Estimated FPS gain

insider · Post by **insider** » Wed Nov 25, 2015 5:46 am

My game is around 11 million polygons now and my GTX 760 is giving me 15 FPS tops @ 1920 x 1080, I need to record the gameplay using FRAPS and require 60 FPS preferably or at least 35 FPS.
I am considering upgrading to a GTX 980 within a month, so will that solve the above issue.

Any advise would be great.

syedhs · Post by **syedhs** » Wed Nov 25, 2015 8:19 am

1) Upgrading to GTX980 may solve or not the problem. If it is solved then it is good..

But if it is not?

2) You have to first profile your game to see where are the bottlenecks.. eg 11 million polygon - is it a total of polygon in the level? When the FPS is 15, is the camera showing all the 11 mil polygon or just partial of it? CPU bound or GPU bound? Many questions.. which can be unique within the games. So estimated FPS gain is not accurate, because if you are CPU bound then upgrading the card may not have effect at all in speed.

insider · Post by **insider** » Wed Nov 25, 2015 12:41 pm

syedhs wrote:1) Upgrading to GTX980 may solve or not the problem. If it is solved then it is good.. But if it is not?

2) You have to first profile your game to see where are the bottlenecks.. eg 11 million polygon - is it a total of polygon in the level? When the FPS is 15, is the camera showing all the 11 mil polygon or just partial of it? CPU bound or GPU bound? Many questions.. which can be unique within the games. So estimated FPS gain is not accurate, because if you are CPU bound then upgrading the card may not have effect at all in speed.

Thanks for replying Syedhs,

The camera shows between 7-11 million polygons but 11 million is the max I have observed.
The CPU usage is around 20-25% and GPU usage goes upto 50%
Anyway just for the info the CPU is an i7 4770K @ 3.5Ghz and 8 Gigs of RAM.
I am running a dual monitor setup so my secondary 20 inch monitor might be adding an extra overhead on my GPU, but assuming I gain 20% by disabling it, it still may not reach the intended 60 FPS.

I am not upgrading for my current project alone, I have plans on buying a 4K monitor too, so is GTX 980 a worthy card and can it handle such situations, I know 11 million is on the high side, but GTX 760 : 15 FPS , GTX 980 : ?
I have all sorts of numbers of the net comparing 760 and 980 but can it handle such high end simulations.

And you can take it as GPU bound my main character is 5 million polygons in itself.

NotCamelCase · Post by **NotCamelCase** » Wed Nov 25, 2015 5:13 pm

insider wrote: ... my main character is 5 million polygons in itself.

I'm intrigued about the main character. Are you using LOD lightly?

insider · Post by **insider** » Wed Nov 25, 2015 7:05 pm

NotCamelCase wrote:
insider wrote: ... my main character is 5 million polygons in itself.
I'm intrigued about the main character.

Haha yes it will be in showcase soon, LOD is in place.
Did a few optimizations and now it is in the range of 4-10 million polygons and FPS is close to 20 max, still nowhere near to what I want and my HUD programming is still left.

So I am not sure if 980 is a go, I can only buy one high end card, 980, 980Ti or TITAN X perhaps, any clues?
If a 980 can do the work then there is no need for a higher end card, besides I change the GPU every 2 years and my 760 will be 2 years old in Feb.

Ofcourse if I buy it and it doesn't help, I'll be rather pissed.

al2950 · Post by **al2950** » Wed Nov 25, 2015 7:19 pm

Are you sure you are GPU limited!? How many subMeshes are you trying to draw per frame? What is the rough rendering pipeline? Buying a new GPU is a quick and easy fix if $$$ dont matter, but I am not sure it will make as much of a difference as you think/hope!

c6burns · Post by **c6burns** » Wed Nov 25, 2015 7:37 pm

I have a 970 and it would rip that amount of geometry up ... but then is the geometry even what's bottlenecking you now? I have doubts

I have limited experience on modern desktop hardware as 99% of my work is with mobile GPUs ... but I would say batch count is a fairly important determinant of frame rate, especially if we are talking pre 2.1 ogre and how it renders. With a really high batch count upgrading hardware will likely not bring about as big an improvement as you expect.

