Number of CPUs

[judicator]

Hi, just wondering what's the most CPUs anyone has ran LW on. I have access to a 32CPU Xeon box running W2k3 or W2K but not sure if LW will work on it.

Thanks.

[mattclary]

Only one way to find out! I see no reason why it wouldn't work. You SOOOOOOOO have to run some render benchmarks for us!

[trentonia]

As far as rendering goes, http://www.respower.com runs LW on 550GHz of processor power spread across something like 300 cpus.

[Lynx3d]

Don't mix up a renderfarm with an SMP system...

You can "only" set a maximum of 8 threads in Lightwave to render, so (at least) 24 cpus would be idle on a rendering job inside Layout. I don't know what happens if you launch a couple of screamernet nodes on a single machine...never really used itm because i'm not an animating guy (yet)

[electroNIX]

I think, it depends on the amount of memory, if you only have 128MB for each CPU... this might be a problem...
But If you have a really monstrous machine with some 64GB or so RAM... it would be way too cool to render very big single frames across those 32 CPUs with amleto... :D
So no need to be an "animation guy" to utilize that beast...

[Lynx3d]

"Up to 9 clients on the same host

In case your clients have multiple CPU, you may start up to 9 Amleto clients on the same host."

That won't really be enough for 32 CPUs...? Or how exactly does Amleto work?

[ChrisS]

Originally posted by Lynx3d
Don't mix up a renderfarm with an SMP system...

You can "only" set a maximum of 8 threads in Lightwave to render, so (at least) 24 cpus would be idle on a rendering job inside Layout. I don't know what happens if you launch a couple of screamernet nodes on a single machine...never really used itm because i'm not an animating guy (yet)

Before LightWave was multi-threaded, that was the only way to use multiple CPUs for rendering (multiple instances of Screamernet). I use scremernet for all of my big renders and it's great for batch rendering. It's also a wee bit quicker than multi-threading IMHO.

The only trick would be writing/editing 32 different command lines for each instance of Screamernet. BTW, does anybody have a text editor trick to edit command lines to sequentially number lots of command lines (ack1, ack2 etc...)?

[trentonia]

"Don't mix up a renderfarm with an SMP system... "

I'm not confusing an SMP system for a renderfarm, I'm just saying that by the time you buy a handful of machines for $8000.00 and spend a bunch of time setting them up, networking them together, installing screamernet, and all that, I will have rendered hours and hours and hours of animation for the same cost on the renderfarm. Also, you have to take into account the maintenance, etc. on your mini farm. I just let http://www.respower.com take care of everything. I don't have to spend one second on ANY of the worries. I just pass the render expense onto the client. You can't beat it.

greg

[Extent]

No you are confusing them. Respower's system is not 300 processors in a single box, it is a renderfarm. He's asking how man processors in a single computer, thus SMP. I've only run on a dual system. Back in the day I wanted to build a passive backplane system, but I got bored of the idea :).

I would rather spend 8k on render nodes than give it to Respower for normal use. Because after you've used up your 1600 GHz hours of use I could easily blow thur that workload in under 4 days and then keep on rendering for free. I would defenatly use respower for true final renders if I was on a superhard time budget because I'm sure they render on more than 16 machines at once. How much maintainance do you think simple rendernodes need? I maintain 6 normal use, internetted, desktop computers at home and it dosen't take any effort at all:rolleyes:

I would *love* to see a screenshot of that system running 32 instances of Screamernet, all rendering:D :D :D :D

[Lynx3d]

(...)and then keep on rendering for free.

Ever added up what even a single PC needs electric power a year? :) You definitely can't call that "for free" on a dozend PCs (at least over here)...but anyway, respower's prices really get's quickly to the point where you could aswell buy your own nodes...and if rendering is all they do maintenance goes to zero...and if you know how to do it with two PCs you know how to do it with 20 PCs...

[electroNIX]

well, you could open several instances of amleto and assign each to a different CPU via the Task-Manager... this can also be done with a shortcut.
you can also assign the priority to each instance via the shortcuts...

I dont know the exact parameter for that right now, but you can find this on a lot of Windows2000 pages, maybe even on microsobs site...

[Extent]

Originally posted by Lynx3d
Ever added up what even a single PC needs electric power a year?...

While I can't give any concrete numbers on my system I'm working on the assumption of the computer using 150 watts/hour (number pulled from this site Link, Electricity questions (http://www.michaelbluejay.com/electricity/questions.html) ) And a baseline KiloWatt/Hr rate of $0.12 . With that power consumption the full load power costs for an entire year is about $157.68. If that's a 1GHz machine that puts your GHz/Hr cost at 1.8 Cents an hour in electrical cost. I suspect actual costs might be actually lower with faster processors (I have no idea what kind of system he tested power usage with) with enough memmory to avoid swapping to the harddisk (so they can go into power saving mode) Even if you double the power usage (to 300 watts/Hr) 3.6 Cents in electricity an hour still blows away respowers rates.

The equivalent workcost of 16 1GHz computers for one year solid at Respower would come out to $70,080. Building your own renderfarm and running it for that same period of time would save you $59,557.12 , including the 8k startup cost. Every year after that would save $67,557.12. And like I said, I expect that the realworld numbers would be even lower than the ones I've calculated.

I would defenatly use Respowers services if I needed a scene rendered yesterday but for any other use it really is too expensive.

