Performance Comparison (FPS) Intel i3 + i5 / AMD Ryzen 5 + 7 [stock + OCed]

[Drayke Newall]

1 hour ago, Wulfie Reanimator said:

Even if I grant you that grey textures are somehow bad for FPS on their own (I don't agree), latency does not affect how fast something is downloaded in the overall scale.

You can have 10000+ ping and still download things at gigabytes per second. (Theoretically speaking. It's an extreme example to bring the point across.)

Latency is a measure of time it takes for one packet (or a round-trip) to travel from A to B (or A->B->A). When you're downloading contiguous data like a texture file, latency won't affect the download speed after the first packet has arrived, as the rest will follow just as quickly regardless of distance, because they were already on the way right behind the first packet. (Similarly, the viewer won't wait for one texture to finish downloading before the next -- the viewer is receiving multiple file downloads at once. No delay between each texture.)

Latency is only important for things like communicating inputs. If it takes 1 second after pressing W before your avatar starts moving... that's not pleasant, but you wouldn't notice the ping just from looking at how fast textures are loading.

The act of streaming itself doesn't cause FPS issues. It's the CPU/HDD time spent on decoding a finished download. Let's say that hypothetically you start downloading all the textures on a sim at the same time (ignoring viewer restrictions), but due to a disconnect or you throttling your internet speed to like 1Kb/s, you're not going to experience any lag from textures. Why? Because your computer can and only render what's already on your computer, and rendering no textures is super easy. I guess this is a bad example because you think that the streaming itself causes problems and disconnect/1kbps means there's no streaming. But just go and open some debug consoles in your viewer and watch where the slowdowns happen. It's not proportional to current network activity.

No textures ever finish downloading = No lag from decoding them into usable data = No texture lag.

And then you get disconnected from SL because 1Kb/s is not enough to sustain you.

You seem to think I am suggesting that it is the downloading speed that effects fps or that streaming does. I don't think this. What I am saying is that when the world is grey and has no textures populating it you have a higher fps due to no textures being present and when textures start to load there is a subsequent drop in fps (minor) due to those textures now populating the screen.

It is the same as what Nalates said in that fps can in some cases go up due to no textures being loaded and once loaded/loading the fps reduces. I never said grey textures were bad for fps just that they can affect it, granted marginally and only in the first instance of scene loading.

I don't argue either that once the packet has been sent that you wont receive the rest at a reduced rate of speed. Distance does however matter as far as how the initial download speed is determined (ping is the test of how long it takes for something to be transferred over a distance aka latency, the further you are away), as the further you are away from the source the slower your download speed is https://blog.httpwatch.com/2008/08/14/the-surprising-effect-of-distance-on-download-speed/. I was using latency to describe the distance which whilst theoretically it isn't, the higher it is the further you are away. It is why here in Australia if we download something from the USA our download speed is reduced based on many factors, not due to the actual speed of the download itself but due to the fact it isn't a single round trip and also because of TCP issues, software restrictions etc. as mentioned in the link before mentioned.

All in all what I am say is, that the longer it takes for your internet to download the textures the more time you will see a slight increase (in some cases) in fps due to SL being grey and no textures being present. In the case of benchmarking done by the OP, to compensate or to remove any of this effect and solely focus on the cpu rendering a specific scene, they should have already had those textures within their cache so that it is loading directly from his system and not any potential effect caused by latency/download/distance (what ever you want to call it). It is also why I asked him originally if he cleared his cache before hand to determine whether there were any other outside effects contributing to fps fluctuations.

Edited December 2, 2019 by Drayke Newall

[Wulfie Reanimator]

47 minutes ago, Drayke Newall said:

Distance also does matter as far as download speeds go (ping is the test of how long it takes for something to be transferred over a distance aka latency, the further you are away), as the further you are away from the source the slower your download speed is https://blog.httpwatch.com/2008/08/14/the-surprising-effect-of-distance-on-download-speed/. I was using latency to describe the distance which whilst theoretically it isn't, the higher it is the further you are away. It is why here in Australia if we download something from the USA our download speed is reduced based on many factors, not due to the actual speed of the download itself but due to the fact it isn't a single round trip and also because of TCP issues, software restrictions etc. as mentioned in the link before mentioned.

