Firestorm vs the LL Viewer

[iceing Braveheart]

9 minutes ago, Solar Legion said:

Want to have a client that allows you to set the Texture Memory to some needlessly high value? Code it yourself.

Video tutorial please

[Ansariel Hiller]

6 minutes ago, iceing Braveheart said:

Video tutorial please

Sorry, the viewer and its development process is so crippled, you unfortunately have to use a Wiki to teach yourself. With your experience, this should be easy to do though: Open Source Portal

[Ansariel Hiller]

5 hours ago, iceing Braveheart said:

because firestorm is artificially restricted to the developers personal computer and catzip is not you can really crank up everything and max up all those settings without losing fps or breaking a sweat performance wise with newer and better hardware on catzip

Sorry I don't own a 20K Euros "consumer" workstation GPU with 16+ GB VRAM to fullfil your needs. But if you pay me 20K Euros, I will gladly get you a special build only for you - with superspecial awesome window title mentioning how 1337 the super advanced changes in your special build are!

[iceing Braveheart]

21 hours ago, Ansariel Hiller said:

Yes! Declaring a workstation GPU as consumer GPU! ?

 

On 11/10/2018 at 4:24 PM, Ansariel Hiller said:

16 GB VRAM... on a consumer GPU... in 2018... out of date soon... I almost spilled my drink! Let's check how much memory the current "mainstream consumer GPU" nvidia RTX 2800 TI for reasonably cheap 1300 Euros has: https://www.nvidia.com/en-us/geforce/graphics-cards/rtx-2080-ti/

11 GB - wow! The first 5 GB probably already vanished towards its successor... ?

Titan v $3,000 12GB vrm

https://www.nvidia.com/en-us/titan/titan-v/

nvidia gamer consumer grade drivers not enterprise workstation

https://www.nvidia.com/download/driverResults.aspx/139606/en-us

SUPPORTED PRODUCTS ADDITIONAL INFORMATION

NVIDIA TITAN Series: NVIDIA TITAN V, NVIDIA TITAN Xp, NVIDIA TITAN X (Pascal), GeForce GTX TITAN X, GeForce GTX TITAN, GeForce GTX TITAN Black, GeForce GTX TITAN Z

https://www.youtube.com/watch?v=AtwEx5Y1Y-Q

this is a workstation gpu 

https://www.nvidia.com/en-us/design-visualization/quadro/

this is a workstation driver

https://www.nvidia.com/download/driverResults.aspx/139017/en-us

SUPPORTED PRODUCTS ADDITIONAL INFORMATION

Quadro RTX Series: Quadro RTX 6000, Quadro RTX 5000 Quadro Series: Quadro GV100, Quadro GP100, Quadro P6000, Quadro P5200, Quadro P5000, Quadro P4000, Quadro P2000, Quadro P1000, Quadro P620, Quadro P600, Quadro P400, Quadro M6000 24GB, Quadro M6000, Quadro M5000, Quadro M4000, Quadro M2000, Quadro K6000, Quadro K5200, Quadro K5000, Quadro K4000, Quadro K4200, Quadro K2200, Quadro K2000, Quadro K2000D, Quadro K1200, Quadro K620, Quadro K600, Quadro K420, Quadro 410 Quadro Series (Notebooks): Quadro P5200, Quadro P5000, Quadro P4200, Quadro P4000, Quadro P3200, Quadro P3000, Quadro P2000, Quadro P1000, Quadro P600, Quadro P500, Quadro M2200, Quadro M1200, Quadro M620, Quadro M520, Quadro M5500, Quadro M5000M, Quadro M4000M, Quadro M3000M, Quadro M2000M, Quadro M1000M, Quadro M600M, Quadro M500M, Quadro K5100M, Quadro K5000M, Quadro K4100M, Quadro K4000M, Quadro K3100M, Quadro K2200M, Quadro K2100M, Quadro K3000M, Quadro K2000M, Quadro K1100M, Quadro K1000M, Quadro K620M, Quadro K610M, Quadro K510M, Quadro K500M Quadro Blade/Embedded Series : Quadro P5000, Quadro P3000, Quadro M5000 SE, Quadro M3000 SE, Quadro K3100M Quadro NVS Series: NVS 810, NVS 510 Quadro Plex Series: Quadro Plex S Series, Quadro Plex Model II NVS Series: NVS 810, NVS 510

nowhere does it mention titan's etc

it's not really fair to punish those with decent computers because you could not afford one in the first place

 

[animats]

2 minutes ago, Ansariel Hiller said:

Sorry, the viewer and its development process is so crippled, you unfortunately have to use a Wiki to teach yourself. With your experience, this should be easy to do though: Open Source Portal

I know you're being satirical. But building Firestorm yourself does not work unless you can fix the bugs in the self-compiled version yourself. Been there, done that, filed the JIRA.

