Advice on hardware options before Octane purchase please!

[Jaybee]

Hi all,

Very new to GPU rendering so I have a couple of questions before I purchase a full licence. I know I’m going to have to upgrade my current hardware to be able to use Octane properly and so I’ve been doing a lot of reading of these forums and of the good advice contained within.

I want to add a second (but different) nVidia graphics card to the system, purely for 3D rendering via it's CUDA cores and I will need to upgrade the PSU as well.

Current System is an i7 920 o/c to 3.04Ghz, Asus P6TD, 12Gb RAM, 160Gb Intel SSD, 4 x 1Tb HDDs in an Icy Dock bay, Corsair 550W PSU. Case is a LianLi PC-7 FNB.

I currently have an nVidia Quadro FX1800 as the gfx card and want to keep this as it’s fine for running the 2D apps. I only have space for one extra card and given all the reading I’ve been doing here I’ve reached the following conclusions:

• As I only have 12Gb RAM (6 x 2Gb Corsair) (board can hold 24Gb max) a 4Gb GPU would never be fully “loaded” so a 2Gb or 3Gb card would be all I could use.
  
  Fermi cards are benchmarking faster than ‘equivalent’ Kepler cards (i.e. 580 vs 680).
  
  Fermi cards need more power so a bigger PSU is desireable
  
  Fermi cards are on the whole cheaper for the performance return than Kepler cards.

My initial thinking (when I first thought the more cores you had, the faster it would run) was to go for a 680 or 690. Now, after reading up on the benchmarking threads, I’m thinking more along the lines of a 3Gb 590 (£200 cheaper than a 690) but would a 690 give the same performance as a 590 and become better as Octane is tuned for Kepler?....Decisions, decisions! I would upgrade PSU to Corsair modular to keep cabling tidy. Would an 850W be OK (prob for 690 only) or should I go for the 1200W (overkill but should only draw what it needs?)?

Here's a shot of the internals! Looks like the GTX 590 card should fit but I'm no expert....





Would be grateful for any advice (or reassurance!) and/or things to watch out for on the upgrade.

TIA

Jaybee

[kavorka]

you do need a bigger PSU.
On Nvidia's website, it says how much power each card will take. Here is the 690 (300 W):
http://www.geforce.com/hardware/desktop ... ifications

the 5.xx series does perform a little faster, but has less VRAM, takes more power, and puts off more heat.
the 590 and 690 are actually 2 x80 cards glued together, so a 3 gig 590, actually only has 1.5 gigs of available VRAM for Octane, keep that in mind.

If you only have 1 slot and have the money, go for the 590 or 690 to maximize everything. you wont be disappointed .
def keep that Quadro for display (disable it as a rendering card in Octane preferences), will make Octane work more efficiently and allow you to render in Octane while using your computer for other things with no slow down.

I am not sure about "As I only have 12Gb RAM (6 x 2Gb Corsair) (board can hold 24Gb max) a 4Gb GPU would never be fully “loaded” so a 2Gb or 3Gb card would be all I could use." I have never heard anything like this before.

[pixelrush]

12gb would be a suitable size for a 2gb card without paging to the disk during loading the largest scenes you could fit into it. This would include a GTX690 with 4gb ie 2xGTX 680. 24gb is OK for 4 gb cards.

[Jaybee]

Thanks for the replies guys. I'm learning new stuff by the hour here...

@kavorka I knew the 590/690 was to *80 cards together but I had no idea that the vRam was effectively halved. Thanks for clarifying that, it means I need at least a 4Gb version if I go for the 590 which may be hard to come by. 3Gb is the largest I've seen still available retail.

I'm doing some arch-vis and some studio type product and some compositing into photographs so I use a lot of photo-real textures and output stills in the 4000-5000px longest side range (about 50Mb in 8bit). So I think I might need at least a 2Gb card (or 4Gb *90 series) to build in some safety room.

I guess it's a case of either that compromise on vRAM or a 4Gb 580 or 680 and take the performance speed hit for now. My idea is to get comfortable with an Octane workflow for a few months then build a new system where I could take and transfer over the card I'm about to buy along with another of the same (or better) type. So might be best to go 4Gb 680 for now else the 2Gb of the (say 690) would limit scenes it could be used for even if paired with a newer 4Gb in the future....

....this is not easy is it...

@pixelrush Thanks for that explanation. I had read some posts on voxelization and the need for more system RAM to "load" the card. The P6TD takes 24Gb max. I currently have 12Gb. I could buy some more RAM (6 x 4Gb) but I can get by with being careful on textures etc I think until I get a new rig built.

Might have to do a "best of three" coin flip...!

Is there any thought that the Kepler cores might catch up Fermi in terms of performance (with newer versions of Octane) or is it likely to stay with Fermi ahead on scores until the next generation of cards?

[pixelrush]

The Kepler series have had disappointing sp computing performance relative to Fermi due to the chip architecture. Recent AMD cards make them look fairly pathetic by comparison.
I think Otoy guys mentioned their code is about as efficient as it gets without rewriting everything to be Kepler specific and even then it probably won't amount to much..
Apparently the next generation Maxwell cards are delayed and there will be a refresh of Kepler due to start arriving in the next few months. The info I saw suggested some cards like the 770,780,790 would be slightly rearchitected and get a decent boost of the order of 30% but others in the 700 series would only get an up clock like 10-15%. If you can wait a little while to see what the new ones are like for Octane purposes, it might be a good idea. I would like to think these might be on par with Fermi or perhaps a wee bit faster.

[Jaybee]

Conundrum time

You're right, we are near a new 7 series of cards coming "soon" whenever that is (these IT deadlines have a long history of being pushed back at the last minute). I'd love to wait and see what the 7 series develops into with Octane though I think it might take longer than a few weeks to get hold of any card (I can see a huge rush by gamers if they're up on 6 series performance) with 4Gb, and then, I assume, an Octane release that will work with them. And what if the performance boost is negligible? I may have wasted 4-5 months of a learning curve.

Having thought this through for a number of days I think I'm going to upgrade PSU to 860W & go for a 4Gb 680 o/c now in the current rig, just to get onboard with Octane and get a usable GPU production workflow sorted. I'd do this knowing I can re-use it in a new rig to be built later this year alongside a 7 series or two perhaps..! I like the linear scalability and the way cards of the 4-5-6 gens can be mixed together.

Thanks again for the suggestions everyone, looking forward to many more discussions here on what seems a very helpful forum.