One thing you can do is grab a card from some store with a return policy (like Best Buy here in the USA) and test the card then take it back

Or prep a demo and ask a kindly soul here to test it for you

NotCamelCase · Post by **NotCamelCase** » Wed Nov 25, 2015 8:12 pm

I doubt you're GPU-bound as well. Especially if you're running on pre Ogre 2.1 version. I'd check the existing benchmarks for a known GPU-bound game with different cards to have a rough idea of FPS gain WRT the GPU.

Post by **dark_sylinc** » Wed Nov 25, 2015 8:17 pm

insider wrote:The camera shows between 7-11 million polygons but 11 million is the max I have observed.
The CPU usage is around 20-25% and GPU usage goes upto 50%
Anyway just for the info the CPU is an i7 4770K @ 3.5Ghz and 8 Gigs of RAM.

That usually means you're CPU bottlenecked and upgrading the GPU will barely affect the framerate.
You should use profilers and play by disabling things to see how the affect the framerate.

Polygon count barely tells me anything about performance. It's not the same a model with 11 million triangles and 33 million vertices, than a one with 11 million triangles and 8 million vertices (literally a bit more than 4x work in the vertex shader).
Shader complexity also plays an important role. It's not the same a diffuse map with Blinn-Phong and one light, than a diffuse,normal,specular,roughness map with a complex BRDF and 16 lights. Also how much area a triangle covers in the screen plays a big role.
See http://filmicgames.com/archives/534

And we haven't even touched the CPU part yet.

insider · Post by **insider** » Thu Nov 26, 2015 5:44 pm

Never mind solved reached around 36FPS, 9 million polygons max, 7200 batches and according to the profiler 40% is taken up by rendering, 25% + 14% by two of my functions which I ll be aiming at optimizing as much as possible, anyway the maximum is taken up by rendering and that gives me the excuse to upgrade to a 980 GTX.

@c6burns yes that idea had crossed my mind too

syedhs · Post by **syedhs** » Thu Nov 26, 2015 6:35 pm

One quick and easy way to increase FPS in the case of GPU bound is to move the time consuming code block into FrameListener::frameRenderingQueued. The function is called whenever Ogre has submitted everything to GPU and then, wait for the GPU to finish. GPU bound may mean time taken by GPU is rather large, so move the code in there mean you will save a lot (YMMV) of processing time.

TheOnlyJoey · Post by **TheOnlyJoey** » Fri Nov 27, 2015 2:22 pm

insider wrote:Never mind solved reached around 36FPS, 9 million polygons max, 7200 batches and according to the profiler 40% is taken up by rendering, 25% + 14% by two of my functions which I ll be aiming at optimizing as much as possible, anyway the maximum is taken up by rendering and that gives me the excuse to upgrade to a 980 GTX.

@c6burns yes that idea had crossed my mind too

You might want to look into smart batching, static geometry could be a solution for that if you have a lot of non-moving object.
7200 batches is way to much, Nvidia always suggests batching around 1000 batches max which reduces a lot of cpu/gpu swap (which is a big part of performance issues).

Also with such a huge scene, screenshots would be appreciated

Post by **dark_sylinc** » Fri Nov 27, 2015 11:09 pm

At 7200 batches with Ogre 1.x you're definitely CPU bounded.

Like TheOnlyJoey said, start batching (Static Geometry, New InstanceManager); or just move to Ogre 2.1

syedhs · Post by **syedhs** » Sat Nov 28, 2015 12:17 pm

Majority (or dare I say, most) games project are CPU-bounded..

Ogre Forums

Estimated FPS gain

Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain

Re: Estimated FPS gain