What determines the FPS, from my experience

[JessieHoo]

Hello, I have recently had some hardware changes, so I gather some first-hand experience on what is most important for the FPS of SL. I would love to share them. 

This may be long so let me first sum up my conclusion:

1. SL performance in high to ultra graphic settings depends primarily on CPU, SSD and much much less on (even the best) GPU. Spend most of your budget on a good CPU and SSD if you want good SL experience. 

2. SL may be the most taxing 3D application (not sure if SL is a game). Even with my latest PC it is impossible to push FPS above 60 on ultra settings at 1920*1080 resolution.

I switch between two graphic presets: Normal and Photography. As the name suggests, Normal preset is for normal play and Photography is only used for taking photos. The major difference is that normal does not have shadows on while Photography has all shadows. As shadows is a huge performance hit even only Sun/Moon is turned on. 

Normal:

Photography:

I always use Firestorm Release 55786. I play windowed. Resolution is 1920*1071, AntiAliasing is at 2X. 

I initially played with the following PC:

CPU: i7-5820k , first OCed to 4.0 ghz, then to 4.4 ghz; SSD: Samsung 960 EVO 1 TB; Graphic card: first GTX 980Ti then upgraded to RTX 2080 foundation edition; Memory: 3*8G DDR 4 at 2133. 

As you can see, my initial PC is already at the high end. The 960 EVO SSD is already one of the performance level SSD. I always make sure that both the game and my cache is set on the SSD of course. Yet at the very first, when my CPU was at 4 ghz with GTX 980 Ti, my FPS in the game is very low: even at the "Normal" setting, in my home my FPS stays at about 10-30. It is similar in other regions with lots of detailed mesh objects or people. If I use the Photography setting, my FPS will drop to below 10, which is hardly playable (it requires at least 24 FPS for human eyes to believe that you are animating in the world instead of playing slides)

My first change is to OC the CPU to 4.4 ghz. Performance almost stayed the same with no visible FPS gains. Then I upgraded from GTX 980 Ti to RTX 2080. I had about 10% increased FPS, which is about 1 to 2 FPS in my home. So, upgrading your graphic card alone cannot get you a lot of performance boost, even if the GPU is a very modern one. GPU usage is at about 10%.

I need to do simulation work in my field of study that requires a lot of CPU power, so I decided to buy a new computer. It certainly can help the SL too. So I have had a chance to run SL on my new PC.  It is a very modern one:

CPU: i9-9900k at 5.0 ghz (single core) or 4.7 ghz (all 8 cores); SSD: Samsung 970 Pro 1 TB; Graphic card: RTX 2080 foundation edition; Memory: 4*8G DDR 4 at 3000. 

And this time, I have a huge FPS increase. At the same place, my FPS is about 2 to 4 times as the original ones. I now have constant 50-60 FPS. If I uncheck limit FPS to 60 I can have 50-80 FPS, all the the same place with the same graphic settings. Even with Photographic settings, I can still have 15-50 FPS. GPU usage now is about 25%. Clearly GPU is being more utilized since the bottleneck on CPU abated. 

So from this, the conclusion seems to be that on high to ultra graphic settings, the FPS of SL depends mostly on CPU and a lot less on GPU. If you want to have a good performance, invest on latest Intel CPU with high frequency. GPU is less important. SSD should be important as well but I do not possess enough information on that aspect, (Samsung 970 pro is not significantly better than 960 EVO), but it is very likely that you will need a high performance SSD.

It is very surprising to see that i9-9900k at 4.7 ghz can be 3 to 5 times faster than i7-5820k at 4.4 ghz in SL. 

[animats]

Quote

It is very surprising to see that i9-9900k at 4.7 ghz can be 3 to 5 times faster than i7-5820k at 4.4 ghz in SL. 

That is interesting. Here's a comparison of those CPUs.  They're not more than 20% apart in anything. They have about the same size L3 cache, even. Finding out why that's happening is worthwhile.

A good test in SL is to visit Tralala's Diner. It's a very cluttered, well-done scene. Go to the entrance to the outdoor market, which is near the collapsed freeway, face the market, and wait about two minutes while everything loads. Then report your numbers with different settings.

