Hi,
I am currently looking to buy a new graphic card. I am eyeing on the Palit GTX 680, however I am not sure which one should I get, the 2 GB or 4 GB version. I don’t do gaming (very rare) and I’m not a gamer. My interest is mainly for Blender use.
Does Blender benefit at all with 4 GB ? Cycles? Or should I just get the 2 GB ?
I use dual monitor 24" and 22".
Would love to hear your opinion.
Thank you.
I’d go with 4GB. Any a little more complex scene quickly starts to use alot memory. It sucks later when you run out of memory on your 2GB card and wish you had bought the 4GB version.
Unless you play games the 6xx series is pretty much useless to you, the 5xx series have better performance in cycles so I’d go with a 580 with as much memory as you can get your hands on.
I’m currently using GTX 560 Ti 1 GB and find it slow… Why the 6xx series would be slow? Sorry hardware noob here. :spin:
Apparently Nvidia realized that their gaming cards were faster than their professional cards (quadro/tesla) in terms of GPGPU computing. This is bad for them as they want people to buy their professional cards for gpu rendering so they intentionally crippled the performance for such tasks in their gaming cards starting from the 6xx series and up.
Thanks NinthJake and Tame.
NinthJake,
I would love to read more about this. Where did you get that information? Please.
Thanks.
Just found this thread…
Interesting read.
This is a convenient assumption to make, but you don’t have the facts to back it up. The reality is that the Kepler architecture is different and that it doesn’t suit itself as well to some workloads as the previous generation.
Also, this issue only affects Cycles GPU rendering, where the 6xx cards would be about 30% slower than the respective 5xx cards. However you can get the 680 with 4GB whereas the 580 tops out at 3GB. GPU memory is only important if you want to do GPU rendering with Cycles: if your scene does not not fit into onboard memory, you cannot render.
Well giving the fact that Blender Internal Render development has been discontinued for about 2 years and all efforts have been diverted to Cycles development, I think it makes more sense to go with the 4 GB rams then, since Cycles is GPU based.
Yes it is different and yes it is not suitable for GPU rendering but why did they choose this architecture if not to prevent the use of “cheap” gaming cards from being used in a professional line of work which their much more expensive cards are marketed to? Especially considering that the Quadro K5000, 5000, 6000 and Tesla K10 cards are all based on the Kepler architecture.
I guess I don’t have any solid written proof on this but that would basically mean an Nvidia employee openly admitting that they are trying to milk you on your money…
Agreed, a very interesting read; sadly, we don’t know if he has a vested interest in professing what he’s discovered. It would be great if someone like Tom’s Hardware was to do a comparriosn - I could get behind that /nod.
I’d say they used this architecture because it is much more efficient for workloads that are relevant to games. Depending on the workload, a GTX6xx can be twice as fast as a GTX5xx even in compute applications. Also, Kepler is a two-stage architecture, GK104 (Geforce 6xx, Tesla K10) and GK110. The “full” Kepler for now is only available as the expensive Tesla K20. The Tesla K10 however is essentially two GK104. If people didn’t have the same complaints with the expensive K10 as with the Geforce 6xx, then you’d have a point - but as far as I heard, they do.
Nobody here has so far benchmarked either the K10 or the K20, so until that happens you can only speculate how much deliberate limits NVIDIA has put in there to not cannibalize their professional market. The GPGPU rendering market isn’t really that big. Tesla/Quadro GPUs do have exclusive features that are only relevant to supercomputer-style applications, but that’s not our problem.
Moved from “General Forums > Blender and CG Discussions” to “Support > Technical Support”