Two Nvidia Cards

Comments

• Charlie Judge Posts: 13,251

  October 2016

  The VRAM is not additive. The whole scene will attempt to load to both cards; so, if your scene is greater than 2GB, all rendering will default to your CPU.

  Assuming your scene is less than 2GB, all the cuda cores iof both cards will be used in processing. So, effectively you will have just slightly less than 1950 effective cuda cores.
• mcorr Posts: 1,104

  October 2016

  Charlie Judge said:

  The VRAM is not additive. The whole scene will attempt to load to both cards; so, if your scene is greater than 2GB, all rendering will default to your CPU.

  Assuming your scene is less than 2GB, all the cuda cores iof both cards will be used in processing. So, effectively you will have just slightly less than 1950 effective cuda cores.

  That's interesting and really helpful (thanks!)... so all the cudas become available, but only if the whole scene fits into the VRAM?

  And just to be sure, the vram that is sought out, or used, is the one with the highest number of GB?
• BeeMKay Posts: 7,019

  October 2016

  You might find this interesting: https://helpdaz.zendesk.com/hc/en-us/articles/207530513-System-Recommendations-for-DAZ-Studio-4

  Itay will try to fit the scene into each card separately, that means that if your scene is 1.8GB (That is a dressed and haired Genesis with a very basic background...), and the card that runds the monitot just has 1.6GB free VRAM, the monitor-card is out, and only the non-monitor card with the full 2GB available is "used".

  If your scene is 2.1GB, both cards are aout, and the render will go back to CPU, adding the full 2.1GB to your RAM.
• mcorr Posts: 1,104

  October 2016

  BeeMKay said:

  You might find this interesting: https://helpdaz.zendesk.com/hc/en-us/articles/207530513-System-Recommendations-for-DAZ-Studio-4

  Itay will try to fit the scene into each card separately, that means that if your scene is 1.8GB (That is a dressed and haired Genesis with a very basic background...), and the card that runds the monitot just has 1.6GB free VRAM, the monitor-card is out, and only the non-monitor card with the full 2GB available is "used".

  If your scene is 2.1GB, both cards are aout, and the render will go back to CPU, adding the full 2.1GB to your RAM.

  got it ...thanks!

  How about the part (C) of my questions? What will happen to the screen refresh?
• Charlie Judge Posts: 13,251

  October 2016

  mcorr said:

  Charlie Judge said:

  The VRAM is not additive. The whole scene will attempt to load to both cards; so, if your scene is greater than 2GB, all rendering will default to your CPU.

  Assuming your scene is less than 2GB, all the cuda cores iof both cards will be used in processing. So, effectively you will have just slightly less than 1950 effective cuda cores.

  That's interesting and really helpful (thanks!)... so all the cudas become available, but only if the whole scene fits into the VRAM?

  And just to be sure, the vram that is sought out, or used, is the one with the highest number of GB?

  Then, of course, you will also only have the processing power of the cuda cores in that one card. And, I'm not sure, but it may have to be set manually to only use the card with the most VRAM.
• mcorr Posts: 1,104

  October 2016

  Charlie Judge said:

  mcorr said:

  Charlie Judge said:

  The VRAM is not additive. The whole scene will attempt to load to both cards; so, if your scene is greater than 2GB, all rendering will default to your CPU.

  Assuming your scene is less than 2GB, all the cuda cores iof both cards will be used in processing. So, effectively you will have just slightly less than 1950 effective cuda cores.

  That's interesting and really helpful (thanks!)... so all the cudas become available, but only if the whole scene fits into the VRAM?

  And just to be sure, the vram that is sought out, or used, is the one with the highest number of GB?

  Then, of course, you will also only have the processing power of the cuda cores in that one card. And, I'm not sure, but it may have to be set manually to only use the card with the most VRAM.

  Got it ... Thanks!
• mcorr Posts: 1,104

  October 2016 edited October 2016

  How can I find out the amount of RAM a scene takes up? If I know this, I can minimize the size of a scene until everything can be held in available vram (and therefore, render faster).

  Post edited by mcorr on October 2016
• BeeMKay Posts: 7,019

  October 2016

  The expensive version: http://www.daz3d.com/iray-memory-assistant

  The "best guess" version would be, start your task manager. Load DS. Note down how much memory is consumed. Create your scene. Look at Task manager again, take note of new numbers. Substract first reading. That will give you a very rough guess to the size of your scene. But like with the memory assiastant, there are always additional factors playing into it, like normal maps and other factors, which consume more memory in render than the mere size of them.

