Performance Issues with Alpha 13

[LuckLock]

Prior to Alpha 13, the regular Solar System simulation ran on my decent computer with an average of 60 FPS. I can barely get 20 FPS with the new update. I have a CPU meter up, and it says I'm only using about 30% of my resources when playing Universe Sandbox. Is anyone else experiencing these problems?

My specs are:
AMD FX-4100 Quad Core 3.6GHz CPU
8GB DDR3 RAM
NVIDIA GeForce GTX 650 1GB

[dylan]

I'm not

[Cryo]

Same here, the collisions are Hella laggy, now one collision  equals lag of 9 at once.....

[C7]

Hi LuckLock, can you please run a performance test and in your results? You can put "Atten : C7" in the feedback email so I can find it easily in our feedback system.

If it's running slowly, you can try one of the older (and less "safe") integrators, that may perform a lot better for you.

Trying to select a different integrator or compute device may help. Sometimes our testing can report a device as performing better then another, when that doesn't pan out in practice. Without getting too technical, some devices don't handle certain cases very well, and will fall off greatly in performance during those times.

Try playing around with the settings and see if something works well for you.

To access the integration and compute device settings, Select the sim icon (little gear in the left corner),
Simulation Settings tab, and look for the "Gravity" section.

In there you can choose IntegatorID and Integration modes, as well as the default compute device, and accuracy restrictions.

A brief overview :
Fixed -> This is the fastest, but will not respect accuracy. May cause bodies to fly out at high timesteps.

Unsafe Global Adaptive -> This is the default for Alpha 12.1, more respect for accuracy, but higher cost

Global Adaptive -> This is the default for Alpha 13, Our new in development integrator. Should have better performance (computationally), but may be slower (in fps) for now. Respects accuracy of the simulation. In a lot of cases this can be faster then the Unsafe version.


In regards to the collisions, try setting the Non Attracting body count to be lower. (We changed the particle spawning logic, so they spawn a minimum amount, which is related to the currently set cap)

To adjust this, go to Home -> Settings -> Performance, and you will find the options there.

We're still collecting performance data to come up with better defaults. But for specific computers, you'll get much better, and smoother gameplay if you adjust this manually.

It will tone down the amount of stuff flying off, but it will perform a lot better.

[LuckLock]

Thanks for the amazing support! I apologize, but I've never sent a feedback email. Do I just put it in the comment box after running a performance test, or do I put it in the "Your email" box? I have already submitted a performance test containing "Atten : C7 Feedback Email: Atten : C7" in the comment box, but if this is not the correct way to do it, I will do it again, but correctly. Again, thank you for your support.

[C7]

We might need to make that more clear. The feedback has a place to put your email address, so we can respond to you directly. It will automatically go to the correct place when you send it though.

No worries though, I got the message fine.

It looks like you're running into some GPU bottleneck issues, two things that should help.

1. Turn down the anti-aliasing a bit, try it off or at something like 2x.

2. Since your gpu is getting tied up, it might be good to use your processor as the computation device for physics instead. I don't see your cpu listed as an OpenCL device, even though it's capable of running it, and would be quite good at it.

You can install AMD's APP (Accelerated Processing Platform), use the latest stable version, which is 2.9.1

http://developer.amd.com/tools-and-sdks/opencl-zone/amd-accelerated-parallel-processing-app-sdk/

This should add support for your CPU to work as an OpenCL device.

Hopefully this will help.

[LuckLock]

Installing AMD APP fixed all performance issues for me. The game runs almost flawlessly with all settings on High. My performance score has gone from around 15 to over 30. Thank you for the support.

[C7]

Awesome, I'm glad to hear that. Thanks for giving me all the information I needed to help!

[RobertGimRo]

Same here, the collisions are Hella laggy, now one collision  equals lag of 9 at once.....

Strange i Dont Get Any Lag With The Collsions atm And iam very glad that i dont I Can Finally Enjoy this game

[C7]

If you're getting collision lag, the best thing to do is turn down the number of max attracting, and max non-attracting bodies. It tends to be if you're over the body count your hardware can handle, it will start to chug. That's why some users run fine on the default settings, and others do not.

We're still working on automated performance tuning, so for now it's best to tweak it till you're happy.

If you want slower and crazy detailed, you can do that, or you can step down to lower counts if you want a more seamless experience.

[Crimson]

Galaxy collisions seems to have seamlessly improved (not the case for planetary collisions).

However, the game seems to constantly stutter (temporarily freezes or "hiccups") every few seconds regardless of what I'm doing. This has only been noticeable for version 13 and 13.1. Fps is much lower in general, even for simple simulations i.e. The Sun and the Earth.

[C7]

You can use the old integrator if you like. It's labeled as the middle option, "unsafe" if you check the integration modes. This is in the Simulation options, under the gravity heading.

Adjusting the tolerances may help as well. Since that effects when the simulation is slowing down to ensure things maintain accuracy.

There's also a possibility there's an error in your logs. Whenever Unity dumps out stack traces to the log it slows down really badly.

So, posting your logs would be useful, so we can potentially rule that out as a possibility.

[Crimson]

Thank you for the response, I've attached my log. When I decrease the tolerance as well as altering the time per second for Type 1A Supernova there's virtually no difference. The time displayed in the bottom corner freezes at ~2.44 seconds. Changing the tolerance rises to around 2.45 seconds but doesn't seem budge from there. It behaves in the same manner as pausing the simulation.

I'm testing out the unsafe integration mode. There is a noticeable increase of ~15 fps although it seems to constantly fluctuate regardless of IntegrationMode.

On a side note, I know I certainly don't have great specs, but there is a difference in performance since version 13.

Thanks for the help!

[The_Fool]

The AMD APP SDK had no effect on my performance score.  I went from around 130 down to 57 with alpha 13.  My computational device is my Intel i7 5930k with Hyperthreading disabled.

[C7]

Can you post your logs please The_Fool? We're collecting as much data as we can to see what users are having an issue. (The same goes for anyone else who's having these issues, log files and performance reports are really useful)

You can send in the performance reports by running the performance test and using the feedback window. You can add Atten : C7 Slowdown to the comments section, so I can find them easily.

So far initial data is showing that GPUs take a big performance hit with the new integration mode. So if you're using Global Adaptive, select your processor as a computation device.

If you're using a GPU try using the Unsafe Global Adaptive.

I checked your logs Crimson and it is in fact using your GPU, can you test with your cpu to see if that helps?

[Crimson]

Running simulations using my cpu on global settings notably increases fps by quite a bit, although it's still overall choppy and constantly stutters.

Type 1a Supernova still freezes around the same time, never exploding.

I've attached another log running simulations on my cpu.

Thanks.

[C7]

We've found the bug in the supernova simulation, and it could effect other situations. Seems it's the cause of a lot of subtle bugs.

We also found and identified the case of the stuttering, so we're working on a solution for that as well.

Turns out too many threads were getting spawned when the workload was low.

[Crimson]

Thank you for taking the time to address and constantly fix bugs. I hope my logs were somewhat valuable.

Out of curiosity, are logs more helpful by sending through the feedback system or providing them on the forums and does whatever choice of method affect which logs have priority over others?

[C7]

I check all of them, so it all gets to the right place, one way or another. Just use whatever is most convenient for you. 

Good reproduction steps, and data are the two key things that are needed for us to be able to resolve an issue quickly.

That's why we always ask for logs and crash dumps. One of them may contain a clue that can point us to the exact right place. If we can't reproduce as issue here, it's almost impossible to tell if it was fixed. Tends to slow things down a lot, since we have to wait for a release, hope for confirmation, and repeat as needed.