The main thing with recent games is that you need to get away from 4 threads.
4c4t CPUs perform quite a bit worse than 4c8t CPUs now - particularly if you are paying attention to 1% and 0.1% lows.
But some games do really benefit from having 8 physical cores.
Deus Ex: Mankind Divided shows a significant improvement with 8c8t vs 4c8t in my testing.
What's also surprising is that it doesn't seem to scale well beyond 8 threads. 8c16t performs worse than 8c8t. I'd be interested in how that ran on a 3950X which has 16 cores - to see if it's an issue with SMT, or how the engine splits up jobs.
It's the 1% and 0.1% Low values which really show a difference moving beyond 4 cores though - look at the 6c12t results compared to 4c8t. Minimum, Average, and Maximum frame rates are almost the same, but 0.1% Low is 37% higher.
Many sites are still only looking at average frame rate, or avg/max/min.
I've also done more testing which highlighted that a basic chart like the one above is insufficient for representing performance in many games. It worked for
DXMD, but you really need percentile charts and "time spent beyond X" charts like the ones The Tech Report use.
My testing in Watch Dogs 2 actually looked worse for 8c16t than 4c8t despite 8c16t clearly being better in-game. The way it scales with cores is quite interesting too. Maximum/average performance goes down a bit, but minimum and 1%/0.1% Lows go up - which is more important for how a game actually plays.
I don't really think the "game and stream" tests that AMD push are really that valid.
For one thing, NVIDIA's GPU encoder now produces higher quality results than the x264 Fast preset which most people use for CPU-based encoding if it's a single-PC setup.
But I've been doing a lot of testing in OBS myself recently for recording and streaming games, and even though I have an 8c16t Ryzen 1700X, I'd never use CPU-based encoding. CPU-based encoding is far more likely to drop frames, and affects game performance considerably more than GPU-based encoding.
Though it does not apply to streaming, it's particularly noticeable once you start trying to record high frame rate gaming at 120 FPS or higher. 180 FPS seems to be about the limit for my GTX 1070 with the resolutions I'm recording at, but using CPU encoding means the games will never reach those frame rates in the first place.
There are no special benefits to pairing an AMD GPU with an AMD CPU.
I would be hesitant to buy any PC hardware right now if you are planning on keeping it around for the entire next generation without upgrading it.
We don't have full specs for the consoles, and we don't know how current hardware will perform in next-gen games. By the time next-gen consoles are out, Ryzen 4/Zen 3 should be available.
If you have a high-end NVIDIA GPU like a 2080 Ti it will probably do just fine with next-gen games, but I'd be concerned about buying mid-range PC hardware today that is supposedly on-par with what the consoles might be offering based on leaked specs. It's probably going to under-perform a year or two after launch.
Personally, I'm planning on upgrading to a 16 core CPU for next-gen.
If the consoles are using 8c16t Zen 2 CPUs, I want the option of putting that work onto 16 physical cores instead - that should help push next-gen games to higher frame rates, since the gap between console and desktop CPU performance will be narrowing.
I do wonder what that means for Intel though. I'd be very surprised if they had 16-core mainstream CPUs, and if they did, they're unlikely to be using the standard ring bus design which has helped keep latency low and plays a big role in their gaming performance. I don't think they've had any CPUs beyond 10 cores which use a single ring bus.