[Ansariel Hiller]

2 minutes ago, iceing Braveheart said:

Titan v $3,000 12GB vrm

Cheap outdated crap! That's so 2016 and not future proof! Come up with something better!

[Ansariel Hiller]

Just now, animats said:

I know you're being satirical. But building Firestorm yourself does not work unless you can fix the bugs in the self-compiled version yourself. Been there, done that, filed the JIRA.

If you happen to use a proper operating system and not one consisting of a pile of incompatible distributions, it works as a charm. But when using "pro" OS like the not named one, one should assume its users know how to deal with such minor issues! ?

[Solar Legion]

Ok, time to stow the snark. Seriously.

[animats]

As I understand it, the problem with texture memory is that the viewers don't handle running out of it very well. They thrash, moving items in and out of texture memory. The viewer can downsize textures to make them fit, but this is inefficient.

I've looked at the texture system code. All the code is there to do all the hard cases. In parallel, even. That's well written, although under-commented. What's not there is a centralized strategy module to decide what to do first. That's why you end up standing in front of a blurry sign for two minutes waiting for the texture system to get around to loading the biggest thing that's right in front of you.

[Solar Legion]

As far as the Titan V is concerned ..... No, it is not a "consumer" product. It is a flashy product, targeted at those with more money than sense.

Again, if you want to have a Second Life client that uses more VRam for the Texture Buffer than it really ought to be using, code it yourself.

Don't know how?

Too bad.

The average user cannot afford to throw around all that much money and as irksome as it is (very irksome) many end up attempting to use the modern equivalent of a glorified Word Processor to run Second Life.

I'll type it one more time for you: Want to use more VRam than you really need to, for a Texture Buffer?

Build. The. Client. Yourself.

Yep, that's enough snark for the time being.

Edited November 12, 2018 by Solar Legion

[Love Zhaoying]

17 minutes ago, Solar Legion said:

I'll type it one more time for you: Want to use more VRam than you really need to, for a Texture Buffer?

Build. The. Client. Yourself.

I don’t put it beyond his capability, at this point.

[animats]

1 hour ago, Ansariel Hiller said:

If you happen to use a proper operating system and not one consisting of a pile of incompatible distributions, it works as a charm. But when using "pro" OS like the not named one, one should assume its users know how to deal with such minor issues! ?

The Linux parts work fine.  I'm just irked about one place where the LL code with a bug fix hasn't been integrated yet. Self-compilers have to work around that or the viewer crashes regularly. This will probably be cleaned up as animesh and EEP go in, since that forces a major resynch with the LL code.

Self-compiling should be encouraged. That's where new developers come from.

[Alyona Su]

3 hours ago, iceing Braveheart said:

I was looking for a better viewer and also tried the kokua you mentioned it also appears to be crippled by the 2GB limitation as well are any of the viewers that are modernized updated and or maintained?

https://ibb.co/nujfqA

You may have 50GB video card. There is a reason the Viewer sets only a small part of that for reserved. All your video memory is being used all the time. That setting is not a limit on how much video memory your video card is using to draw the SL world. But I'll not try to explain that here because I'm not a techy person, though it has been explained before and I have a basic understanding what that setting is for and it's not what most people think.

Edited November 12, 2018 by Alyona Su

[cykarushb]

1 minute ago, Alyona Su said:

You may have 50GB video card. There is a reason the Viewer sets only a small part of that for reserved. All your video memory is being used all the time. That setting is not a limit on how much video memory your video card is using to draw the SL world. But I'll not try to explain that here because I'm not a techy person, though it has been explained before and I have a basic understanding what that setting is for and it's not what most people think.

Sorry, no. Very little video memory is used by anything other than vram intensive games. There is a reason that the default VGA windows drivers run on 64mb of system memory, it doesnt need any more and wont use any more. The reason the viewers have video memory limits come when cards didnt really have that much vram, but SL would use it all, so you had the slider to make sure your card wasnt to have all its video memory eaten in SL by textures and not have enough for other things which would then start using system memory. Example, say its 2006 again, youre rocking a 320mb 8800GTS and when you play SL in a super detailed space, youre using all your video memory, so the youtube cat videos youre trying to play in the background, while they use very little vram, is now using system memory instead which is most likely 667mhz DDR2, so gg your video, its now stuttering because its essentially running off core2duo igpu since it cant get the little bit of vram it needs from your anemic 320mb 8800 GTS. God forbid you tried to run two viewers without a 512mb card.

