Attention GTX 580/590 owners

Wed Apr 02, 2014 5:44 pm

Wow thanks for all your answers guys! it definitely looks like that I need to buy a card with more than the 1,5 GB Ram.. It actually happens so, that I can buy 2 new 3 GB GTX 580s for the same price as one GTX 590. This would give me more Ram to work with, but take up much more space in my computer. Frank how do you manage to run your system with so many fermi cards at once? Could you share some info about your setup?

Again a big thanks to everyone who answered

Best
Henrik

Wed Apr 02, 2014 5:48 pm

FrankPooleFloating wrote:A 580 3GB will absolutely destroy a 640 4GB and make it crawl away in shame. I got two of mine for $200 each and the others for $150 each...

No doubt it would in terms of power but I don't see very many 580 3GBs on ebay for less than $300 each and most are around $400 whereas I can get 3 640s for $400 all day long.

Also there's the question of whether you actually have 3GB. Isn't it 1.5GB x 2 in a 580?

Wed Apr 02, 2014 5:56 pm

xxdanbrowne wrote:Also there's the question of whether you actually have 3GB. Isn't it 1.5GB x 2 in a 580?

No, the 580 came in 1.5 & 3GB versions. It's the 590 that split the 3GB into 1.5GB for each GPU.

Wed Apr 02, 2014 6:04 pm

kavorka wrote:I do think this is a bad time to want to set up a farm. If/when the cloud comes out, it might be much better for cost/efficiency than making your own farm. You also have to deal with electricity and maintenance.

My biggest question about the cloud rendering vs getting a local beasty rendering GPU/farm, is the bandwidth needed to use it. People are discussing weather 3 or 4GB is enough VRAM for a GPU, but I'm thinking more alone the lines of "how long is gonna take to upload that much scene data to the cloud before anything starts rendering?"

Wed Apr 02, 2014 6:12 pm

Nope
The GTX 580 have two models, one with 1,5 gb and the other with 3gb... all for one GPU.
After buying the 590, Nvidia launches the 580 with 3gb and i had to buy one... 1,5gb simply wasnt enough.
Nowadays with my two Titans, i use the 580 only for the system... the GTX 580 is a great card...
Now we are, here at the office, pretending to invest in 4x 780 Ti with 6gb, when they come available here in Brazil... we are just waiting to see if Nvdia will launch the GTX 880 with 12gb by the end of the year... if they are coming, we will have to wait a little more...
Fabio

Wed Apr 02, 2014 6:18 pm

My biggest question about the cloud rendering vs getting a local beasty rendering GPU/farm, is the bandwidth needed to use it. People are discussing weather 3 or 4GB is enough VRAM for a GPU, but I'm thinking more alone the lines of "how long is gonna take to upload that much scene data to the cloud before anything starts rendering?"

This is a good question. In my area, we have relativly slow internet connections, and there are not enough buildings to have a chance to get optical fiber in a close future.

There is an other problem imho with the cloud. Some of you may remember that I moaned a lot when I lost my main client in the past when Refractive Software decided that Octane will require Online connection. Some clients (for industrial prototyping) have very strict confidential rules, and some require complete privacy and refuse to allow the data to be outsourced through the net on a distant renderfarm.

As an artist with few knowledge in security networks, I can't give them any warrantie and can't affort engaging my own responsibility if the cloud is used. So local rendering can help keeping clients.

When working under non disclosure agreement, each time I am not at home, the HDD drawers with clients data are put in a safe.
I can't on one hand sign a non disclosure agreement and take care of data, and on the other hand sent data in the cyberspace, out of any control.

Wed Apr 02, 2014 6:55 pm

schodt wrote:Wow thanks for all your answers guys! it definitely looks like that I need to buy a card with more than the 1,5 GB Ram.. It actually happens so, that I can buy 2 new 3 GB GTX 580s for the same price as one GTX 590. This would give me more Ram to work with, but take up much more space in my computer. Frank how do you manage to run your system with so many fermi cards at once? Could you share some info about your setup?

Again a big thanks to everyone who answered

Best
Henrik

Yo Henrik:

Corsair Obsidian Series 900D
Rosewill HERCULES 1600W
ASUS P9X79 WS
Intel Core i7-3820
Corsair Vengeance 32GB

Two of my 580s are currently water cooled with The Mod http://www.overclock.net/t/1203528/offi ... ka-the-mod and the other two still have shrouds/fans. I will likely go full custom loop in near future. But might just do the mod on the other two... we will see. I listen to dark ambient all day long and need to silence these two bastards yet - so they stop messing with my listening enjoyment.

Don't go less than 1500W for four 580, 680, 780 or Titans. Hope this helps Henrik.

Fri Apr 04, 2014 6:51 am

You got a pretty sweet setup there Frank. It looks like a great idea to watercool the cards, I'll definitely have to look into that.

Thanks again!

Fri Apr 04, 2014 7:22 pm

xxdanbrowne wrote:
FrankPooleFloating wrote:A 580 3GB will absolutely destroy a 640 4GB and make it crawl away in shame. I got two of mine for $200 each and the others for $150 each...
No doubt it would in terms of power but I don't see very many 580 3GBs on ebay for less than $300 each and most are around $400 whereas I can get 3 640s for $400 all day long.

Also there's the question of whether you actually have 3GB. Isn't it 1.5GB x 2 in a 580?

Sat Apr 05, 2014 6:39 am

I would never set up a farm right now.

I've been thinking ever since I started using Octane, that Next Limit better get moving on GPU rendering for MaxwellRender, or they will be signing their own death certificate. I really think that they believe that relying on render farms will save them the hassle of learning GPU programming.

Other than that, I need all the VRAM I can get...I am a poly-heavy guy and always will be. We all have our way of working and that one is mine. Since I build my models in ProE/Creo, I do not get the advantage of LOD control and caging...I'm stuck with single resolution obj export, unless I spend the extra time doing models at different resolutions for different distances in a scene. Until I buy BalancerPro, Re-UVing everything for a different resolution mesh is out of the question. So my Octane scenes are going to be VRAM heavy for awhile. One GTX Titan would be perfect for me.