This article is about the choice of the video card. It is often asked "what kind of video card should I use". The answer is not so simple. It depends.

One of the most asked questions is about the video card. What should I use? The answer is not very simple. It depends.

Everybody wants to have the best performance within his/her budget. That is completely natural, but what is realistic and offers value-for-money, or - in my terms - BFTB 'Bang-for-the-buck'?

Note that the list of 'supported cards' by Adobe is completely irrelevant, out-of-date and incomplete. The only thing that matters is that the video card has at least 1 GB of dedicated VRAM available.
For CS6 the 'hack' may be necessary to enable hardware MPE for nVidia cards only and for CC even that is no longer necessary.

Let's have a look at your workflow. Is it all about rendering for previews? You want to have a snappy experience when scrolling through your time-line? You use many accelerated effects? You don't use third party plug-ins like Neat Video, Magic Bullet Looks or Colorista often? OK, you are in category A.

Rendering for previews is not really important to you? Don't use many third party plug-ins? Your main interest is in speeding up export times? OK, you are in category B.

This already looks contradictory: 'Don't use many third party plug-ins' mentioned twice for both category A and B. This is one of the things that make the question about which video card to use so difficult to answer. Well, the first thing you should note is that third-party plug-ins hardly, if ever, can use hardware acceleration. Mostly those plug-ins are ill-suited to use a video card at all to improve on rendering speed for previews or encoding. Often they are also threading miserably and do not use the full capacity of the processor, slowing things down significantly. The only exception I know are Pixelan plug-ins like Dissolve Master, that thread very well and use the video card extensively. It is my experience with Dissolve Master that even 'red line' segments of the time-line do not need to be rendered for fluid playback.

Category A

A lot of editors fall in this category, since they hate choppy or jerky behavior when scrolling through their time-line, they regularly want to check if their CG work is exactly as was intended, the looks of their video transitions and effects are correct, they have the client present to show progress or similar situations. If any of these apply to you, then a hardware accelerated video card is very advisable. It can make a huge difference in performance.

Later on I will show you a table with suggested video cards that would make a nice addition to your system to help performance, but AMD cards are not among them, nor are nVidia Quadro cards, as will be explained below.

Category B

If you belong to category B, the first thing you should do is to read Tweakers Page - Exporting Style to help you understand in what situations a top-notch video card, or even two cards in the case of CC can be helpful and when the impact of the video card is minimal. Depending on your export style, it is quite common that the video card is not used at all or only minimally.

Often people look at GPU usage during export and are surprised that the GPU is not used. That clearly shows that encoding is CPU bound. That is not a bug in the software, that is a bug in (unfounded) user expectations.

Only if you export a lot with resizing, frame rate changes or blurring going on, is a good video card with hardware acceleration helpful to reduce encoding times. Otherwise you will profit more from extra cores, higher clock speeds and bigger L3 cache. Dual video cards are generally a waste of money for Category B users, since they are idle most of the time anyway. If the majority of your work does not use resizing and usually consists of exporting HD to full HD 1920 x 1080, there is no sense to get a top-notch video card, since it will hardly be used, if at all. The CPU carries the burden of the encoding.

General advise


For the moment ONLY nVidia GeForce cards are advised. CS6 does not support AMD cards and AMD is still considered the 'New kid on the block' as far as CC goes. AMD is still in its infancy, the drivers and Adobe support is still too new to effectively compete with nVidia's CUDA acceleration, which has been proven in the past years and have stable drivers from nVidia and stable support from Adobe. nVidia CUDA support is a mature technology with a proven track record. Add to that the recent price increases of AMD cards and they have effectively outpriced themselves in the competition with nVidia.

FYI: Among the 500 supercomputers in the world, more than 20% use GPU processing to accelerate performance, like hardware MPE. 67% use nVidia and only 3% use AMD. Just to let you know...

Benchmark results with AMD cards show they are about two times slower than equally priced nVidia cards. Even the latest AMD cards lack support for feature level 12_1 in DirectX 12 in contrast to nVidia's latest cards.

nVidia Quadro cards are a different story. The simple reason is that all Quadro cards are way overpriced and underperforming in comparison to the GeForce cards. The ONLY reason one may choose a Quadro card over a GeForce card is when 10 bit output to a costly 10 bit monitor is required. See the graph below. Based on August 2013 prices in Europe excluding VAT and only comparing to GeForce cards that perform equal or better than the Quadro card in question. The Green bars show the relative performance index of the GeForce card when compared to the Quadro card. So for example, the GTX 760 is 11% faster than the K5000, despite being almost 1/8 of the price. The Red bars show how much more the Quadro cards cost compared to similar performing GeForce cards.


Even the much touted 'Maximus' configuration, at least by Adobe, is ill advised. It is a very sure way to blow vast amounts of money away, but without any tangible performance gains. A single GeForce easily outperforms a 'Maximus' combination of a Quadro and Tesla card for a fraction of the price. With that out of the way, let's talk about the amount of memory on the video card, but remember that only GDDR5 memory should be considered. GDDR3 is just too slow.

