If you need a new card, do yourself a favour and stay away from all AMD cards. The only brand to get is nVidia and the specs you must look for are:
- At least 1 GB VRAM
- At least GDDR5 memory
- At least 192+ shader units or CUDA cores. The more the better and even more important, look at the best memory bandwidth you can afford.
Looking at your budget range, comparable in cost to the HD 6670, look at the GeForce GTX 550 Ti. In PR it is around 11 times faster than any AMD card for rendering. The next step up to look at are the GTX 660 or GTX 660Ti
Look at the MPE Gain Chart
Thank you for taking the time to reply and for including your table.
If I read it correctly the NVIDIA cards have the Mercury Playback Engine and the AMD cards do not, is that correct? Following on from that logic, there is a gain in rendering from the Mercury Playback Engine re reduced rendering times that the AMD cards cannot achieve through lack of MPE?
In connection with LucidLogix, does this become a factor where it can pass on the task of rendering and editing to the internal processing available from an i7 sandybridge CPU? If LucidLogix can pass the rendering and editing task to the i7 sandybridge CPU instead of the GPU then surely the advantage of the MPE from the NVIDIA cards is lost?
I thought that with LucidLogix and sandybridge CPU's the intensive tasks of rendering/encoding/editing video were removed from the GPU and passed on to the CPU for more efficient processing. However, I am a total novice in this area so could be wrong on this?
Please forgive my ignorance but what is "PR", as a novice I am not familiar with this?
My understanding of the way LucidLogix Virtu operates is that for gaming applications it allocates the task over to the graphics card and for non gaming applications it passes the task to the CPU. LucidLogix Virtu is not purely for gaming, it allocates video tasks to the best engine to carry out the task, for games that is the GPU and for other tasks it defaults to the CPU if the application supports LucidLogix.
Games do not belong on an editing rig. LucidLogix Virtu does not belong on an editng rig. PR will not benefit from having it installed and it can only lead to problems. If your claims about Virtu are correct, installing it will reduce performance by a factor 11 if it does not give problems. Who, in his right mind, would deliberately reduce performance by a factor 11?
The periodic table, I forgot about Adobe plagiarising atomic structure to introduce their own versions of the "Elements". I was always under the impression that "Pr" was a lanthanide element of name Praseodymium, however, never let science get in the way of commercial opportunism.
Thank you for the explanations and your help however, I fail to see how LucidLogix Virtu will reduce performance by a factor of 11 if there is an NVIDIA card in the machine, that is impossible. Either the NVIDIA card will process the rendering or the i7 sandybridge CPU/PC RAM will process the rendering, how can there be a performance degradation of 11 fold?
It will only happen if LucidLogix takes over. If not, then there is no sense installing it in the first place, because by the nature of the product, it will remain resident and run in the background, stealing CPU cycles and memory and thus reducing PR performance, not to speak of potential incompatabilities.
Speaking in terms of the periodic table, where you start with
you may end up - after installing LucidLogix Virtu - with
which is very short lived. LV stands for LucidLogix Virtu of course, not Livermore.
High-end video cards are good at what? Processing graphics (video), and lots of it! Therefore, Adobe wrote the 64-bit Premiere to take full advantage of the hundreds of CUDA cores found in nVidia cards to unleash the parallel-processing goodness onto HD video editing, accelerating the effects rendering and such. To suggest that LucidLogix would switch the load over to the CPU for improved performance is totally backwards and counter-productive, which is what Harm is trying to tell you. The Mercury GPU acceleration relies on the nVidia GPU, not the CPU.
Get yourself an nVidia card and you will see a great performance improvement in Premiere!
Adobe has a short list of approved nVidia cards, found here - http://www.adobe.com/content/dam/Adobe/en/products/premiere/cs6/pdfs/compatibiltiy-list-pr emiere-pro-cs6-only-hw-compatibility.pdf but many others will work with a minor software modification, just Google "Studio 1 hack" for more info.
This video explains how the GPU acceleration benefits Premiere - http://tv.adobe.com/watch/short-and-suite/gpu-acceleration-and-nvidia/
Would you please be able to clear up a point for me?
In the MPE Gain Chart there are performance figures listed for quite a few NVIDIA cards that are not on the current Adobe Premiere Pro CS6 recommended list.
For example, the current Adobe list has GTX 285, GTX 470, GTX 570 and GTX 580 as recommended cards and all of these are in the MPE Gain chart. However, not on Adobe's list but in the MPE Gain chart are cards such as GTX 550 and GTX 480 plus a good few others.
Are these cards that are in the MPE Gain chart but are not in the Adobe list the ones where they need to be unlocked or hacked to enable the MPE? Or, do they have MPE Gain out of the box without any firmware or software updating?
Thx for any advice you can provide.
When you install PR or an update, like CS6.01, the file:
'cuda_supported_cards.txt' is created new, so each time you update and your video card is not in the list of this text file, you have to add it manually. That is the 'hack', just a simple text line added to this file. Once you have done this, you can use the hardware MPE acceleration.
The people who have submitted their data have done this all the time. That is why we now have 625 systems with 13 different video cards in this chart, while only 5 of these are officially supported. Remember that to be included in this chart, there have to be more than 12 observations after correcting for outliers. So there are even more 'hacked' video cards in use, but they are excluded from the chart, because there are not enough observations to publish statistically reliable data, as are the supported cards like the Quadro 5000 and 6000.
If you look at the chart now, you may get the impression that the 680 (avg. 4.92 s) is way faster than the 570 (avg. 8.91 s). This is caused by two factors:
- Limited number of observations for the 680, and
- 680's are only installed on the very latest Intel PC's. The older generation is also installed on much older and slower systems, resulting in slower figures.
The decisive factor for video card performance is memory bandwidth. CUDA cores are important, but not as much as the memory bandwidth.