Computer not fast enough, Vray + Sketchup

[lm1985]

Hi all,

 

First ever forum post so apologies if this is a stupid Q;

 

Have been asked by a client to render an image at 300DPI to be printed 4m wide by 2.5m High. with my current computer spec I will be grey and old before this renders (im 28)

 

Im using vray 1.49 for sketchup pro 8 computer spec as follows;

 

-Gigabyte GA-B75-D3V Socket 1155 VGA DVI 8 Channel Audio ATX Motherboard

-XFX HD 7850 Core Edition 2GB GDDR5 Dual DVI HDMI Dual DisplayPort PCI-E Graphics Card

-Microsoft Windows 7 Home Premium w/SP1

-Intel Core i7 3770 3.4GHz Socket 1155 8MB Cache Retail Boxed Processor

-HGST Deskstar 1TB Hard Drive SATAII 7200rpm 32MB Cache - OEM (for storage and files)

-SanDisk 128GB Pulse SSD - 2.5" SATA-III - Read 490MB/s Write 380MB/s (for programs)

-Corsair 16GB (2x8GB) Ddr3 1600mhz Vengeance Memory Kit

 

I need to severely upgrade my PC and am willing to pay 1k+ if it dramatically improves performance.

 

Thanks in advance

[Chris MacDonald]

The chances of them needing the render to be 300DPI are slim to none at that size (which would work out at 47244 x 29527 pixels!). It's just something that clients request without really knowing what they're talking about. 300DPI is for things that you view from about a foot or two away - good luck reading a 4 metre advertising board from two feet away!

 

I'd just render out at 72dpi for those dimensions (11338 x 7086, so still pretty big). It might also be worth pre-empting the printer coming back with "this file isn't at 300 DPI" by resizing it in photoshop to 300 DPI - they will never know.

 

As for your PC...

 

Wait for dtolios to get involved in the thread!

[Dimitris Tolios]

I know that Vray for Sketchup (VfSU) is a pain as thereis no estimated time to completion bar etc, but when you say "it takesforever", in which region of forever you are referring to? What kind ofperformance improvement do you expect over that?

 

Because the fastest upgrade available in the region of$1K (tho you prob mean Quids) for you, will be 50-100% better...if what youneed is in counted in hours, and what you are willing to pay (or afford) cangive it back in "days", you probably need to re-evaluate either yoursettings, or your upgrade budget, or your expectations...few years ago, GIrenderings would need 1-2 days for a 3000px wide image, in the fastest desktopPC available: 20, 30, 40 hours of rendering for a single image, before PP wasto be expected. Now we "grow old" if it is not done in a couple ofhours? Give us a sense of what you are getting, and where you need to be.

 

Regardless of theabove:

Fastest upgradeyou can do with you current system: switch to a 4930K + X79 board +cooler. Around50% faster than what you have right now. Would be in the$800-900 range (dependson cooler chosen, don't know in pounds). Would mean taking the time to setupnew Windows + all your apps etc ofc - I would not recommend to anyone swappingOS drive to a different motherboard. Windows are pretty tolerant, but you mightend up spending more time troubleshooting afterwards than going for a cleaninstall (esp. when you have an SSD that speeds things up).

 

Smarter move?

Fastest buy you can do within most budgets: go forrendering nodes.

You can probably get a couple of FX-8350 / 16GB renderingnodes within your budget or a couple of i7 based nodes for 20-25% more than theFXs. That will give you (in either case) about 3 times the current renderingpower (or 3x faster results), for a bit more than 1.5x the speed you would getgoing for a single CPU approach.

 

At any rate, this kind of resolutions are a killer forVfSU, as it is a 32bit app and doesn't handle memory well past the 3GB~3.5GBrange.

Probably you already know that you will have to renderdirectly to a file, turning Vray frame buffer off or switching it to previewmode to limit the % of memory spent on just displaying the progress in fulldetail, that might end up with the program crashing mid-way.

Chris nailed it as far as "requirements"go...nobody does this kind of renderings on a single frame, on a single PC etcand expects a "clean and fast" result. It could be a composite ofmany smaller frames, or it could be a large single one, but for Blinn's shake(James Blinn), don't try to render300dpi billboard signs! It is a waste.

 

I did my thesis project presentation in VfSU, needed topopulate a 3mx1.2m board (10ft x4ft) that would be seen from pretty darn close(not like a billboard)...even then, I went for images 5-7000px wide...could notafford the time for bigger, or the hassle for VfSU crashing halfway, few daysbefore the "bigday"...

 

72dpi in really largeformats are passable, 100 dpi are ok, 150dpi is great.

 

300 dpi are for things people will hold in their hands,11x17s or so, and even then,most people "demanding" it would have noidea what it really means -it’s just an industry meme...

 

Anecdote: I clearly remember the early days of D-SLRs,where stock and news agencies would "downplay" 8~11MP images (nativesensor resolution) as sources that were not meeting their "300dpi digitalfile" rules...

 

What if my EOS 20D would produce sharper results @ 8MPthan what I was getting through the same lens and a 4800dpi Nikon film scanner($1800+ machine) using 35mm film? What if the 1Ds @ 11MP was even better?

 

The scanned image was bigger so it had to be better,right? 30+ MP, 48bit, yadayada…if it is TIFF and 300dpi, the prophecy isfulfilled. That's all they’ve cared for.

Edited November 20, 2013 by dtolios

[lm1985]

Ok I appreciate it was a slightly vague question. I find it difficult to explain my rendering time as it is so varied. I would say an average render at approximately 4000x3000 pixels may take 1/2 -1 day.

 

I did mean £ sterling yes. I feel if i could 1/2 or 1/3 my rendering times i would be over the moon and spending £1500 on upgrades would be more than worth it.

I have considered a building a render farm as used PCs at slightly lower specs seem to drop considerably in price, i assume this is because everybody is constantly striving for the best of the best!

Something else that i see has been discussed often is the GPU rendering ability for VfSU2.0.

(Another vague question coming) I understand the more nodes = faster rendering however would using Vray 2.0 with an upgraded graphics card 'significantly' reduce the need for lots of nodes.

 

 

I think my real problem is i'm finding it difficult to compare components performance with render time.

[Dimitris Tolios]

The biggest the job, the more "scalable" the results are, as inefficiencies due to network delays etc get ironed out: a render job that would take a fewminutes doesn't scale perfectly between multiple machines, as it takes sometime for the network to communicate assets / model etc to all nodes, somethingthat costs seconds here and there.

 

Same seconds or minutes of delay in a job that will last hours either way, are a drop in the ocean.

So, it is safe to say that a rendering job that would occupy a single 3770KCPU for 10 hours, will be completed roughly in 5 hours by 2x 3770Ks and 3h20min by 3x 3770K etc.

 

FX8350s are pretty good when all cylinders (cores) are firing, so I wouldsay that as far as rendering nodes go, a 8350 will match or in some cases surpass a 3770K.

 

BUT, the scene preparation from Sketchup geometry + materialsfor the VRay engine is single threaded. That

means that the machine initiatingthe rendering (your workstation) will be better off being intel CPU based, and ofc the higher the clock, the better. An overclocked i5 will surpass a stock i7 in this process (same generation), which depending on scene complexity might last pretty darn long.

Time savings are actually meaningful when you are often testing even small resolution frames, where the time it takes for the scene to be "translated" for VRay to start, might last as long or even longer than the actual rendering (or it might feel this way).

[lm1985]

Thanks all for your comments guys, I've certainly got some thinking to do. FYI the 'prospective client' has decided to go to a large marketing company instead due to the resolution, doh