  As for your C question, That would fully depend on the size of your scene. As long as it fits on both cards, two cards are faster than one card (more CUDA cores). But from my own experience (having rendered with a GTX 660 2GB card for a year), 2GB is not a lot in any case for Iray, and you'll really have to shuffle around a long to fit things on the card, unless you want to render Vicky in her Temple with a sword. 
• Charlie Judge Posts: 13,251

  October 2016 edited October 2016

  mcorr said:

  How can I find out the amount of RAM a scene takes up? If I know this, I can minimize the size of a scene until everything can be held in available vram (and therefore, render faster).

  This utility may be of help: http://www.daz3d.com/iray-memory-assistant

  That said, only very basic scenes will fit on a 2GB card and 4GB VRAM card is often considered a minimum for practical DAZ Iray use. I have a 4GB card and really wish I had opted for more VRAM.

  Post edited by Charlie Judge on October 2016
• mcorr Posts: 1,104

  October 2016 edited October 2016

  Thank you all for the really useful information and advice. I now know how to proceed.

  BTW (just for the record) ... it's not a 760ti that I already have, but a 660ti .... and the 750ti is the one I am thinking about buying. It's an extra 620 (or something) cudas, which should help noticeably with rendering times, but ... the vram limitation will remain a problem if I only buy the 750ti.

  At the moment, I would be happy even if I could only speed up rendering of the initial basics (setting up the skin, posing the character and getting the lighting right ... all of which would--I think--fit into 2GB). Just being able to do that would be well worth an extra 100 bucks since I spend so much time in that initial phase of the creation process (where you are endless tweaking the lights, background hdri, etc.). How long it then takes to render the whole finished scene that doesn't fit into vram (character with clothes, background, props, etc.) is of less concern to me: I just hit render and go eat something, or surf the internet until things are done.

  That having been said, if I can find an older and inexpensive 4GB nVidia card (not the newer 1xxx ones, but one that will run on my present rig), then I'll go for that. That'll be worth another 100 bucks. Then when all the presently new iray-geared software, hardware, drivers, etc. are stable and their costs come down, I'll just buy a whole new rig (CPU, etc.) with the best graphics card(s) I can afford.

  Post edited by mcorr on October 2016
• Charlie Judge Posts: 13,251

  October 2016

  How about a GTX 750 Ti with 4 GB VRAM: https://www.bhphotovideo.com/bnh/controller/home?O=&sku=1186118&gclid=CjwKEAjw7svABRCi_KPzoPr53QoSJAABSvxffh9bZfrIvN88VXNpP8zUKR2xITdH-6NCrZ1oYv10YRoCXT3w_wcB&is=REG&ap=y&c3api=1876%2C52934714882%2C&Q=&A=details
• BeeMKay Posts: 7,019

  October 2016

  https://www.amazon.com/EVGA-GeForce-Quieter-Graphics-04G-P4-2974-KR/dp/B00NVODXR4

  Refurbished, you can get it at $179, used starting from 210. It's worth the thought if you intend to spend 200 on additional cards anyway.
• mjc1016 Posts: 15,001

  October 2016

  Something to consider...is your power supply able to handle it?
• bad4u Posts: 684

  October 2016 edited October 2016

  The brandnew GTX1050Ti with Pascal architecture are starting around $160 for 4GB models, only need about 75W and GTX1050 usually don't need additional power supply cable attached. The Ti version has 768 cuda cores, and if you plan to go for a new rig some day, it even might be worth to keep that one for running the monitor then and use the larger card(s) for iray exclusively. The even cheaper non-Ti version only has 2GB, so only the Ti is worth considering, at least you are not investing in older technology, I'd say. Maybe read some tests.

  Post edited by bad4u on October 2016
• Mattymanx Posts: 6,996

  October 2016

  Consider having two identical cards, even by the same manufacturer.  Also make sure you register your cards.  EVGA has a good reputation for warrenties.  Zotac is good too.
• jardine Posts: 1,215

  October 2016

  Mattymanx said:

  Consider having two identical cards, even by the same manufacturer.  Also make sure you register your cards.  EVGA has a good reputation for warrenties.  Zotac is good too.

  thanks for the reminder, mattymanx!

  j