Viewer benchmark test results

I've spent half a day to run through some series of tests to benchmark the main viewers around. That's the current offical 1.21.6.0 viewer and the available 3rd party viewers that are based on the same code base:

Imprudence 1.1 RC1
KirstenLee's S16
Cool Viewer 1.21.6.0 R16

Results in summary...let's see how a table looks like here

Viewer..............................High (default)..........Ultra
=====================================
Imprudence 1.1 RC1...........32.6 FPS..............23.7 FPS
Official 1.21.6.0...................33.7 FPS..............24.1 FPS
KirstenLee's S16.................34.3 FPS..............24.9 FPS
Cool Viewer 1.21.6.0 R16.....35 FPS................26.4 FPS

The field is pretty close and likely no differences are noticeable in everyday usage.

As I can't post pictures here you are invited to read the details on my blog. Link below

Did you average the frame rates in different sims, or chose 1 sim in particular?

I usually chose Abbotts Airport to benchmark- lots of stuff to rez.

I used the same sim, the B@r Rose store in Black Cat. It has a number of textures but not much/stable traffic making the samples comparable. I did not move, let it full rezz for a couple of minutes and the frame rate stabilize. Each test was at least repeated twice to confirm. So I think the results are valid, however they will likely differ when moving around, loading textures, rezzing etc. making it more of a gamble.

What I think is interesting is how close the results are. All in a range of about 5%. I dont think someone can really notice a difference. But this test doesn't tell anything about how fast textures load.

Was there any difference in the number of short freeze-ups between the viewers?
I always get those, even with good FPS.

@Conifer

Not quite sure what you are talking about. I only know about freeze ups from Mac, the tests were all done using Windows viewers.

There is a not really reproducible effect that sometimes a viewer runs slower at the first start. Closing and restarting usually fixes that. Also when changing the graphic settings sometimes there is a slowdown, sometimes not. One reason I have done the tests several times.

How about installing Ubuntu Linux and trying a viewer on that.

I've been told SL can run faster under this.....

Be interesting to see how it compairs to your current benchmark results

Pointless to do that as this is a complete different environment. You would compare apples with oranges. Perhaps someone wants to run the same tests for all viewers under Linux?

I assume it will run slower under Ubunbtu. AFAIK Ubuntu still uses i386 instead of i686, meaning it would run unoptimized and slower. Secondly using OpenJPEG will show the same result as Imprudence, lower performance. This all may be compensated by other factors...who knows?

I'm not going into a religious debate about Linux, I use it in a VM when I need it. For me it's still not ready for prime time desktop usage for several reasons. But thats just me

Short freeze ups? o.O

That only happened when my 1.6ghz dual core laptop had 1gig of ram.

Ubuntu is my primary OS and SL runs slower in it than on my Vista install

Some say it is an ATI issue. I wish it were not so.

Kinda disagre with that.

I know what you are saying but the PC is the fixed point, not Windoze.

If you can install anything onto a PC and make SL run faster than at the end of the day we have SL running faster on a PC

Whether that be Linux, XP, Vista or DOS

The OS it only a bit of software running in the pc to run the progam.

But, that aside it's strange how different peoples experiences vary.

I have heard it said that Quad core CPU makes a diference, and others say it does not.
I have heard than SLI graphics make a difference and other say it does not
I have even heard it runs faster under Linus, and some say it does not.

You would not think it wouldbe so dam hard to get a straight answer.

I think we need a standalone Sim/Homespace, set up wih areas of various detail/objects.
then (not that I guess it's possible) something written than will run a benchmark and fly your avatar (or viewpoint) around this sim (that never changes) and gives you a full rating at the end of it.

then we'd all have something to work with.

Like the old DOOM and QUAKE benchmarks built into the software.

I'm running SL on either openSUSE 11.1 or vista home premium on both of my computers, and on a general note i can say that it is noticeably faster on linux. on both of them. both have nvidia graphics and enough ram for SL...
yet, on vista SL crawls to a stop more often than i'd care for, bundled with "not responding". on average every 5-10 minutes.

@Piggie

You forget that the viewer is only part of the overall software. You would not only benchmark this but always the combination of all parts viewer+OS, given the same hardware platform. And such results would likely differ between for instance Ubuntu which is targeted at simplicity for the user and compatibility (bloated and slow) and a sleek 64bit Arch Linux. Just to name two. Also results between XP and Vista, and their 32 and 64 bit versions, will be different.

Overall I don't see any benefit in doing such a test unless you test various viewers on the same hardware and software platform and by doing so eliminate all variables other than the viewers itself.

If you're getting those frame rates in Bare Rose, you've got a pretty nice system. If it only took a few minutes for the textures to stabilize, you've got a pretty good Internet connection.

I suspect any differences would be more noticeable on older hardware.

Edit: people going for performance are likely to be interested in the 19.x and 18.x based viewers as well.

Hehe..no, that was not the main store but a branch which is pretty much a secret tip to hunt for some great freebies *hint*. In the main store it would probably need half a day for one viewer alone .

The problem with older viewers is that there is no real common basis to test with different versions. All handle settings a bit different and have partly different features, besides older LL viewers don't run anymore to compare against. I've done a test for the 1.22 RC tho, which was about 10% slower. Perhaps I will try with Nicholaz 1.18.0.6 and CV 1.19.0.5, should at least give some indication.

One thing I'm thinking about is if these tests are really usefull. The rendering itself seems pretty much on par between all four. And it was basically a static test not inculding any of the issues we all see, to name only slow loading of textures. But I don't really know how to do a dynamic test similar to a game benchmark by running a kind of pre-programmed sequence. Such results would be more valuable. If anyone has an idea?

Script a prim to move you over a fixed path, and snapshot the frame rates at specific points on the path?

Neat Results, Thanks

Not really that surprising, considering the variety of hardware and software people try to run SL on. People are quick to assume their own experiences are universal truths.

I'll bet a quad core CPU in a system with a bottom feeder video card makes a lot less difference than one in a system with a good video card. The same with SLI in a system that's CPU bound.

So you're probably getting straight answers, within the limits of the experience of the people giving you the answers.

What about Linden 1.22.9?

And I just went back to the LL viewers because I now find the most recent RC to the fastest and most stable LL viewer in a very very long time.

I'm curious about Boy's system specifications when he conducted his tests, as I'm not sure if mine are comparable or not; but here's some preliminary results I took:

About Second Life information:


Graphics setting: High (default), AA and AF off:
Average FPS: 33.2

Graphics setting: Ultra (default), AA x4, AF on:
Average FPS: 21.4

As an aside, there were 15 avatars present at the time, according to my radar, and two of them simply would not leave my field of view (I didn't ask, but just wanted to note this). May try this again later tonight when traffic is lighter.

Just an aside, but Boy is a female.

More details including this is on my blog .

Faster likely, but more stable?

Makes perfect sense as I written before. This test is more or less static with all and everything rezzed. And all 4 test candidates showed pretty much the same results. That is for the rendering engine only. It doesn't test anything in regard of texture loading or dynamic behaviour in general. Would be a shame if 1.22 would be worse

Okay, I looked. We may well have comparable systems after all. The only notable difference I saw was that you're running a single-core system at about 150% of the clock speed of my dual-core system. I don't know how much of a difference that will make in rendering speeds; but, if our systems are comparable, it appears that SL on Linux might be a hair faster, but not enough for anyone to notice. We'll see what happens when I repeat my test later tonight.