Ryzen review: DCS 1.5 and 2.0 at 1440p and VR

This is a two part review where I take a look at Ryzen, the GTX 1080TI, 1440p, and VR across both DCS 1.5 and 2.0 (NTTR only). First, the system specs:

CPU: AMD R5 1600X clocked at 4.0 GHz all cores.

GPU: EVGA GTX 1080 TI SC2 ICX (@ factory overclock only)

RAM: 16GB DDR4 @ 2667 MHz

Motherboard: MSI B350 Tomahawk

OS: Win 10 Home

Monitor: 27” 1440p TN panel (supports 60Hz and 75Hz)

VR: Oculus Rift CV1 (purchased a week ago)

All tests use the following settings: high preset and for VR, pixel density in DCS set to 1.4. Oculus home was closed for 1440p tests and minimized for VR testing. Additionally, I turned off ASW for the VR tests.

An important note about VR: VR has a highly rigid display timing of 11ms, 22ms and 44ms. This is due to always on v-sync plus a second synchronization between the left and right eye. As such, it displays 90 fps, 45 fps and 22 fps. It is physically incapable of showing 60 fps, as an example. It can alternate between 11 and 22 ms frame timings, giving the illusion of 60 fps but it will also be super microstuttery when doing so, since it can't actually do the 16ms required for 60 fps.

Starting out with 1440p in DCS 1.5, I spawned at Batumi airfield as that particular area is the absolute worst performing area in all of DCS 1.5 and 2.0. As you can see from the two following pics, facing forwards pins core 11 at 100% while the GPU is at a lowly 46% usage, fps is 129 though. Looking to the right tanks it to a pathetic 39 fps with only 25% GPU usage. This isn’t a one off, the rest of the 1.5 test goes the same way for both 1440p and VR.

With that out of the way, I took off and headed north along the coast. Flying over the town of Batumi matched what was seen from the runway, almost no GPU usage and horrible fps when looking in some directions, but then shooting up past 90 when looking in others, and not just out to sea either.

Once past the CPU crushing city of Batumi and getting a little altitude, things improve dramatically but are still highly dependent on view direction even if there isn’t much of a difference between them in terms of what’s drawn on screen. This spot has fps from 120 to the high 50's, depending on where I looked. Once again, note the heavy CPU bottleneck, the GPU is barely touched regardless of view direction.

Moving to everyone’s favorite part of the map for dog fighting, we see much improved frame rates, though it’s still the CPU holding us back. But at 150+ fps, does it really matter?

Moving to other parts of the map has results everywhere between the two extremes. DCS 1.5 is incredibly CPU dependent in all areas, often ignoring the GPU almost entirely. Overall I think 1.5 is certainly playable with the 1600x and in some areas where there are fewer draw calls, it runs great. But then there are spots like Batumi where it chugs. It’s worth mentioning that the I5-4690 this chip replaced had it even worse. On average, I gained about 10 fps in the Batumi CPU grinder at any given graphics setting.

Now let’s look at VR, but I’m sure you can guess how this is going to go…

Yikes. It doesn’t get better once the wheels are up either…

So where does it get better? Getting away from Batumi of course. In the same places the 1440p result got above 60 fps, VR finally hits the first proper display step of 45 fps. In the mountains and most of the plain areas, it runs pretty well. And hell, I even got 90 fps when consulting the witch doctor, sacrificing a goat to the gods, chanting the correct verses and waiting for the stars to align properly.

But I really wouldn't bet on 90 fps. And don't think a 7700k is going to save you either. It's 25% faster in single thread, due to 25% more clock speed when on liquid cooling and hooked up directly to a high voltage powerline.:P

Final thoughts on 1.5:

The engine is trash, everyone already knows that, including ED which is why they are remaking the map to use the newer engine. Despite that, the R5 1600x handles it far better than my previous I5-4690 ever did. In the Batumi CPU torture test, the 1600x is, on average, about 10 fps faster at the same settings. Additionally, the low fps is due entirely to draw calls. This creates some interesting options in VR, since pixel density can really improve the quality and it falls entirely on the GPU to handle. I chose 1.4 PD for this test to keep it consistent with 2.0 and NTTR. When actually playing 1.5, I do not fly near the Batumi/Kutaisi area and I also crank that PD up to 1.7 and am considering going even higher. In fact, no reason not to run on anything less than super ultra detail except with medium view distance and 10km trees, as those are the only two things calling on the CPU. That applies to 1440p as well. Hell, I could probably run a triple 4k setup without losing a single fps over Batumi.

I know DCS 1.5 wasn’t encouraging, but this review is about to get a whole lot more positive. Let’s start with NTTR at 1440p. Spawning in the lovely F-5E at Nellis, I saw 130 fps when looking ahead. Looking to the sides had little impact, the lowest being 122 fps.

