRQ in software-only mode is severely taxing to the CPU: MRQ in software-only MPE mode really needs at least 12 to 16 threads in order to perform acceptably fast. ... View more

Shawn, Thanks for that link. However, it redirected to a "Page Not Found" portion of the site until I removed the extra "/http://" at the end of the address. Then, I realized that the MPE software-only mode really needed a strong hexa-core CPU just for that mode to perform nearly as well (relatively speaking) as the MPE GPU accelerated mode. The 4-core, 8-thread i7s do not perform nearly as well in MPE software-only mode, especially with Maximum Render Quality (MRQ) enabled - and I would expect the cheapest LGA 2011 CPU, the i7-3820, to perform significantly slower than the i7-3930K in MPE software-only mode (again, due to the 3820's four-core, eight-thread configuration versus six cores and 12 threads in the 3930K). On the other hand, with GPU acceleration enabled in Premiere's MPE, the Sandy Bridge quad-core CPUs perform nearly as well as the i7-3930K (again, relatively speaking). ... View more

Harm, Regarding the "Budget" versus "Economical" system: Unless one is going to use part of the RAM as cache, Premiere Pro CS5.5 works best with about 2GB of RAM per logical core. This means that the i5 system doesn't take much if any advantage of more than 8GB of total installed RAM. I discovered this today when I ran the PPBM5 benchmark on "Randall's Flying Pig" (my i5-2400 auxiliary editing rig) with both 8GB and 16GB of RAM. However, for an i7-2600K or 2700K system, one would need 16GB of RAM in order for CS5.5 to perform its best. And for an i7-3930K, 32GB is recommended. ... View more

Nope. Remember, Harm stated "No GREEN disks at all"! This means that the Green drive should not even be inside your PC's case at all - but instead put into an external enclosure with eSATA and/or USB 3.0 capability. Here are three reasons why: 1) Their much slower than minimum spindle speed: The WD Green drives spin at a very slow 5400 RPM (as I noted in a previous post). 2) Most Green drives also have relatively slow random access speed - as much as 22ms instead of the 15ms of the average 7200 RPM hard drive. 3) Many Green drives (and all of the Western Digital Caviar Green drives) spin themselves down after a firmware-set period and cannot be overridden or disabled at all whatsoever! That means that the drives will simply ignore any software- or Windows-set spin-down times unless those times are less than the firmware-set time. In other words, if you have Windows to never power down any drives, the Green drives will power themselves down anyway. And every spin-down and spin-up can (and does) cause crashes in Premiere Pro, especially in the middle of a long encode. However, points 1) and 2) are not as critical as point 3). The constant spinning down and spinning up of the Green drives will destabilize an editing system that expects every single installed drive to perform at its maximum speed and performance at all times. This is the basis for my (and Harm's) assessment of Green drives in an editing system. ... View more

In this case, get an external USB 3.0 or eSATA enclosure and put the Green drive into that. Then get the third internal 7200 RPM hard drive. ... View more

In this case, get another 7200 RPM hard drive from another source: Green drives are often too slow in both sequential transfer speed and random access speed to be of much use for anything besides backups. In the case of the Caviar Green, its rotational speed is only 5400 RPM despite Western Digital's "variable-speed" claim. ... View more