SL is usually limited by single-thread CPU speed. The usual situation is that the main thread is at 100% utilization. A few other threads can use maybe 25% of a second core fetching and decoding assets. The GPU is seldom the bottleneck, even with a $80 NVidia board.

Edited November 10, 2018 by animats

[JessieHoo]

Thanks for the reply. I went to Tralala's Diner and stood here as the attached picture indicated for several minutes. When I use my "normal" setting, I was getting 30-35 FPS. When I use "photography" setting (which as you can see is the ultra quality plus a few further tweaks), I was getting 15-25 FPS. Graphic card usage was around 40% (an incredibly high figure considering I am using RTX 2080). I did not test using my old CPU since I have already changed my PC.

Another interesting finding is that overclocking the CPU has negative performance impact on SL. When I overclocked my i9-9900k to 5.3 ghz in single core and 5.0 ghz for all cores (which is a 0.3 ghz increase of CPU frequency), my FPS actually reduced a bit (around 3-5 FPS). I am guessing that maybe SL is also sensitive to memory clock. And when I overclock, I may accidentally changed the clock of the memory (I am using automatic overclocking feature of my motherboard which sets everything for me), but it is not convincing even for myself.

[cykarushb]

Weird to note, tried that place, similar view, every single slider to the max and every box ticked in (yes i know this is dumb:) 1280x1024

~15-17fps after everything loaded

And ive got a drastically lower tier system than you, with an i5 4570, GTX 970 and 12gb of 1866mhz ddr3 running windows 7 enterprise, using the regular LL veiwer

1280x1024 is kind of weird and i know, i use this monitor because it runs at 120hz and i like the 5:4 aspect ratio more

can you grab a screenshot in 1280x1024?, id like to see what the comparison would be if we were in the same resolution

 

 

[Wulfie Reanimator]

Not shown in the picture: Anisotropic Filtering, OpenGL Vertex Buffer, 2x AA, 1024MB Texture Memory Buffer
FPS fluctuates around 24-29 FPS. Enabling Advanced Lighting with Sun/Moon + Projectors (shadow quality 1.5) drops to around 16 FPS.

If I set my settings exactly as shown in the OP, my FPS is around 29-32 FPS.

i5-3570k (3.4GHz, 4 cores/threads), 16GB RAM, GTX 1060 (3GB), maximized window on a 1920x1080 monitor (but my desktop is 3840x1080).

Changing the window resolution to 1280x1024 made the FPS counter occasionally reach 33 FPS.
At 1280x1024 window size and matched settings, GPU load is at 31-42% and overall CPU load is at 29-32% (core 4 is at 45-60%, others dip between 14-30%)

Edited November 12, 2018 by Wulfie Reanimator

[cykarushb]

Thanks for posting a 1280x1024 to see a comparison. I think ive got something wrong going on then if im seeing those low fps numbers on 1280x1024.

I went and busted out the 1080p monitor and installed firestorm to get a reading based off of OP's settings and resolution as well.

1080p, same settings as the top:

I got this screenshot at a bad time, its averaging more like 55 fps, note the little spot of sub 20fps when i was not in that tab so it averaged to 42.

These images are big so its gotta be multiple posts (most xenforos are like 20mb per post, whats up LL?)

 

 

 

 

[cykarushb]

Max settings, 1080p, 4.0 shadows and 16x aa and all that:

Roughly 8fps tops, expected

I also tried bringing the resolution back down on my other monitor, same max settings, 12-14fps in 1280x1024. Which is a weird change because most games im looking at almost double the framerate.

Edited November 12, 2018 by cykarushb

[JessieHoo]

48 minutes ago, Wulfie Reanimator said:

If I set my settings exactly as shown in the OP, my FPS is around 29-32 FPS

Thanks for showing this, but I think in my settings, the slider of "performance" is between "high" and "ultra" for my normal setting and is on "ultra" for my photography setting. While on yours, it is between "low" and "mid". I do not exactly know what does this slide does but it should have some impact. And, we were not on the exact same spot. I have a lightning source (the yellow bulb) in the screenshot as you can see. Having that while with LOD and/or shadows enabled can dramatically reduce FPS. As we know that the FPS of SL can change vehemently even if you move a tiny distance or angle. I will return to that place and test my FPS at the exact same spot shown in your picture.