But what about over Vegas? This is typical:

Note the 99% GPU usage above, we have a GPU bottleneck with a 1080 TI! It’s at 109 FPS, but still, it isn’t fast enough for the 1600X here.

Heading out of town brings fps up to 140-160, it largely stays in that range while on the deck. And it isn't just in the open desert either.

Climbing up a bit and looking back at Vegas give us a very respectable 130-ish fps. If only 1.5 ran this well when looking at Batumi.

And look at those mountains (and the fps).

Right, so that went amazing well. So how’s VR? Starting over back at Nellis, looking ahead shows an uninspiring 45 fps, but both the CPU and GPU are well over 50% usage at least. Looking to the side changes nothing, except upping GPU usage slightly.

Heading towards the strip brings the GPU usage up but has no change otherwise, it remained at 45 fps all the way into the main event.

Getting out of town finally sees the FPS go up a little bit, but it is a bit jittery given it’s a 90Hz display, not 60Hz.

It also isn't consistent, it doesn't take much to drop it right back down to 45 since none of the frames reach the 11ms mark.

We could drop the graphics to low and try to get 90 FPS or up them a little to fully utilize the GPU while accepting a permanently locked 45 fps, but that seems like a poor choice. I see only one sensible solution to this problem, graphics level: stupid. Setting everything to as high as it will go (forget flat shadows, use high and default!), including setting PD to an utterly insane 2.5 and we have…

And thus I successfully made NTTR run as badly as DCS 1.5 near Batumi, though due to a very different bottleneck! Incidentally, this is effectively displaying DCS at absolute max detail at 7020x3900. That’s 27.4 million pixels, or equivalent to a triple 4k setup. So what about stupid detail at 1440p? Here ya go. Still totally playable.

Final verdict on 2.0, works pretty well. 1440p runs fantastic and you can pretty much max it out with the 1600x and GTX 1080 TI. If in a large battle, you may want to tone it down slightly for a bit more GPU headroom for explosions and such, but it works. You probably aren't going to have a massive battle in the Las Vegas strip either, then again... I have this great MP mission idea!

For VR, I find it's smooth enough. Realistically, it's never going to go above 45 fps but you can throw an awful lot at it with high detail settings and still keep that 45 fps. End result, it's a good experience. I greatly prefer NTTR in VR over a regular monitor at 1440p. Spotting isn't a problem either. The resolution is poor, but that just means the pixels are bigger, thus easier to spot. My actual VR settings are basically high, but with the full shadow effects instead of flat nonsense. It looks really, really good despite the low resolution like that. With ASW on, it's even pretty smooth, though there can be some graphical artifacting with props and rotors.

Great information, thank you! Unfortunately, I wouldn't hope for the 2.0/2.5 engine to provide better CPU utilization.

And you'd be wrong, should have waited for part 2. :P

What's going on here is 1.5 is basically DX9 pretending to be DX11, thus nearly all graphics processing is done on the CPU itself. DCS 2 is properly implemented DX11 and thus it mostly falls on the GPU to process. The CPU still has to pass it the basic model info, hence the high cpu usage in an empty mission, but it is dramatically lower than if it had to do everything. One can only imagine how bad it would be over Vegas if it were rendered in the 1.5 engine. There is still a lot of room for improvement though. Splitting the graphics thread off from the game logic (including physics) would free up a fair bit for more units as well as probably allowing 90 fps in VR. Another thing that can be done is having AI units occupy n threads instead of just 1. That's a relatively easy thing to do and would allow for some crazy big missions for those of us with high core count CPU's.

Unfortunately this does have its limits with DX11 as the graphics will still be single threaded. In order to have multiple CPU cores feed the GPU, DX12 or Vulkan must be used. I don't see DCS making that switch anytime soon. I tried to teach myself Vulkan and it's, umm, complicated. OpenGL and DX11 are idiot proof by comparison.

Grest write up Beastybaiter!

So basically 1.5 is CPU limited and 2.1 is GPU limited? Im tempted to go for a 1600x from my 2700k@4.6ghz but if 2.5 is gonna be GPU limited then im better of going for a GPU upgrade.

2.0 isn't totally free of CPU usage for the graphics, you can't see it in the 1440p results but CPU usage is very high (90%+ for nearly all of it). It's simply that the GTX 1080 TI tops out first in most cases. These were empty missions other than the player aircraft, so in a more interesting mission, a cpu bottleneck is likely. Though it shouldn't be a problem, I really haven't had any issues with it in any sort of reasonable mission. Getting away from Vegas will also collapse CPU usage. You can see that in the 45 fps VR result when looking at a mountain. So in general play, NTTR runs very well in both VR and 1440p. Sometimes it has a GPU bottleneck, other times CPU, but in both cases it stays pretty smooth so long as you don't go too crazy with the AI units.