• Login
  • Register
  • Dolphin Forums
  • Home
  • FAQ
  • Download
  • Wiki
  • Code


Dolphin, the GameCube and Wii emulator - Forums › Dolphin Emulator Discussion and Support › Support v
« Previous 1 ... 660 661 662 663 664 ... 1174 Next »

Low FPS but low GPU load with 4xSSAA; not CPU limited
View New Posts | View Today's Posts

Pages (4): 1 2 3 4 Next »
Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
Thread Modes
Low FPS but low GPU load with 4xSSAA; not CPU limited
10-21-2012, 09:32 PM
#1
DolphinRocks
Unregistered
 
On SSBB I can run LLE, DX9 4xIR, no SSAA at 59-60 fps with my specs at 31% GPU Load.
As soon i set 4xSSAA fps drop to ~40 at 58% GPU load.
Why isn't my GPU stressed more to get higher FPS? I can't be CPU limited from the first example, can I?
Running Dolphin x64 3.0-787

Specs:
i7-920 OC to 3.6GHz
ATI HD5870 1GB OC to 900/1300MHz
Reply
10-21-2012, 09:37 PM
#2
DefenderX Offline
The comedy never ends.
*******
Posts: 4,617
Threads: 186
Joined: Dec 2011
Your GPU is not strong enough to handle 4xSSAA + 4xIR. That's it.

The CPU is not the problem.
[Image: IEZtsj]
[Image: 8fhx2zum.gif]
Website Find
Reply
10-21-2012, 09:41 PM
#3
DolphinRocks
Unregistered
 
I thought that was the case but why is it being underutilised? Shouldn't it be at 100% load for it to struggle?
It can handle up to 2.5xIR and 4xSSAA, I just checked. This is better than 4xIR without SSAA, right?
Reply
10-22-2012, 06:58 AM
#4
AnyOldName3 Offline
First Random post over 9000
*******
Posts: 3,505
Threads: 1
Joined: Feb 2012
It's always better to add IR than SSAA, because they're basically implemented the same, but SSAA gets scaled down more times than IR, so loses more quality.

Anyway, sometimes GPUs don't show full load, even when they're the bottleneck. This is because you aren't necessarily using all the GPU's resources (for example the Graphics RAM isn't in high demand by dolphin), but part of the actual chip is all in use.
OS: Windows 10 64 bit Professional
CPU: AMD Ryzen 5900X
RAM: 16GB
GPU: Radeon Vega 56
Find
Reply
10-22-2012, 10:06 AM
#5
lamedude Offline
Senior Member
****
Posts: 360
Threads: 7
Joined: Jan 2011
My guess is at that resolution that 32 ROPs can't keep up with the 80 TMUs and/or 1600 shaders.
Website Find
Reply
10-23-2012, 11:02 AM (This post was last modified: 10-23-2012, 11:02 AM by NaturalViolence.)
#6
NaturalViolence Offline
It's not that I hate people, I just hate stupid people
*******
Posts: 9,008
Threads: 24
Joined: Oct 2009
Quote:It's always better to add IR than SSAA, because they're basically implemented the same, but SSAA gets scaled down more times than IR, so loses more quality.

wtf am I reading. No it doesn't.

Quote:Anyway, sometimes GPUs don't show full load, even when they're the bottleneck. This is because you aren't necessarily using all the GPU's resources (for example the Graphics RAM isn't in high demand by dolphin), but part of the actual chip is all in use.

Video ram capacity and bandwidth aren't measured at all when the GPU load is calculated.

Quote:My guess is at that resolution that 32 ROPs can't keep up with the 80 TMUs and/or 1600 shaders.

It's a possibility. If I recall shader throughput is the only thing that is measured by GPUZ (which is what I assume he's using).

It's more likely that he's running out of memory bandwidth. There is an easy way to test that. He can change his video memory clock rates and retest for changes in performance.

Quote:I thought that was the case but why is it being underutilised? Shouldn't it be at 100% load for it to struggle?
It can handle up to 2.5xIR and 4xSSAA, I just checked. This is better than 4xIR without SSAA, right?

2.5xIR + 4xSSAA = (2.5^2) x 4 = 25x
4xIR + no SSAA = (4^2) x 1 = 16x

4x IR + 4xSSAA = 64 times the native resolution. That's a crazy high internal resolution of about 21 megapixels per frame.
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."  
-Ron Swanson

"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony
Website Find
Reply
10-23-2012, 11:05 AM
#7
Axxer Offline
Why would I help you?
*******
Posts: 1,078
Threads: 28
Joined: May 2012
How would I go about testing where my bottleneck is in the actual GPU (such as whether it is memory badwidth or GPU clock or shader clock)?
Desktop:
OS: Windows 7 64 bit
CPU: Intel Core i5 3570K @ 4.4 GHz
GPU: EVGA NVIDIA GeForce GTX 550 Ti (Factory OC'ed)
RAM: 16 GB @ 1600 MHz
Dolphin Revision:3.0, 3.0-784, 3.5, latest 3.5
Find
Reply
10-23-2012, 11:24 AM (This post was last modified: 10-23-2012, 11:26 AM by NaturalViolence.)
#8
NaturalViolence Offline
It's not that I hate people, I just hate stupid people
*******
Posts: 9,008
Threads: 24
Joined: Oct 2009
Quote: How would I go about testing where my bottleneck is in the actual GPU (such as whether it is memory badwidth or GPU clock or shader clock)?

1. Pick a clock rate
2. Adjust it
3. Benchmark the scenario
4. Goto Line 1

Or alternatively adjust settings and benchmark (although this is tricky to do in a reliable way since many settings affect more than one part of the GPU and/or 3D graphics pipeline).
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."  
-Ron Swanson

"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony
Website Find
Reply
10-23-2012, 11:53 AM
#9
Axxer Offline
Why would I help you?
*******
Posts: 1,078
Threads: 28
Joined: May 2012
Ok. Basically I should change things and test to see if it makes a speed difference, and if it doesn't move onto something else?
Desktop:
OS: Windows 7 64 bit
CPU: Intel Core i5 3570K @ 4.4 GHz
GPU: EVGA NVIDIA GeForce GTX 550 Ti (Factory OC'ed)
RAM: 16 GB @ 1600 MHz
Dolphin Revision:3.0, 3.0-784, 3.5, latest 3.5
Find
Reply
10-23-2012, 03:21 PM
#10
lamedude Offline
Senior Member
****
Posts: 360
Threads: 7
Joined: Jan 2011
Bandwidth looks like the culprit.
5870 can't even reach a 1/3 of its peak fillrate in a synthetic test.
Website Find
Reply
« Next Oldest | Next Newest »
Pages (4): 1 2 3 4 Next »


  • View a Printable Version
  • Subscribe to this thread
Forum Jump:


Users browsing this thread: 1 Guest(s)



Powered By MyBB | Theme by Fragma

Linear Mode
Threaded Mode