Nvidia vs amd for cycles?

Sorry if this has been mention before Im just trying to get some clear answers on what card is best for cycles before I fork over a ton of money for one. I was reading up on the post john lancaster made a few posts below this one where he’s testing the R9-390x. But I’m still unclear as to which brand is better for faster renders. cuda cores or stream processors?

From what I keep hearing gtx 5xx cards are better for cycles. but theres other cards out now that have many more cores. Especially amd’s which have twice the amount as nvidia cards but apparently theyre not as powerful core for core.

So Im just looking for some insight before I spend the money.
Also, not to sound all noob, but running 2 of the same card will give me double the power correct? I would have to upgrade to a better psu but thats something Im considering.

Thanks for any advice or tips. all is appreciated.

It’s up to the developers and AMD themselves, pretty much the same way it’s up to Lux devs and Nvidia on the Luxrender side of things.

You could buy an AMD card now and don’t get proper support or even Nvidia current levels for years.

You shouldn’t compare cores as they’re different for every architecture. You should compare performance and all around support as I highly doubt that you will be doing just that with your card.

If you ask me, Nvidia is still the way to go until some serious improvements are done on AMD’s side.

For now nvidia is better. AMDs/OpenCL support was abbadoned for very long time, and it will took some? time for chasing CUDA level of support.



source: http://www.blender.org/manual/render/cycles/features.html

And about performance.
Cores don’t influence directly on performance cause it is different architecture.
Only way to compare them is by benchmarking (and ofc how much power do they need and how many RAM they have) but it is also not very reliable since a lot can change.

My R9-390x does the BMW2.7 benchmark in 52.66-54 seconds.
AMD cards seem to like the largest possible tile size.

Plus this 390x has 8GB which is pretty great for rendering.

I just changed my graphics card to Nvidia and it’s a joy.
you get open cl support and you get Cuda support so your’e not giving up on anything(maybe some money but how often do you change your graphics card?)
I had enough waiting for AMD to fix their drivers and software to start supporting AMD but it’s just stuck for too many years

Even if blender will work with AMD without any issues you still need other software (at least once in a while) and nvidia is much safer option.

Moved from “General Forums > Blender and CG Discussions” to “Support > Technical Support”

In this list, branched path is really slow on GPU like volume (and most of the time, volume is really easy to fake). Transparent shadows are active in buildbot. The BF and AMD are working on the rest. Note that the experimental kernel isn’t really usable because it eats your memory and you have not a lot left for your scene. So in the end, Nvidia is better if you want non-faked volume in most of your scene (for once in a while, CPU is ok and sometime even faster) and you don’t have heavy scenes requiring a lot of ram.

Thanks for the replies everybody. Im thinking 2 nvidia cards might be the way to go. I didnt know all that but Im glad I do now.

thanks john. But what are tiles?

Normally when you render it splits it up into several “boxed” areas or tiles, these affect memory usage and speed. My R9-390x seems to do best when it renders the whole thing in 1 “tile”.

If you’re going Nvidia then a pair of 970s or 980s would be the way to go. 980tis if you have the money.

2 970s are about equal in performance (a bit more for blender) to a 980ti.

ok that makes sense. thanks john and everybody else. and thanks for the link matali.

OK OK OK But if that all thoses things guys said is pure truth there is ALSO a big part that can change final perforance level you considering the brand you choose. I’am in professional archiviz and i’ve the habbit of heavy scenes rendered on GPU onboard Memory.Last past month i go for 2XGTX 970 while keeping my hold HD 7970 in an other PC waiting for opencl fix in blender cycles as many user !
BUT since last AMD opencl fix i’ve experimented some custom benchs that change all for me !

pavilion barcelona : 100 samples

7970 : 58 sc / 100 sample [B]tile size= 512

970 : 92 sc / 100 sample tile size= 256

and i test some my own big commercial scene on both 7970 an one 970 the OLD 7970 continue to outperforms the 7970 ! And i’ve the same result i use LDR envmaps for both and the final result is simply the same !

so what i want to say is that you have to choose but considering the final goal and sacrifice you do by choosing this one or this other one (AMD or NVIDIA).

For Nvidia:

GTX 580 / GTX 970 / GTX Titan similar performance but very different price : But support sss/transparent shadows/ etc

But here you get lower speed but with all features (but for me SSS is rarely used and can be greatly faked/ Transparent shadows is cool feature but not really noticeable in most of rendering/ and for HDR .LDR work great and save lot of memory on both NVIDIA an AMD).

For AMD :

7970 beat by 50% gtx 970 / GTX titan

R9 390 ====>> 2x GTX 970 or 1xTITAN X
8Gb ====>> 2x 4Gb or 12 GB - 4GB
350 $ ====>> 700 $ or 1000 $

so now buying 2x R9 390 is like buying two TITANX at 700$ instead of 2000$

in addition support for all thoses missing features are coming in next months !

Now i change my two 970 for two R9 390 nitro from sapphir.
[/B]

For the moment AMD cards can’t render volume materials and can’t GPU compute .hdr enviromnent map.