Adding a Second Nvidia Card

Granville

I bought a new system just before iray was added to DAZ studio, so I didn't pay attention to things like ram on the card. Now I am asking Puget Sytems to add a card to my existing GTX 760. They are going to upgrade my power supply too. My hope is to drive 3 monitors off one card and render with the new card.

• What card should I add?
• Are there any things to avoid?
• What driver conflicts might I run into?

Thanks in advance for sharing your expertise.

nicstt

Plug in the card, connect power leads.

Make sure your PSU can handle it.

You might need to update drivers, but if they're fairly recent you should be ok. I'm using two cards on 358.91. Beware game ready; as the name suggests they are game ready - and usually for specific games, and can mess stuff up.

Do NOT SLI the cards. Disable in software I believe works, if you want SLI for games.

mjc1016

Any Nvidia card with 4 GB of memory or greater...the 900 series being the 'ideal' ones.  As long as you aren't mixing GeForce and Quadro, there won't be any driver conflicts...they all use the SAME drivers.

nicstt

Card to add is a whole different subject; personally, I'd wait for Pascal due in a month or three.

But at least 4GB or GPU RAM, preferable a 980ti or a Titan, a 980 isn't bad, and 970 is decent value for money.

 

TabascoJack

I've got a 970 and a 760 in my rig.  I only render on the 970 as it has 4GB.  The 760 powers two monitors and the 970 powers the third.  More importantly, I upped the memory in the box to 16GB from 8.  

When Daz / Iray starts needing to move memory around, 8GB isn't enough to avoid paging to disk.  The extra 8GB of RAM avoids the situation.

kaotkbliss

remember, if you're a gamer as well (like me :) ) and you get 2 different cards, you can't SLI them. They can be from different manufactures, but they must be the same model. Dissabling SLI in the nvidia control panel should work to use the cards with Daz (then re-enable to game)

If you don't game, then this whole post is pointless :D

joseft

As mentioned above, you should wait a month or so until nvidia releases the new range of GPU - Pascal. Their development road map in terms of what the next generation of GPU has over the previous is becoming exponential rather than linear - so the Pascal cards are really a big upgrade over the 900 series GTX cards. Which means what you get for your money for a Pascal card will be much better than the 900 series cards. Now is definitely not the time to be buying new video cards for that reason.

GTX760's support 4 displays, so you should have no problems driving your 3 monitors off the 760 and using your new card to power renders. Just stay away from SLI, and use Iray render settings to ensure it is only using your new card for the renders

Mattymanx

I would recommend having two identical cards.  Have SLI setup for gamming and you can disable it in the software for Iray rendering.  YOu can render with both cards in Iray to spped thigns up a lot.  If you need to do other things while it renders then you may have to use only one card

alexhcowley

The latest Pascal rumour is that Nvidia are going to announce it at their GPU technology conference next week, most likely in their CEO's keynote speech on Tuesday. 

Cheers,

Alex.

Granville

Wow, Thanks everyone for sharing your vaste expertise. I didn't know about the Pascal architecture. I have saved up $1100 so far. I will keep saving each month.

Widdershins Studio

So Pascal was announced officially but we probably won't see cards for consumers till late 2016 going into 2017.

I'm tempted now to buy that second 970 I have been itching about.

Wondered if others are still going to wait at this point.

ScarletX1969

Pay attention to the CUDA cores on the Nvidia cards.  As another poster said, have two identical ones help, but is not necessary.  I have 2 GT 740s and a GTX 940.  I use one GT 740 for three monitors and the other two for rendering.  I have 32GB of RAM.  Render time is still gonna depend on the complexity of the scene but it has taken a scene that took well over 12 hours to render and rendered it in 10 minutes with my setup.

FSMCDesigns

LOL, and I thought gamers were fanatical about hardware and performance, they have nothing on you guys, LOL.

I currently have a 6 yr old i7 64 bit system with an older 9800 gtx 2 gb card and can get good Iray renders using either CPU or GPU in about the same time luxrender takes. Can't wait to see what the new rig coming next week (especially with gaming!!) does - skylake 6700K , 16 gb DDR4 2800, and a 970 4GB card. Even as a gamer it's real hard to justify over $500 for a vid card

Widdershins Studio

Hehe, I used to game a lot, not so much at the moment. Since I've got more serious about this I've fussed over hardware far more than when I was a gamer.

I'm on the fence about a second card. The benefit it brings is more speed - I might just decide to convince myself to be more patient. 

bad4u

Widdershins Studio said:

So Pascal was announced officially but we probably won't see cards for consumers till late 2016 going into 2017.

I'm tempted now to buy that second 970 I have been itching about.

Wondered if others are still going to wait at this point.

970 is definitely a good choice for the next 1-1.5 years at least, especially at its justifiable price tag.