[Wulfie Reanimator]

4 hours ago, JessieHoo said:

Thanks for showing this, but I think in my settings, the slider of "performance" is between "high" and "ultra" for my normal setting and is on "ultra" for my photography setting. While on yours, it is between "low" and "mid". I do not exactly know what does this slide does but it should have some impact. And, we were not on the exact same spot. I have a lightning source (the yellow bulb) in the screenshot as you can see. Having that while with LOD and/or shadows enabled can dramatically reduce FPS. As we know that the FPS of SL can change vehemently even if you move a tiny distance or angle. I will return to that place and test my FPS at the exact same spot shown in your picture.

The slider doesn't update its position when you adjust/increase your settings manually. The slider is just a preset selection, but now that I know you're using the presets, I'll edit my post later when I redo the test as well. 

What about the other setting tabs?

Edited November 12, 2018 by Wulfie Reanimator

[bigmoe Whitfield]

Shadows and high draw distance are frame rate killers.  11 years I've been in SL, with all kinds of hardware.

[Wulfie Reanimator]

1 hour ago, bigmoe Whitfield said:

Shadows and high draw distance are frame rate killers.  11 years I've been in SL, with all kinds of hardware.

OP is talking about the effects of hardware on performance though, not settings. At least in the OP.

[bigmoe Whitfield]

1 hour ago, Wulfie Reanimator said:

OP is talking about the effects of hardware on performance though, not settings. At least in the OP.

Shadows is a huge effect on hardware.   so it's very relevant.  they physical drag the hardware down enough with taking huge amounts of resources. 

[Wulfie Reanimator]

The way shadow calculations are done is the same for everybody*. Hardware is not the same for everybody. OP wants to know which components have the biggest impact on viewer performance.

If everybody is using the same settings - shadows or not - at the same place, and report their specs and where the load is, they can gauge what has the biggest impact on FPS.

The more factors we match, the more factors we can ignore. You're focusing on something that is not comparable between people. Me enabling shadows (or any setting) has a different impact on my FPS than anybody else doing the same.

Edit: Wew, failed to quote.

Edited November 12, 2018 by Wulfie Reanimator

[Love Zhaoying]

1 hour ago, Wulfie Reanimator said:

OP wants to know which components have the biggest impact on viewer performance.

What was the answer?

[Wulfie Reanimator]

12 minutes ago, Love Zhaoying said:

What was the answer?

I would say CPU, but at the same time I don't have ways to clearly and definitely prove it objectively. As I'm writing this and standing in the test location, my FPS is steady just below 30 FPS, but not a single CPU core is under pressure and my GPU even less so. Temperatures are low and there is barely any disk or network activity.

Even then, even if my CPU was being maxed out, that wouldn't mean that SL was super CPU heavy, because my CPU could be underpowered, malfunctioning, overheating, I could have the "power saving plan" enabled in Windows, etc. This is why you compare with others. ?

Edited November 12, 2018 by Wulfie Reanimator

[Love Zhaoying]

1 minute ago, Wulfie Reanimator said:

I would say CPU, but at the same time I don't have ways to clearly and definitely prove it objectively. As I'm writing this and standing in the test location, my FPS is steady just below 30 FPS, but not a single CPU core is under pressure and my GPU even less so. Temperatures are low and there is barely any disk or network activity.

Even then, even if my CPU was being maxed out, that wouldn't mean that SL was super CPU heavy, because my CPU could be underpowered, malfunctioning, overheating, etc. This is why you compare with others. ?

I’ve seen posts where “texture thrashing” (dealing with all the physical files) was a performance hit, to which I always reply: install SL on SSD. 

[Wulfie Reanimator]

Just now, Love Zhaoying said:

I’ve seen posts where “texture thrashing” (dealing with all the physical files) was a performance hit, to which I always reply: install SL on SSD. 

That would obviously be a performance killer because disk access is slower than any other part in your computer. That said, my SL cache is on an SSD. Every texture in my view is loaded (no discarded or ignored textures according to Firestorm) and I'm not moving my camera which would cause new textures to be downloaded or accessed. No other people are in my view either.