*apologises for going horribly off topic*

[asmith47]

So does that mean it is ok to power down the disks on render nodes? I started a thread in LW General about heat buildup in my server room, and we got sidetracked into why it isn't practical to rework the AC system at this time. :rolleyes:

[Extent]

I dont see why not, I used to keep my disks spun up all the time but with my new dual system the heat was getting too much. a render node should only infrequently use the harddisk if it's got enough ram for the scene your rendering, shutting the harddrive down will save power, noise, and heat. It dosen't hurt you at all while the system is sitting crunching numbers, it will just delay the next harddrive write or read op. by a couple seconds as the disk kicks back up. While rendering short frames the disk wont have long enough to spin down, while doing longer renders the extra couple of seconds really isn't much to speak of

[garg]

Hi guys,

I'm one of the Amleto programmers. (Just to introduce myself :rolleyes: )

Up to now as we did not encounter someone with more than 2 cpu per computer we fixed the number of instance of Amleto to 9. This can be changed in no time if someone needs it.

Amleto work in 2 different ways depending if you render a single frame across many computers or render many frames across many computers.

1 frame many computers: we called it segments rendering. Amleto set multiple limited regions (one per segment) and render all those "scenes" on different Amleto clients (can be on one or more computers). After all segments have been rendered, our software merges segments and produce one final frame. We smooth merging to improve the merge when the scene contains things like radiosity or DOF.

Many frames many computers: it work more or less like screemernet. Each frame (or block of frames) is sent to different computers.

You may try the trial of Amleto:

http://www.rayserver.com/

Hope this give a bit more information.

Alain Bertrand

[asmith47]

Thanks. I will try enabling the hard disk power down in the BIOS and see if there is trouble.

[ResPower]

Hey guys -

Saw some of the posts about costs. The idea of ResPower is that you get ALL of the computers when you render (as soon as they become available) in order to have speed. Not to offload computation.

Let's make the math easy. Say you bought 1 computer at $1000. That amount of money would get you 2000 GHz hours on our farm. Yes, you'll have the computer forever, but it would only double your outputput speed (1 render node + 1 workstation). If you bought 10 computers, you'd end up spending $10,000, but only increasing your speed by 10.

On the other hand, you could render 250 times faster than that for $1,000 on your final render.

In order to go 250 times faster, you'd have to spend $250,000 + infrastructure costs + monthly upkeep.

Many of our customers use us on a weekly basis - they do modelling and wireframing stuff, then upload for the final renders, with antialiasing and so on turned way, way up. That way they can spend more time tweaking.

So the bottom line is, yes, you can buy your computers, but you'd never be able to afford so many; they'd be sitting there idle while you're modelling and tweaking; and the more you buy, the bigger the pain it is to support them, repair them, etc.

Obviously I'm biased, but unless you're a big company like dreamworks or pixar, it makes sense to outsource and only pay for the time you're rendering. Or else make lower quality animations and buy a couple of computers. Electricity isn't the big thing in the equation.

[asmith47]

Great post, ResPower, you make some valid points, but it does make sense for small houses to have their own farms. Even a relatively long FX shot is only a few hundred frames. At maximum antialiasing and motion blur and whatever, you don't need that many GHz/h to render the average scene at video resolution. With Screamernet, I can use suitably qualified workstations as nodes as well as my farm computers:

5 rendering computers at 2x933Mhz----------9330
3 w/s @ 2x933------------------------------5598
1 w/s @ 2 Ghz------------------------------2000
1 w/s @ 2x2 Ghz----------------------------4000
1`w/s @ 2.5 Ghz----------------------------2500

--------------------------------------------22794

so already I have almost 23 GHz in-house over 19 nodes that I can use after hours and 12 GHz dedicated to rendering when we are using all the other workstations. This means that I have 5% or so of your whole operation, available over half the time (overnight), and 2.5% all the time. 24/7. Paid for, except for the electricity. More than enough to test render things all day and batch render a bunch of stuff at night. Not only that, these days it would be pretty cheap to set up a rack with a bunch of 1u dual xeon servers as the prices for GHz and RAM keeps plummeting. I wonder if you have factored that sort of depreciation into your business model! :eek: I would assume that as hadware prices decrease so should your charges per GHz/H.

As for maintainance, the farm servers work like a charm once they are set up, and workstations need maintainance mainly because of all the different BS software they have to run, and you need to deal with that, render farm or no, so it doesn't add to your cost either.

Electricity costs are always important for sound business and environmental reasons.

PS: I will be setting up an account with you guys and testing it out because one very salient point is you can save someones rear end if they are on deadline and need results asap to keep relations good with the customer or for the occasional shot that has so many frames and high resolution textures that I have outstripped my capabilities :D

Don't get the idea that I am dissing your business, I hope to have enough work that I will be using it! I just won't use it as much as you want me to. :p

[asmith47]

And as for power down on the disks, we have had no trouble getting frames back with this feature enabled!

[asmith47]

Also on the side of outsourceing is the space taken up in your facility, rent on that square footage, and perhaps more importantly the potential added value that the square footage dedicated to in house computing could gain the organization if it was used for something else. It is an interesting argument and time will tell if respower is a viable idea. I have a feeling that it is, if properly managed. Being sure of costs is an inexact science, but the effect of those costs on competitive advantage is always present, so the invisible hand will eventually let us know!

[ResPower]

Hey Asmith -

No, I didn't think you were dissing us at all!

And you're welcome to give it a tray. If you need help figuring it out feel free to give us a call.

Saving folks' butts is what we're all about! Definitely something we're into doing.

http://www.respower.com