Graphics cards and Volocity

I have been thinking recently about replacing my graphics card so Volocity will run even better. Volocity Visualization in particular does some very clever things which rely on technologies implemented in the graphics card. Without these our experience with Volocity Visualization is not as it should be.

 

I was not sure what kind of graphics card I should get, so with the help of one of my post-docs, we had a look around and I want to share with you some of the useful information that we found out.

Reading the system recommendations seemed like a chore, but it provided us with all the information we needed.

Let’s start by looking at the system recommendations, concentrating only on the graphics cards. What we are looking for is a single graphics card (single GPU) with at least 512 MB video memory. More memory is better as long as the other requirements are also met. Multiple cards in one machine will not be an advantage for Volocity.

For optimum performance the card must support OpenGL 2.0 or above. This means we are looking for a dedicated graphics card, we can’t rely on integrated graphics as they don’t have nearly enough memory. Neither do we need to spend thousands of dollars on workstation graphics cards such as the NVIDIA Quadro or ATI Fire ranges.  The bias of the technology implemented in these cards is not in the right direction for Volocity and they are very expensive.  We can therefore save our money to upgrade the RAM in our machine instead.

If we wanted to buy an Apple Macintosh, then life is reasonably simple and the above rules make the choice easy.  When we looked on the Apple website (http://www.apple.com/macpro/), we saw that we have six choices.  After we had ruled out cards with too little memory, multiple cards and the most expensive NVIDIA Quadro option we were left with the NVIDIA GeForce 8800 GT 512MB - an excellent choice. Not only does it perform well, it also comes third out of the six when ranked by increasing price- that’s good sense in my book!

Nvidia GeFore 8800GTWatch out for that OpenGL 2.0 requirement.  Graphics drivers in Mac OS X 10.4.11 might not support OpenGL 2.0 for older cards.

However, we wanted to buy a card for our Windows machine, so we had to do a little more research, but it did not take long and as I’ve said before, a little effort now saves time and money later. Luckily for us there is another group of users with far greater combined purchasing power than us.  They are the people who want to play high speed, highly realistic games in 3D environments. The two major manufacturers of video chip sets are competing for these customers and developing better chip sets all the time. If a card is desired by these customers then it is a good indication that it should be considered for a Volocity system too.

Following the same rules as above about memory and type of card, we came to the conclusion that the NVIDIA GeForce GTX 200 series, particularly the 280 and the 285 are best in class. The chips are incorporated into cards by manufacturers.  Cards based on the same chipset are all pretty similar so our advice is to go for the manufacturer that supplies what you need, i.e.

  • The right type of PCI card for the motherboard
  • Drivers available for OS

Advances in the drivers can make a difference to performance so it is always a good idea to download the latest version from the NVIDIA or ATI website.

nvidia ati logohttp://www.nvidia.com/page/home.html
http://www.amd.com/us-en/

Then, thinking about the machine, we need to remember three other points:

  1. These cards are big - they typically take up the width of two expansion slots in the machine.
  2. The recommended cards require a lot of power; this will be an additional load on the power supply which may need to be upgraded.
  3. These cards generate a lot of heat and have large fans to cool the processor - this is the main reason why they are so big.  Make sure the main cooling for the case has not been impaired in any way.  A card that is not supplied with enough power or runs too hot will at best run with advanced features disabled at worst damage the machine.

The next problem we ran into was with availability of the card we wanted in the system that we’ve configured. When it was not possible to configure exactly what we wanted on the supplier’s website, we decided to buy the machine with the cheapest graphics option and replace the card ourselves. It’s a small inconvenience but will save us time and money in the long run. In the Vipoir lab, we budget for an upgrade every 2 or 3 years.  We spend money keeping our software up to date so we want to take advantage of new functionality.

Let me know if you’d like me to look at any other aspects of the system recommendations in more detail in another tip (simon.vipoir@improvision.com).