THAT, is why that slider exists. Its not needed at all now, not in the slightest. Even that same generation the 8800GT had 1gb vram variants because GDDR3 got cheaper and cards started coming with more vram. SL's slider capped at whatever the high end was and wasnt really updated since.

It doesnt really need to have such a limit in 2018 since most people are running cards with at a minimum 1gb of video memory nowadays, even integrated intel HD graphics would use 1.5-2gb of memory. So the LL viewer and its 512mb limit is pretty low.

Now the question is, do you actually need SL to be able to use that much video memory exclusively for textures. Because outside of textures if i set that slider to the minimum, SL uses like less than 300mb of video memory for everything else. Max the slider it usually doesnt jump above 800mb total usage in most places, meaning im definitely maxing the video memory slider in the stock LL viewer. Moving over to firestorm however and unless i absolutely max my settings and LOD, and up the resolution scaling, ill maybe top over 1-1.2gb of video memory. WIth that 2gb of video memory for textures available i dont max that at all, barely touch it.

Now, video memory isnt the exclusive place textures are constantly held, theyre saved in system storage or cache, then depending on their priority they either get pushed into system memory for the GPU to grab them, or directly into video memory. Also despite some arguments on here, youre not loading an entire uncompressed texture into vram.

Despite SL's really large quantity of textures that can use a lot of video memory, youre just not going to ever touch 2gb of usage outside of extremely detailed spaces in very high resolutions at basically max settings. Which i dont think anything except for top tier workstation cards can handle anyway (or should i say, compensate for due to the CPU bottleneck...)

3 hours ago, iceing Braveheart said:

 

Titan v $3,000 12GB vrm

nvidia gamer consumer grade drivers not enterprise workstation

 

Titan V is whats called a toe, it "tests the waters" for a future line. The titan V was a workstation tier low binned quadro given gaming drivers and a premium price tag, to see if high end consumers would pay the extremely high price premium for the mediocre jump in performance. (this is also the ENTIRETY of the RTX lineup because the Titan V was a relative success)

Even if it is a gaming card, it is not a relevant card in the slightest considering it was thousands of dollars and almost nobody bought one, ironically the market that did buy them was pretty much the same market that buys Teslas. Nor is the highest of high end considered the average card for gaming, high end cards are a very small portion of the consumer GPU market, most people are running current or previous gen mid tier offerings, such as the 1050ti, 750ti, 650ti, 550/560ti, etc... And thats just the people who really bother to look into the GPU they buy, a large portion of people run integrated graphics, older APUs, weird oem cards like GTX 555's and 4gb GT 730's.

Check out the steam hardware survey for peoples PC's, you can see what the most common hardware is: https://store.steampowered.com/hwsurvey/videocard/

The big noticeable chunks, the first 4 cards before you get into obscured relativity by percentage margins are the 1060, 1050, 1050ti and 750ti. Meaning at most these people have 6gb of video memory. Many have 2 or 4gb, some 750ti's had 1gb. As it goes down there are the odd thrown in 1080ti's and 1070's, but then right along side them are 960's and 970's, mobile maxwell gpus, modern low end, significantly older cards such as these few *****s, an INCREASING NUMBER, who are still running 6600GT's for some unknown reason. May god have mercy on their framerate.

I used a 6600GT recently and it was terrible.

tl;dr, the high end is not the baseline, youre not gonna use that much vram anyway and really most people dont even have that much video memory, the average SL user probably isnt rocking 1080ti SLI, theyre more than likely on something like a 1050 or equivalent, maybe integrated graphics. Id like to see LL do something like what steam did here, since i know SL tracks your system specs with each hardware change, see what every user is running.

 

[Alyona Su]

23 minutes ago, cykarushb said:

Sorry, no.

Okay, you win. I already said I'm not a geek-nerd.

Oh, by the way: TL;DR is supposed to be at the top of the post, not the bottom, Because of it kind of defeats the purpose way down there.

Edited November 12, 2018 by Alyona Su

[Ansariel Hiller]

Okay, @iceing Braveheart convinced me:

[cykarushb]

1 hour ago, Alyona Su said:

Okay, you win. I already said I'm not a geek-nerd.