NVidia indeed officially announced the first Pascal chip GP100 yesterday, showing samples at GTC2016, but not even one working card yet. The GP100 with Pascal architecture supports 3584 cuda cores, 16GByte HBM2 memory, fast NVLink interconnect, produced in in 16nm FinFET+-

BUT following Jonah Alben, NVidia chief architecturer for that chip, the GP100 is not thought for consumer or gamer cards, but the pros and supercomputing first, especially HPC and deep learning - and the price tag most probably will reflect that. You can expect first cards for that later this year, and some manufacturers might offer cards with GP100 for consumer/gamers later, but rumors say these will only use GDDR5/X then instead of HBM2 and probably cut back on (for gamers) less important double precision calculations also.

What annoys me a lot more is, that despite using the 16nm it is a huge chip , with 610mm² even larger than the Maxwell with 601mm² (I think that's used on 6GB GTX980 and/or Titan, if I remember correct) and it will consume even MORE power, not less. That means while a GTX970 usually is rendering at ~150Watts and the GTX980 at ~250Watts, the new GP100 Pascal is announced with 300Watts.

Btw. CUDA 8.0 will support Pascal, it should be available in June.

Tesla GP100 cards should be available to research facilities and cloud provider later in 2016, but the GP100 chip should not be available to OEM manufacturers for cards until first quarter 2017, so don't expect gamer cards before second quarter to mid 2017. This is info from Jen-Hsun Huangs presentation of Pascal architecture on GTC2016, who is NVidia boss I believe.

I don't think there's any reason to wait for Pascal if you want to render NOW, maybe they get affordable in two years or so, when less high-end chip versions for consumer market have been announced. As it is now I don't expect Pascal cards for under $1000 for the next 1.5 years at least.. but good luck waiting ;)

 

mjc1016

bad4u said:

Widdershins Studio said:

So Pascal was announced officially but we probably won't see cards for consumers till late 2016 going into 2017.

I'm tempted now to buy that second 970 I have been itching about.

Wondered if others are still going to wait at this point.

970 is definitely a good choice for the next 1-1.5 years at least, especially at its justifiable price tag.

NVidia indeed officially announced the first Pascal chip GP100 yesterday, showing samples at GTC2016, but not even one working card yet. The GP100 with Pascal architecture supports 3584 cuda cores, 16GByte HBM2 memory, fast NVLink interconnect, produced in in 16nm FinFET+-

BUT following Jonah Alben, NVidia chief architecturer for that chip, the GP100 is not thought for consumer or gamer cards, but the pros and supercomputing first, especially HPC and deep learning - and the price tag most probably will reflect that. You can expect first cards for that later this year, and some manufacturers might offer cards with GP100 for consumer/gamers later, but rumors say these will only use GDDR5/X then instead of HBM2 and probably cut back on (for gamers) less important double precision calculations also.

What annoys me a lot more is, that despite using the 16nm it is a huge chip , with 610mm² even larger than the Maxwell with 601mm² (I think that's used on 6GB GTX980 and/or Titan, if I remember correct) and it will consume even MORE power, not less. That means while a GTX970 usually is rendering at ~150Watts and the GTX980 at ~250Watts, the new GP100 Pascal is announced with 300Watts.

Btw. CUDA 8.0 will support Pascal, it should be available in June.

Tesla GP100 cards should be available to research facilities and cloud provider later in 2016, but the GP100 chip should not be available to OEM manufacturers for cards until first quarter 2017, so don't expect gamer cards before second quarter to mid 2017. This is info from Jen-Hsun Huangs presentation of Pascal architecture on GTC2016, who is NVidia boss I believe.

I don't think there's any reason to wait for Pascal if you want to render NOW, maybe they get affordable in two years or so, when less high-end chip versions for consumer market have been announced. As it is now I don't expect Pascal cards for under $1000 for the next 1.5 years at least.. but good luck waiting ;)

 

And there you have it...the wait is over/too long to be practical.

joseft

bad4u said:

Widdershins Studio said:

So Pascal was announced officially but we probably won't see cards for consumers till late 2016 going into 2017.

I'm tempted now to buy that second 970 I have been itching about.

Wondered if others are still going to wait at this point.

970 is definitely a good choice for the next 1-1.5 years at least, especially at its justifiable price tag.

NVidia indeed officially announced the first Pascal chip GP100 yesterday, showing samples at GTC2016, but not even one working card yet. The GP100 with Pascal architecture supports 3584 cuda cores, 16GByte HBM2 memory, fast NVLink interconnect, produced in in 16nm FinFET+-

BUT following Jonah Alben, NVidia chief architecturer for that chip, the GP100 is not thought for consumer or gamer cards, but the pros and supercomputing first, especially HPC and deep learning - and the price tag most probably will reflect that. You can expect first cards for that later this year, and some manufacturers might offer cards with GP100 for consumer/gamers later, but rumors say these will only use GDDR5/X then instead of HBM2 and probably cut back on (for gamers) less important double precision calculations also.

What annoys me a lot more is, that despite using the 16nm it is a huge chip , with 610mm² even larger than the Maxwell with 601mm² (I think that's used on 6GB GTX980 and/or Titan, if I remember correct) and it will consume even MORE power, not less. That means while a GTX970 usually is rendering at ~150Watts and the GTX980 at ~250Watts, the new GP100 Pascal is announced with 300Watts.

