Date: Wednesday, February 08, 2017
Over the course of the last week we have dove into the depths of AMD and NVIDIA drivers. We have tested AMD and NVIDIA video card drivers on the latest GPU and previous generation dating back to summer of 2015. It has been a grueling, intense, detail oriented, yet an enjoyable and nostalgic visit into GPU performance over time in eleven games. We have learned a lot from this testing, and were reminded of several performance oriented game facts as well. Overall, what a great experience and a ton of work, not that I am complaining.
In our first evaluation, AMD Video Card Driver Performance Review, we took an AMD Radeon R9 Fury X and AMD Radeon RX 480 and tested drivers from launch (or first Windows 10 release driver) to current Crimson ReLive Edition drivers.
We had a positive experience of performance progression over time in a linear fashion of about 5% to 6% on both video cards. This was particularly surprising on the AMD Radeon RX 480 since it has only been out 6 months and has had less time to mature, yet AMD seems to have pulled out some consistent performance gains in that short time.
In this evaluation we have now done the same testing on NVIDIA drivers and video cards over time. We have taken the GeForce GTX 980 Ti and tested from the first official Windows 10 release driver to the now current driver 18 months worth of driver updates. We have also taken the GeForce GTX 1080 and tested from its launch driver to the latest driver culminating 8 months of driver improvements. Our results are solid, and we can come to some conclusions.
Simply put we did not see a lot of NVIDIA driver performance improvements over time on either video card. When we did experience driver performance gains over time the actual percentage difference was very small, generally well under 5%. There was the occasional game where a bug fix, or new game support was added that boosted performance. However, from "Game Ready" driver to "Game Ready" driver, the improvements were nil to none.
In newer APIs like Vulkan API in DOOM and DX12 in Gears of War 4 we saw the largest improvements on the newer GeForce GTX 1080 video card using the Pascal architecture. In Hitman DX12 for example the GeForce GTX 1080 received performance improvements, while the GeForce GTX 980 Ti was stagnant.
In Gears of War 4 we saw a greater improvement with the GeForce GTX 1080 than we did with the GeForce GTX 980 Ti. The Maxwell GPU GeForce GTX 980 Ti does not support Async Compute in Gears of War 4. However, Pascal GPUs (GTX 1080/1070/1060/1050) do support Async Compute. Only in DOOM with the Vulkan API, did we see both improve almost equally.
Outside of Vulkan API and DX12 API games, both the GeForce GTX 980 Ti and GeForce GTX 1080 were rather stagnant in terms of performance updates in DX11 games. We just did not see anything worthy of noticing while playing or changing the gameplay experience. Most performance differences fell within what we would consider a margin with the real-world game testing we do.
In our AMD Video Card Driver Performance Review Conclusion we specifically pointed out how well the AMD Radeon R9 Fury X is holding up in game performance. With the GeForce GTX 980 Ti we were able to play every game well at 1440p with the highest in-game settings. Only Watch Dogs 2 performed poorly, but it was also poor on the AMD Radeon R9 Fury X as the game has proven to be a real graphics hog.
When we compared AMD Radeon R9 Fury X and GeForce GTX 980 Ti performance using the latest drivers we found these to be very competitive. Both video card launched at an MSRP of $649, and today seem to be almost evenly matched in performance. We did find that for the most part the GeForce GTX 980 Ti slightly edges out performance over the AMD Radeon R9 Fury X, but we are talking by small percentages. When it comes to a game like Hitman DX12 we saw the AMD Radeon R9 Fury X stomp all over the GTX 980 Ti.
This leads us into some thoughts we have about DX12 on the Maxwell architecture (GTX 980/Ti) series of video cards. According to our testing in Hitman DX12 and Gears of War 4 DX12 the GeForce GTX 980 Ti did not receive or benefit from performance upgrades via drivers. Whereas, the GeForce GTX 1080 (Pascal GPU) did. We also saw improvements to DX12 games on both AMD GPUs.
The Pascal GPUs (and the AMD GPUs of the last couple of GCN iterations) support DX12 and are able to optimize performance with it better. This is something important to consider as more DX12 games are released and support new features like Async Compute.
It seems compared to AMD, that NVIDIA driver improvements are more about bug fixes, and "Game Ready" day one game support, rather than performance improvements over time. On the other hand AMD focuses on performance improvements alongside bug fixes and new game support over time. This may very well be a question of resources for each company and access to game developers.
In this review, and the previous one, we have seen a clear pattern of AMD having a more consistent progression of performance on its GPUs from driver to driver over the course of the video card's lifetime compared to NVIDIA. The 6 month old AMD Radeon RX 480 has achieved the same performance advantages as the AMD Radeon R9 Fury X has in 18 months is quite telling. On the other hand, the 8 month old GeForce GTX 1080 has been flat in terms of performance improvements. The performance you got out of it at launch is the same performance you get out of it today.
So that begs the questions... Does AMD launch video cards with performance left on the table in terms of drivers? Does NVIDIA launch video cards that are optimized to the utmost out of the gate?
Does AMD keep its driver engineers' noses to the grindstone eking out every bit of performance that it can find as time passes? Does NVIDIA let performance optimizations go undiscovered over time?