What about the amount of VRAM per installed card?

As a rough guide, how much VRAM should be installed on the video card (with a minimum of 1 GB GDDR5), think along these lines:

  • For SD material, 1 GB is enough
  • For HD material, go for 1 - 2 GB
  • For 3K or 4K, go for 2 - 4 GB
  • For 5K+, go for 4 - 6 GB

Keep in mind that the requirements can vary significantly, depending on the complexity of the time-line. If it happens that the VRAM is depleted (not enough memory), hardware MPE is automatically turned off rather ungracefully and stays off for the rest of the encoding. If you use two video cards in CC, each card uses its own memory and cannot use the memory on the second card. The rendering or encoding process works like this: Frame 1 rendered on card 1, frame 2 rendered on card 2, frame 3 rendered on card 1, etc. If either card runs out of memory, again hardware MPE is turned off. Simply put, two video cards with 1 GB each can not be treated as a single card with 2 GB, so the risk of turning off hardware MPE automatically is bigger with these two 1 GB cards than with a single 2 GB card. Frame renders on each device are done independently with their own memory, sharing no GPU resources.

Luckily EVGA recently accounced the 6 GB version of the GTX 980 Ti, making it a solid option for a Warrior system, a junior version of the Titan.

What about the brand of video card?

Since all brand-names have almost identical specs, be it Asus, EVGA, Gainward, Gigabyte, MSI, Palit, PNY, Zotac and all the others, your decision should be based on price / warranty and clock speed. Performance of a video card is mainly based on memory bandwidth, so a higher clock speed will improve performance over slower clocked cards. EVGA is usually the brand with the best clock speeds on certain models (the Superclocked range).

Now let's get down to business. What kind of card do you profit from most? This can not be answered without considering the rest of your setup. It is all about balance, as explained here: Tweakers Page - Balanced Systems

What kind of video card should I use?

It is a complete waste of money to install Brembo disc brakes or a 7-speed DSG gearbox in a Dacia Duster with a 1.2 litre engine. Or exchanging the engine with a 5.7 litre HEMI V8.

As already said in the Tweakers Page - Balanced Systems article, the video card needs to be in balance with the other components to be considered a wise investment.

Consider what happens when you render a time-line for preview or encode it to the destination format.

  • The source material is fetched from disk.
  • Once the source material is loaded from disk, it is modified to an intermediate format and put into memory.
  • If CUDA acceleration can be used, the intermediate format is uploaded to the video card for processing.
  • In the case of dual video cards, frame 1 is handed over to video card 1, frame 2 is handed over to video card 2, etc.
  • The intermediate results are then downloaded from the video card into memory.
  • The CPU starts encoding to the final delivery format.
  • The final results are then written to disk.

As always, the performance and speed of this complete process depends on the weakest link in the chain. That is the reason the table below considers number of cores, amount of memory, overclocking and disk speed to suggest a video card that is 'in balance' with the rest of the system. Implicitly L3 cache is also incorporated in this overview.


In the case of Category A, there is something to be said for looking at economical or warrior type of cards, because of the frequency of rendering for previews. However, for Category B it might be the opposite, you may look more to budget cards. But if you use third party plug-ins with some regularity and those plug-ins are ill-threaded and do not use hardware acceleration, there may be a solid reason to choose a lesser video card.

The attentive reader will recognize that the uploading from RAM to VRAM and the downloading from VRAM to RAM is largely determined by the memory bandwidth of the video card, the QPI speed and the size of the L3 cache, so it is important to use a PCIe-3.0 16x slot for the video card (because of the significantly higher efficiency of PCIe-3.0 over PCIe-2.0) if your chipset allows that. Selecting from many brands of video cards should, apart from price / warranty, be based on memory bandwidth. Of course you have to figure out what local prices are and what warranty applies in your country. There are just too many differences between countries to give an outright brand to suggest.

If you consider using two video cards, beware that the easiest and least troublesome way to do that, is by using two video cards from the same family and preferably the same generation, so either all GeForce cards, or all Quadro cards. Mixing them can cause serious headaches. The same generation means all cards belong to the 6xx series or all cards belong to the 7xx series and in the case of Quadro all are K-series.

Considering to upgrade your existing video card with a newer one ?


Note that there are significant architectural changes with Maxwell 2 in comparison with previous generations.
For Premiere Pro editors the most significant changes are the fewer CUDA cores compared to Kepler and the memory bus of only 256 bits.
However, next to the improvements mentioned above (ROPS, L2 cache, etc.), the much higher clock speed of the GPU can make the GM204
a worthy successor of the 700 range of cards. How much faster it will be remains uncertain now. We are awaiting the first benchmark results.
  • Claims by nVidia of 30 - 50% performance improvements are probably overly optimistic, but 5 - 10% may well be possible.