Btw. CUDA 8.0 will support Pascal, it should be available in June.

Tesla GP100 cards should be available to research facilities and cloud provider later in 2016, but the GP100 chip should not be available to OEM manufacturers for cards until first quarter 2017, so don't expect gamer cards before second quarter to mid 2017. This is info from Jen-Hsun Huangs presentation of Pascal architecture on GTC2016, who is NVidia boss I believe.

I don't think there's any reason to wait for Pascal if you want to render NOW, maybe they get affordable in two years or so, when less high-end chip versions for consumer market have been announced. As it is now I don't expect Pascal cards for under $1000 for the next 1.5 years at least.. but good luck waiting ;)

 

pretty much the general consensus yep.

I was one of the people waiting on news of Pascal before building my new machine. Given that it looks like it will probably be around this time next year before we see a pascal version of the Titan available to consumers, i decided not to wait around for it.

Which will probably mean i will wait until the Volta generation until i upgrade again

bad4u

To clarify the very last section - like in the sentences before I'm still talking about "GP100 Pascal" when it says "Pascal". No doubt there will be less high-end, non-GP100 chip versions available that are also Pascal based for consumer market and they probably start in the $500 - $1000 range when released this year, but beside missing the HBM2 they might have only 256bit memory interface, compared to the 384bit that GTX980 and Titan already have today. Performance might be in similar range then, but only time will tell. At least these should be more energy efficient then (chip is far smaller compared to GP100).

Personally I'd say if you currently look at GTX970 - go for it, it's good value for your money (though only 3,5GB can actually be used full speed). If you look at GTX980 - only the 6GB versions make sense in my eyes (as these perform very similar to the 12GB Titan), but those are in a price range where you might want to wait where the 8GB consumer Pascal products settle. If you want high-end cards with GP100 Pascal - doesn't make sense to wait, buy what you actually need and upgrade in few years when they get affordable..

LayLo 3D

I just bought a new GPU even though I new about the development of Pascal. I got an EVGA GTX 980 Ti Hybrid. But, I have excellent credit and get these crazy credit card offers all the time, sign up for our card, spend $1,000 get a $300 statement credit, plus 0% interest for 9 months. When I got that offer in the mail, all it said to me was, "you should buy that GTX 980 ti you've been drooling over" lol. Oh yeah, the credit card even gives me an extra 3% cash back on any computer hardware I buy, so I basically got that card for $360!! Who could say no to that? Now I'm not saying people should go run up a bunch of credit card debt... Just saying that's how I justified buying my new card =). But, for the record, I've never paid a cent in interest on any credit card.

Love the card so far, by the way. Stays cool and quiet even when at 100% usage for an hour.

Ghosty12

bad4u said:

Widdershins Studio said:

So Pascal was announced officially but we probably won't see cards for consumers till late 2016 going into 2017.

I'm tempted now to buy that second 970 I have been itching about.

Wondered if others are still going to wait at this point.

970 is definitely a good choice for the next 1-1.5 years at least, especially at its justifiable price tag.

NVidia indeed officially announced the first Pascal chip GP100 yesterday, showing samples at GTC2016, but not even one working card yet. The GP100 with Pascal architecture supports 3584 cuda cores, 16GByte HBM2 memory, fast NVLink interconnect, produced in in 16nm FinFET+-

BUT following Jonah Alben, NVidia chief architecturer for that chip, the GP100 is not thought for consumer or gamer cards, but the pros and supercomputing first, especially HPC and deep learning - and the price tag most probably will reflect that. You can expect first cards for that later this year, and some manufacturers might offer cards with GP100 for consumer/gamers later, but rumors say these will only use GDDR5/X then instead of HBM2 and probably cut back on (for gamers) less important double precision calculations also.

What annoys me a lot more is, that despite using the 16nm it is a huge chip , with 610mm² even larger than the Maxwell with 601mm² (I think that's used on 6GB GTX980 and/or Titan, if I remember correct) and it will consume even MORE power, not less. That means while a GTX970 usually is rendering at ~150Watts and the GTX980 at ~250Watts, the new GP100 Pascal is announced with 300Watts.

Btw. CUDA 8.0 will support Pascal, it should be available in June.

Tesla GP100 cards should be available to research facilities and cloud provider later in 2016, but the GP100 chip should not be available to OEM manufacturers for cards until first quarter 2017, so don't expect gamer cards before second quarter to mid 2017. This is info from Jen-Hsun Huangs presentation of Pascal architecture on GTC2016, who is NVidia boss I believe.

I don't think there's any reason to wait for Pascal if you want to render NOW, maybe they get affordable in two years or so, when less high-end chip versions for consumer market have been announced. As it is now I don't expect Pascal cards for under $1000 for the next 1.5 years at least.. but good luck waiting ;)

 

Interesting thing is that AMD have already put a HBM based card out for the normal consumer in the R9 Nano, you would of thought that Nivida would do the same with their equivalent model..