What matters the most on a GPU: Fancy tech, the Bottom-line performance or something else entirely?
Recently there have been a lot of controversy around a benchmark behind a game from a developer called Oxide games. The game is part of the first batch of games to use the new industry standard DX12 API. The new API promises better performance and better graphics, without having to do any hardware upgrade. Awesome right? Yes, but upon running the benchmarks it shown that the Nvidia cards were not performing better than the AMD cards, and in some cases, even performing worse in this new API. How is it even possible? It can be due to a lot of reasons, but all which points to the preliminary status of the drivers and the game in question. It is the first real demo to showcase DX12, and like everything being first, it brings with it a lot of kinks which will be ironed out further down the road. Unfortunately people just love to jump on the gun and decided took the preliminary results very seriously. The same people started to question whether the Nvidia cards really do support DX12 as they advertised. It is a long story, but PCPer covered it really well.
The Oxide game (an AMD sponsored title) is not end all, be all game for DX 12 showcase. Nvidia and Oxide have recently released some information regarding how a future driver and Oxide game update will implement parallel processing or Async as people like to call it. As of right now, the current implementation of Async in the Oxide game is done in favor of the GCN architecture — AMD architecture. Once Nvidia fully enables Async and more DX12 games are available, with the nearest ones being Gears and Fable, only then we can have a better picture of things stand.
Even if you look to seriously correlate DX12 performance of the cards directly with Oxide game benchmark, things aren’t even as bad as many people (fanboys) tend to make it out. Yes, it is true that AMD GPUs are performing significantly better in DX12 mode compared to DX11 mode, but here is the catch, the AMD cards were performing at an abysmal level to begin with in DX11 mode. It is no secret that Nvidia drivers are the better of the two, part of the reason is because of low driver overhead, but that is a different story entirely so let's focus on the DX11 performance only. The DX11 performance of Nvidia cards are leaps and bounds better than the AMD cards. So when you are moving to DX12 which brings additional performance for Nvidia, it doesn’t look much of an improvement. So in essence DX12 is giving AMD owners a massive boost in performance over DX11 but NOT a massive boost in terms of effective performance.
Also do note, companies like Nvidia and AMD don’t pull out architectures out of their asses, they follow a roadmap. Another scenario might be that Nvidia deliberately didn’t implement Async in Maxwell as a design and economic choice: why use something which won’t be used for several years at least. DX 12 adoption will start but it will take several years for it to take the place of DX11. Before that even happens, Pascal — Nvidia’s next-gen GPU — is going to come and shake things up.
So what matters the most? What makes a card a winner? In my opinion it is simply not just about ticking the performance checkbox or the bragging right of having a fancy tech which makes it decisive. It is about the entire package — the overall experience you get when you enter the manufacturer’s ecosystem with the purchase of their GPU. At the end of the day, big numbers don’t mean much if the hardware in question gives out equally big headaches. So you might want to consider this before buying your next GPU or any computer hardware or device for the matter. Thanks for reading and see you on the next one!