Oh, by the way: TL;DR is supposed to be at the top of the post, not the bottom, Because of it kind of defeats the purpose way down there.

the idea of "too long, didnt read" is meant to catch people with an end synopsis because they scroll past the wall of text

however on here, i just post that as a general summary

[Qie Niangao]

11 hours ago, cykarushb said:

Now the question is, do you actually need SL to be able to use that much video memory exclusively for textures. Because outside of textures if i set that slider to the minimum, SL uses like less than 300mb of video memory for everything else. Max the slider it usually doesnt jump above 800mb total usage in most places, meaning im definitely maxing the video memory slider in the stock LL viewer. Moving over to firestorm however and unless i absolutely max my settings and LOD, and up the resolution scaling, ill maybe top over 1-1.2gb of video memory. WIth that 2gb of video memory for textures available i dont max that at all, barely touch it.

This is all pretty fascinating. This part in particular made me wonder: Why haven't I taken these measurements myself? How would I go about measuring them? Is there a commonly used tool for observing what's using VRAM? If so, is it generally useful for performance tuning, or for anything else besides my unwholesome curiousity about how stuff works?

11 hours ago, cykarushb said:

Now, video memory isnt the exclusive place textures are constantly held, theyre saved in system storage or cache, then depending on their priority they either get pushed into system memory for the GPU to grab them, or directly into video memory. Also despite some arguments on here, youre not loading an entire uncompressed texture into vram.

Despite SL's really large quantity of textures that can use a lot of video memory, youre just not going to ever touch 2gb of usage outside of extremely detailed spaces in very high resolutions at basically max settings. Which i dont think anything except for top tier workstation cards can handle anyway (or should i say, compensate for due to the CPU bottleneck...)

So apparently textures loaded in VRAM remain compressed. That's not intuitive to me. So... every time a texture needs to be referenced, it has to be decompressed -- and it's not worth caching the decompressed bits anywhere for future reference? In this space / time tradeoff, then, space always wins? Or I'm not understanding something.

I'm also surprised SL needs so little texture memory, whether VRAM or system memory. The viewer's (compressed) texture cache -- a separate thing, but still -- it too is smaller than I'd have expected for what seems to take forever to download. Why did I have to wait for ages to download a measly gigabyte or two? Clearly something is happening here that I don't understand.

[Love Zhaoying]

8 minutes ago, Qie Niangao said:

unwholesome curiousity about how stuff works?

Subtle! 

[Phil Deakins]

12 hours ago, cykarushb said:

the idea of "too long, didnt read" is meant to catch people with an end synopsis because they scroll past the wall of text

however on here, i just post that as a general summary

I always thought it meant that the post I'm replying to is too long, so I didn't read it all. I still think that, because it makes much more sense to me.

It's followed by a reply to some little bit that was spotted in the post that wasn't fully read.

Edited November 13, 2018 by Phil Deakins

[cykarushb]

6 hours ago, Qie Niangao said:

Is there a commonly used tool for observing what's using VRAM? 

GPU-Z

Its sister program CPU-Z is also pretty useful for other things. Good to have in the nerd toolkit.

[LittleMe Jewell]

10 hours ago, Phil Deakins said:

I always thought it meant that the post I'm replying to is too long, so I didn't read it all. I still think that, because it makes much more sense to me.

It's followed by a reply to some little bit that was spotted in the post that wasn't fully read.

I've always used tldr both ways - to give a summary version of a long post and to indicate in my reply that I did not read the entire thing. 

When used to summarize a post, I have seen the tldr located in both places - at the top of the post and other times at the bottom of the post.

[LittleMe Jewell]

7 hours ago, cykarushb said:

13 hours ago, Qie Niangao said:

 Is there a commonly used tool for observing what's using VRAM? 

GPU-Z

GPU-Z will tell you how much VRAM is being used, but it won't tell you the processes using it.  However, if you bring up GPU-Z with nothing but your basic system running and make note of the current usage, you could then start your viewer and monitor how much the usage increases.  That will give you an idea of what the viewer is using.  

[iceing Braveheart]

On 11/12/2018 at 5:54 PM, Ansariel Hiller said:

Okay, @iceing Braveheart convinced me:

will there be a slider?

http://hg.phoenixviewer.com/phoenix-firestorm-lgpl/rev/4750d36f9ee5

it looks like a easy change from current 2048 to 16384

[Ansariel Hiller]

1 hour ago, iceing Braveheart said:

will there be a slider?

http://hg.phoenixviewer.com/phoenix-firestorm-lgpl/rev/4750d36f9ee5

it looks like a easy change from current 2048 to 16384

No need for a slider. It will allocate all VRAM it finds for textures.