Wow, that website referenced above is one of the most biased sites I've ever seen posted here. Suggest you don't post any of their articles again, unless you want to be laughed at...
Jarnis, I commend you for trying to explain to the people here the choices FM made for this benchmark, but do understand you are talking to some who will just not accept anything you say. Don't take it personal, just the way things are 'round here.
^Making a 'true' empty statement as an attempt to rebuttal a logical argument without actually having any tangible, logical counter-arguments to the article in question. Stating a conclusion without any premises that support the conclusion is the very definition of a flawed argument.
Let's recap what the article states:
1) NV lacks hardware Async Compute - true
2) NV introduced Enhanced Async Compute but they really meant pre-emption because NV has no ACE engines - true
3) NV positions Pre-emption as some killer feature, nearly an Async "equivalent" in its marketing slides. However, pre-emption in prioritizing key tasks must stop existing workload before a new workload begins. This is not true multi-engine parallelism and this is not the same as ASync Compute. Thus, pre-emption's goal is to prioritize tasks but Async Compute hardware engines can issue tasks in parallel - true
4) In real world DX12/Vulkan games, modern APIs show massive gains in CPU performance. In Doom, a 1.2Ghz 5820K gains 70%+ with a Fury X when running Vulkan. In Hitman, FX8350-70 gains 50-70%. There is no correlation between faster CPU and improved 3DMark score. Normally, this is what we want -- a GPU limited benchmark. However, in this case, the time view data shows that Fury X stalls waiting for serial code, which means the CPU has no work to issue. Hitman and Doom do not exibhit this behaviour. Since all existing DX12/Vulkan PC games show CPU scaling and dependency and Time Spy does not show it, it also proves Time Spy has no relevance to real world DX12/Vulkan games as it does not mimic the programming behaviour/trends of real world modern API games.
5) By definition of DX12/Vulkan, programming should move away from AIB (AMD/NV) driver reliance to the developer. Why? Because to get the full advantage of the new API's "closer to the metal" / "removal of the old API obstruction layer", developers MUST optimize the game engine's code to the specific GPU architecture to maximize coding to the metal. This means using Pre-Emption for Pascal and its large L2 cache to the max, while for AMD running as much code as GCN can handle under Async Compute, shader intrinsics, etc. This is how coding works on consoles, Doom, etc.
By definition then since Futuremark failed to create deep architecture specific optimizations, the Time Spy benchmark does NOT compare the true potential of GCN to Pascal architectures under next generation DX12 games. What it instead compares is ONLY how those graphics architectures can run Time Spy synthetic code -- nothing else. This is 100% true. If next generation games are coded to specifically take advantage of Pascal and GCN under DX12/Vulkan, Time Spy had to do the same -- they did NOT. This means Time Spy fails to capture both the GCN Async Compute console port effect and all scenarios where DX12/Vulkan AAA games that will take better advantage of GCN's DX12 capabilities.
Since we already see RX 480 beating 1060 in 80%+ of DX12/Vulkan games, it is clear that real world DX12/Vulkan gaming performance does not accurately align with Time Spy. Since TimeSpy does not try to measure average GPU performance (TPU charts), but tries to predict future DX12 performance, then the performance delta between 480 and 1060 should roughly match what's already happening in DX12/Vulkan games. It does not:
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/73040-nvidia-gtx-1060-6gb-review-21.html
6) It's impossible to have a synthetic DX12 benchmark that tries to predict performance in future DX12 games without taking full advantage of GCN's most important DX12 feature - Async Compute. Since FutureMark is not doing much in the way of GCN-specific optimizations, in effect they are assuming almost no AAA developer making DX12/Vulkan games will do so on average over the next 1-2 years. 280X losing to the 960 also makes no sense as that is not what we see in modern DX12/Vulkan games.
7) By definition, ALL synthetic GPU benchmarks are worthless for predicting performance in the next generation games because next generation games do NOT use game engines which are used by these synthetic benchmarks. That means Unigine, 3DMark, GPUMark, Catzilla, etc. are all worthless predictors of a real world gaming experience a PC gamer will get in a new game. Unless we actually test a new PC game, 3DMark score tells us nothing about how a GTX1070 or an RX 480 will actually run the game as far as the user experience is concerned.
The people who work at FutureMark will defend their work because it's their job. The ONLY way to gauge how well modern graphics cards perform in DX12/Vulkan games is to test them in real world DX12/Vulkan games.
Another litmus test as to why 3DMark is worthless is if it ceases to exist, it changes nothing about our graphics card choices because we should only care about how a graphics card runs a game we play, not some worthless synthetic score.
There is no need to have synthetic power viruses like FurMark or synthetic GPU benchmarks like 3DMark or Unigine when we can test real world applications for which the graphics cards were purchased in the first place. But eh, if you love playing Furmark, 3DMark, Catzilla, Unigine games, by all means defend synthetics!!
The industry just likes a simple, repeatable benchmark that they can run and it spits out a score in minutes. If more games used in-game benchmarks, there would be even less reason for 3DMark. Also, YouTube and online reviewers want to please all their viewers/readers. There is no doubt that there is a fraction of PC gamers who still think useless synthetics like 3DMark, Unigine and Passmark mean anything. To please them, some reviewers still include synthetics just to not alienate some readership that will read another review that does.