Unusual rendertimes for animation frames

[tricky]

Not sure if this is related to vray, but any opinions would be appreciated.

Last weekend, I sent a backburner job to 14 render machines on our local farm. The network is 1GB.

We have 8 old machines, 3 new machines and 3 workstation machines that were being used in the job. There was no GI in the job, and the frames were 1080p.

Render times for frames were typically as follows:-

NODE 1 - 1hr 50 (XEON 5160 @3GHz, 4GB RAM)
NODE 2 - 1hr 50 (XEON 5160 @3GHz, 4GB RAM)
NODE 3 - 1hr 50 (XEON 5160 @3GHz, 4GB RAM)
NODE 4 - 1hr 50 (XEON 5160 @3GHz, 4GB RAM)
NODE 5 - 1hr (dual XEON 5160 @3GHz, 4GB RAM)
NODE 6 - 1hr (dual XEON 5160 @3GHz, 4GB RAM)
NODE 7 - 1hr (dual XEON 5160 @3GHz, 4GB RAM)
NODE 8 - 1hr (dual XEON 5160 @3GHz, 4GB RAM)
VISION 1 - 50min (i7 2600k @ 3.4GHz, 16GB RAM)
VISION 2 - 50min (i7 2600k @ 3.4GHz, 16GB RAM)
VISION 3 - 15min (!) (i7 2600k @ 3.4GHz, 16GB RAM)
WORKSTATION 1 - 40min (i7 2700k @ 3.5GHz, 16GB RAM)
WORKSTATION 2 - 40min (i7 2700k @ 3.5GHz, 16GB RAM)
WORKSTATION 3 - 40min (i7 2700k @ 3.5GHz, 16GB RAM)

I am trying to understand why VISION 3 (highlighted in red above) is rendering so fast when similar spec machines are 3 or 4 times slower. All except the workstations (which are running Windows 7) are running XP64. As far as I am aware, they are installed about the same as eachother.

Any suggestions?

[super gnu]

did the fast frames look any different to the slow ones?

[tricky]

did the fast frames look any different to the slow ones?

They don't seem to. Here are a couple of test frames. Same spec PC rendered each one. They both have a RAID 0 set up with the same hard disks. Both on XP64 SP2. Same motherboards.

[super gnu]

yep they look pretty identical.

cant imagine what would cause that odd difference. normally youd have one machine that was slower than the rest. but to have one thats 3 times faster is strange to say the least. id find out what is wrong with that machine and make sure all the others get the same treatment!



hyperthreading or any other bios differences?


any other components subtly different ( well subtle wouldnt explain 3x rendertime difference)


are the other machines at 100% cpu?

you get the same difference rendering locally?

tried any other benchmarks on that machine?


if its repeatable, ill give you top dollar for that renderbox

[tricky]

Hahaha.

I guess I need to find some piece of benchmarking software to run on the 3 VISION machines that will test everything. Cinebench? As far as I know (I only built them a couple of weeks ago) they are identical apart from the power supplies!

[super gnu]

ahh thatll be it. the power supply is giving the pc more POWER!

cinebench is probably a good test in this case.



interested to know the final reason for this...

[stevesideas]

If its not the power supply I would definitely recommend checking your bios settings. I remember changing some bios settings and the system ran very slowly...cant remember what it was..sorry if that's not very helpful.

[joconnell]

There's a thing called a snoop filter that caused a load of intel and xeon machines to run slow on supermicro boards - might be an equivalent for you?

[tricky]

There's a thing called a snoop filter that caused a load of intel and xeon machines to run slow on supermicro boards - might be an equivalent for you?

Will take a look. Thanks.

[tricky]

One thing that I *think* is different about the odd machine is that TweekUI was making the machine autologin. All other machines have backburner and spawner running as a service on an administrator account. The fast machine also has them running as services on the admin account, but it is actually logged into a windows session.

Could this have an effect?

[flino2004]

I would return the good ones to exchange them for defect ones like vision 3, and I'll ask for a discount too

[vlado]

One thing that I *think* is different about the odd machine is that TweekUI was making the machine autologin. All other machines have backburner and spawner running as a service on an administrator account. The fast machine also has them running as services on the admin account, but it is actually logged into a windows session.

Could this have an effect?

It's not impossible... I guess you can easily check if you log in on one of the other machines and check the render times. BTW is the 3ds Max Bitmap Pager enabled?

Best regards,
Vlado

[vance3d]

What was the cause for the difference in render times?

V Miller

[stevesideas]

Not sure what the reason might have been but I noticed something similar and then logged in to the faster machine to see I had sent the job with DR on. That single machine was using the other render nodes for DR hence the speed difference. Without DR the times evened out for me...maybe this was the same thing.