SSD or Velociraptor for LW

[JonW]

I run Screamernet from my E5450 which has a Velociraptor 300. I had rendered 900 frames on my W5580 for a Job. Then with SN for a test.

I added a Corsair 120 to the E5450 & installed Lightwave. Copied Scenes, Objects etc to the SSD & ran everything from there.

F9 render x 2 LW instances W5580: 168 minutes.
Screamernet from C drive, Velociraptor: 68 minutes
Screamernet from additional drive, SSD: 61 minutes

These were quick frames to render, 11.2 seconds a frame on the W5580. But because SN is a bit slow communicating, running it from the SSD instead of the Velociraptor has reduced render time by 10%.

Link to some earlier information, post 6.
http://forums.newtek.com/showthread.php?t=123069

Very rough installation of SSD!

[JonW]

Here is a screen shot of 2 nodes on the W5580 with SN running from the E5450, WTM on high speed, SSD left & Velociraptor right.

[Boris Goreta]

I thought LW/SN doesn't access hard drive any more after the scene is loaded for the first time. If a node doesn't have enough RAM it uses hard disk instead so this might be the case but on the other hand you're using your workstation which is probably loaded with RAM.

[JonW]

Other than to be told which frame to render next, & to save the rendered frame so the node can get onto rendering the next frame. It looks as if when a frame is finished & also when a node is waiting for information for the next frame, it needs to get quicker access to a hard drive than I though would have been necessary.

These frames were short, so there is a much higher % of communicating in relation to actual rendering & if frames from different nodes arrive at the same time there must be a bit of a bottleneck.

Looking at the SN window, when it went through it's cycle of saying Ready & rendering next frame etc. It did appear to be much snappier than with the Velociraptor.

So my guess is that if you have an absolute plethora of nodes (I could only wish!) &/or you are rendering relatively short frames, get an SSD or even RAID some SSDs for super quick access to reduce bottlenecks. If you have that much money tied up in say 50 or 100 or more nodes, getting quicker SN access via some SSDs looks like it maybe a worthwhile investment.

[Soth]

it does make even difference when frames are layered exrs that have 50MB each

ive put as much memory in my server as i could afford to not make my customers pay for saving the frames on the disc instead of rendering :P

[JonW]

it does make even difference when frames are layered exrs that have 50MB each

Since you are running a large network/nodes It maybe worth trying running your server assuming it's also a 55/56 CPU/s, with HT off. The computer will be a bit slower overall but each core will be a lot faster. Most of these network programs are only using 1 or 2 cores anyway.

[Soth]

thx for the tip, ill monitor server cpu use, but AFAIK it never ho higher than 10%
have very good RAID card with 1gb of cache

but ill read if turning it off will make the server to use less electricity, i think Ive put too much processor power in that box anyway (inherit 2 old processors from the rendering node)

I could change this two to something with higher clock and less cores maybe, definitely I will do some reading tonight. ;)