Yes, but this only makes sense for TCP because TCP is very pedantic about making sure you received the previous packet before sending in the next one. That article specifically says:

Quote

[...] with standard TCP connections the round trip time between client and server (i.e. ping time) imposes an upper limited on maximum throughput.

This upper limit is caused by the TCP flow control protocol, It requires each block of data, know as the TCP window, to be acknowledged by the receiver. The sender will not start transmitting the next block data until it receives the acknowledgement from the receiver for the previous block.

SL uses UDP for most things, including texture streaming. UDP is one-way, ideal for streaming content (video, files, etc...), because it does not care whether the previous packets were received before the next ones. The viewer handles and re-requests missing data and stitches them in the correct order.

http://wiki.secondlife.com/wiki/UDP

http://wiki.secondlife.com/wiki/Transfer_Manager

http://wiki.secondlife.com/wiki/Image_Pipeline

http://wiki.secondlife.com/wiki/Texture_Console

Edited December 2, 2019 by Wulfie Reanimator

[randakong]

@OP, that's a joke on i9 CPU, the clubs barely 100 people at the time you recording. You barely maintain 30 fps. Why? Your Ms:160 is the ones that lower your FPS. If you got ms: 0 to 2 from your ISP. It makes a huge differences. It just like this. Someone is getting 250ms from other sides of the world, they experience lags and FPS as well.

[Wulfie Reanimator]

10 hours ago, randakong said:

@OP, that's a joke on i9 CPU, the clubs barely 100 people at the time you recording. You barely maintain 30 fps. Why? Your Ms:160 is the ones that lower your FPS. If you got ms: 0 to 2 from your ISP. It makes a huge differences. It just like this. Someone is getting 250ms from other sides of the world, they experience lags and FPS as well.

You're 3 years late.

[Jackson Redstar]

can't really benchmark SL anyways, SL is like a woman - sometimes she is in a good mood and everything goes smooth and easy, next miniute, bad mood and no matter what you do it's all horrible! 😂

[randakong]

Do u guys know why I said that? Because I had FX 6300 stocked 16GB ram dual and GTX 1650 super got way more FPS than OP. That's the only reason I could think of. My ms:2 which aka pings from my ISP. I had 30mbps download connection.

[Kavarek]

On 11/24/2019 at 12:05 PM, Bas Curtiz said:

My goal was to find out which CPU is better for SL (without paying a 500 bucks worthy CPU like the i9-9900k)
I'm aware SL doesn't use multi-cores, but single-core: so in theory it benefits from the highest clocked frequency.

I finished my CPU performance comparison FPS-wise in SL with same hardware/settings/viewpoints/avs/etc. for:

Ryzen 5 3600 vs. Ryzen 7 1700x [stock + overclocked]
check it out here: https://youtu.be/VDpN-w-0sts

Intel i3-9350kf vs. i5-9600k [stock + overclocked]
check it out here: https://youtu.be/my-J89BM6Nk

Conclusion below the vids.

I will give test comparisons  on similar systems.

On the I9 10 core + GTX 2080 Super, with out shadows, advanced on draw at 500m. 4k resolution it would handle 120-140 fps with moderate  detailed area and perhaps 16-20 avatars.  Shadows on-smoothed 50-60 but in heavier loads 35-40 fps.

AMD system Ryzen 5 6 core, Rx 6700, it couldn't handle it near as close.  with advanced on 50 fps shadows on 28-30.

AMD Ryzen 7 12 core and RX 6800. Much better but still average 10-20 fps less the the intel/nvidia system with a lower GPU rating. 

I can only say i get better performace on Intel + Nvidia gpus, except for a bad driver version 3 versions ago that really caused issues with firestorm. 

 

I9 aren't going to be very used. it will balance load very well, but hardly be touched compared to other CPU's, but you are talking a big price difference for something being used like second life. 

Second life simply is not optimized, and uses old OPENGL factors. If the viewers and core are rewritten to utilize newer opengl functions, everyone could see a big improvement across the board. Better yet a direct X version but i don't see that happening. 

 

 

Edited February 27, 2023 by Kavarek