NVIDIA VOLTA's TITAN V DX12 Performance Efficiency

We take the $3,000 NVIDIA TITAN V video card using the new Volta architecture and find out if its DX12 and Async Compute performance is more efficient than the last generation Pascal architecture in gaming performance.


Last week we evaluated the $3,000 NVIDIA TITAN V in thirteen games at 1440p and 4K for a full gaming performance evaluation. In that evaluation we found the NVIDIA TITAN V to perform between 30-40% faster than the GeForce GTX 1080 Ti in today’s games at either resolution. The gains were real, massive, and constituted a greatly improved gameplay experience.

Article Image

In that evaluation we used a mix of DX11, DX12, and Vulkan API’s to show the best overall picture of performance in games. NVIDIA TITAN V is based on a new architecture named Volta which supersedes the Pascal architecture which GeForce GTX 1080 Ti is based on. Given this information we wanted to find out if NVIDIA TITAN V using the new Volta architecture is in anyway shape or form more efficient at DX12 performance versus Pascal architecture.

DX11 vs DX12

Has NVIDIA improved DX12 and Async Compute performance with the Volta architecture? These were hot topics during Maxwell and Pascal’s reign. The general understood consensus is that AMD Polaris and Vega GPUs better handle DX12 and Async Compute performance, are more efficient at it and show a performance improvement advantage.

Since we had some time with the NVIDIA TITAN V we wanted to explore if DX12 and Async Compute efficiency and performance has improved with Volta versus Pascal. Now, this is hard to do apples-to-apples because the TITAN V is so much faster than the GeForce GTX 1080 Ti. Therefore, we cannot compare each video card’s gaming performance to each other. Instead what we must do is compare the difference between DX11 and DX12 performance on TITAN V to the difference between DX11 and DX12 performance on GTX 1080 Ti. This is not a "perfect" way to go about this, but we thought that the comparison still might have some merit and watned to see exactly what the data had to say. Given that TITAN V is the only card right now that gives us access to the VOLTA architecture, the comparison to the gains on Pascal still might at the very least be of interest to some of our readers.

In the games that we tested there were a few games that allowed us to play in both DX11 and DX12 APIs. For those games we took extra data on TITAN V while we had the video card it so we could put this review together for you today. We of course are going to compare with the GeForce GTX 1080 Ti Pascal architecture video card.

Keep in mind that for this review we are not looking at the raw performance between each video card, but rather how much better or worse DX12 performance is compared to DX11 performance on a given card versus the other one.

For example, we will start out looking at two distinct facts. Number one, we will first see if DX12 is faster or slower than DX11 in said game on each video card. Number two, if performance is slower or faster we will take that difference and compare that difference to the difference between the competing video card. In this way we can find out which one is more efficient at DX12 and Async Compute.

For example, let’s say in a game that turning on DX12 on Card A causes a performance deficit of 10%. However, on Card B turning on DX12 on Card B also causes a performance deficit, but the drop is only 2%. That would mean Card B is more efficient, there’s less of a drop in performance, even if overall DX12 is still slower than DX11. However, if let’s say Card B has a positive impact with DX12 and performance is faster, while Card A is slower, then that of course shows Card B is better as well at DX12.

It sounds a bit confusing but follow along in the text and we’ll break it down nice and easy under each game on the following pages. In the conclusion we’ll talk about it and come to a conclusion if Volta is better at DX12 performance than Pascal was.

While we are only using a handful of games, we do recognize this can all change with future games and DX12 support. Also keep in mind that TITAN V is not a gaming video card and does not have as optimized drivers for gaming as GeForce GTX 1080 Ti does. This is important to keep in mind, but surely the TITAN V gaming drivers have matured greatly since its launch.

Game Setup

The games we are using are as follows, with some information:

Deus Ex: Mankind Divided - In this game we have generally found DX12 to be slower than DX11 on Pascal. Therefore, we will see what happens with TITAN V and then what the difference is between DX11 and 12 even if it is slower, which is "more slower" etc.

Rise of the Tomb Raider - We have also found performance to be slower in DX12 in this game, so again we’ll see what happens and how much slower each card is.

Sniper Elite 4 - Now with this game we have found DX12 to be faster on Pascal GPUs in the past. In fact, every video card so far, we’ve thrown at this game is faster under DX12. Therefore, it will be interesting to see how TITAN V does. Will it be much faster at DX12 than Pascal, or not? Also, this game uses Async Compute.

Tom Clancy’s The Division - This game has varied in our testing. For the most part we have experienced improved performance under DX12 on Pascal, based on the settings used. It does however vary with what settings are being used. Uniquely we’ll be using PCSS Shadows at 1440p which is something we don’t usually test with.

Wolfenstein II - We are testing this game in the Vulkan API but we are using this game to test Async Compute Off versus Async Compute On. This is a newer game and does support Async Compute toggle option in the graphics settings. There was some controversy at the launch of the game as to Async Compute’s benefits on Pascal. It was slowing the game down. However there have been a lot of patches since then, and now we need to find out if its faster or slower. It will be interesting to see how TITAN V handles it as well since the drivers are not very optimized for Vulkan API gaming.