Will upgrading my GPU help render faster with Iray?
Blastedlabs
Posts: 60
The first question I have is, how much does the GPU actually contribute to rendering with Iray? 30%? 60%?
Currently, my PC has one EVGA Geforce GTX 660ti. Last time I looked at graphic cards, they had stagnated, and a $300 graphics card was barely a 10% upgrade.
A random GFX card on Newegg that costs $300 compares in the following ways.
- 1050 Mhz (980 Mhz) +70
- 4 GB 512 bit (2 GB 192 bit) +2 GB
Also, if I did upgrade, would I be better off with 2x $150 SLI cards?
Thank you!
Comments
SLI won't work with Iray. I would recoend getting a card with at least 4GB. If the scene won't fit in the cards RAM the card won't be used.
Iray renders MUCH faster on GPU than CPU. A GPU can have thousands of CUDA cores all rendering in parallel.
Which $300 graphics card are you comparing?
I honestly don't remember. It was a Newegg sale item. Realistically, these are the two cards I'd probably look at based on current price and this new information. One is faster with 4mb memory, one is slower with 8mb memory.
GIGABYTE G1 Gaming GeForce GTX 970
http://www.newegg.com/Product/Product.aspx?Item=N82E16814125684
SAPPHIRE NITRO Radeon R9 390
http://www.newegg.com/Product/Product.aspx?Item=N82E16814202164
Wow! That is great information, ty. Is there a way to tell how large a render is? Is it just the size of the .duf file?
Iray is meant to be used with Nvidia GeForce cards. I just went from a GTX-750Ii to the GTX 970 w/ 4 gb, and tripled the CUDA and doubled the memory. I am extremely happy with the speed increase. MUCH faster. I use the CPU and GPU combination.
Dont; go Radeon if you are going to use IRAY.
Good luck
It's really not the "speed" of the card (e.g. clock speed or memeory speed) that matters most for Iray. Rendering speed is dependent on the number of CUDA cores. Two caveats: 1) if the scene doesn't fit in the VRAM, IRay defaults to CPU for the render 2) Whle Cuda cores are (sort off) shared by multiple cards, the VRAM is not. Thec scene must fit in each card individually or the card will not be used for the render.
And, as said before, don't use SLI for Iray. 4GB is probably the minimum memoery you want to get.
Sadly, currently there is no reliable way to tell how much VRAM the scene will need before the render.
The frequency of the cards won't tell you much. There have been huge changes in the hardware between the 660 and the 970!
The 970 is approx. 2 to 3 times faster in most things.
On the 'Compute' side the difference is even bigger with the 970 being 5 or 6 times faster than the 660.
The 970 is a good choice if you can't stretch your budget to a 980ti, which is the current sweet spot for price/performance/memory.
(There's nothing better than spending other people's money )
Only NVidia cards will help with IRAY rendering. 980ti is about the best (consumer card) atm - unless you go for a titan.
980 next best,
then 970, 960, 950
IMO 970 is a good concompromise cost/performance wise
Don't get less than 4GB or memory.
Make sure you have a good PSU and that it can cover the extra power required; that it has the required leads for the card.
Obviously make sure your motherboard can support the card, or the extra card if that is what you're doing.
Ha.
I added a weak little GTX 740 with 384 cuda cores. What used to take my cpu an hour and a half now takes about 20 mintues gpu only. Just saying for any broke daz artists out there.
I've been considering a 740 or 750 as a stopgap while I decide when to upgrade the whole system, good to know it can make an appreciable difference.
Interesting; but how much and what type of memory does the card have? I can't find a 740/750 in the local shops with more than 2GB, and that's not really enough for a proper full scene, or more than one figure. Memory type can also be an issue, GDDR5 is newer and faster than GDDR3.
I was using a GT640 with 2GB GDDR3; it was slower (just) than the CPU; it was whilst I was considering what to get, and once I'd got the 970, as I wanted to be sure before I got better, I kept it on as the monitor driver. It struggled at times,
http://www.scan.co.uk/products/4gb-evga-gt-740-sc-28nm-1334mhz-gddr3-gpu-1059mhz-384-cores-dvi-mhdmi
http://www.scan.co.uk/products/4gb-gigabyte-gtx-750ti-windforce-2x-oc-gpu-pcie-30-(x16)-5400mhz-gddr5-boost-clock-1137mhz-dl-dvi-i-
I got the one with 4GB GDDR5 http://www.amazon.com/EVGA-GeForce-GDDR5-Graphics-04G-P4-3748-KR/dp/B00KJGYOGG/ref=sr_1_1?ie=UTF8&qid=1452968820&sr=8-1&keywords=evga+gt+740+4gb
A GPU will help out tremendously. But be sure to get one with as close to 3,000 CUDA cores as possible, if you want to be able to use the viewport effectively for real-time rendering. Also, the number of GPU's seems to scale porportionally (at least in theory). In other words, two GPU's will typically provide almost twice the rendering speed as one. I have two GTX 780 Ti's, and I can usually manipulate my camera in the viewport in real-time, without having to wait for iRay during its black and white "I'm thinking" screen.
Out of curiosity, has anyone experimented with Daz 4.9 beta to see how much faster the latest version of iRay is?
.
What about the possibility of doing render passes?
And would it work for you to render different elements (eg. two different characters) and different times, as different images, then combine them in eg. Gimp?