does now the nvidia 600 series card get a lot of improvements, so it can render faster than the 500 series (on the same level <like gtx560 with gtx660 and so on>)?
No. The only exception is the Titan, but if you compare the current line-up to the numerical predecessor, you get worse GPGPU-performance. There might be exceptions with low-end cards but they don´t serve any purpose anyway.
time to stock up on GTX 500 models before they are sold out 
I am happy with my MIS GTX 570 which is crazily silent, and got it for 260 but now I see 570s for around 100 to 200 $.
So sad, I like how NVidia put so many CUDA processor. At first, I thought that it will be amazing. But after read the posts about that… err
What? so cheap! When I want to by 570, it’s around 350$ -_- just a year ago…
okay, so if there is any significant progress, please make me stay tuned! Thank you!
but they don´t serve any purpose anyway.
The only problem I see with buying these older card is that they seem to only come in the 1GB memory configuration. I have found that Cycles can not render a single object with a decent amount of displacement without erroring out because it has run out of texture memory. This forces me back into CPU mode which effectively makes me abandon Cycles completely. Who wants to wait on that when other render engines are faster in CPU mode?
For simple scenes, the low-end cards can work, but you may be setting yourself up for a surprise if you rely on Cycles for production work. At some point you are going to add in one more object or one more image map and bam, you will be forced into CPU mode.
If you are purchasing a new card, purchase for memory size, not speed. Even the slower cards will still be faster than CPU rendering.
Not only that, but most supported lowend cards ( GTX460 is not what I meant, I am talking about GT630/640-style cards) are actually slower than decent quad-cores. Thats why I said they hardly don´t serve any purpose- for Cycles rendering at least.
Hey Atom,
that can be very true when you do a lot of character work.
In my field with product design the 1.25GB is more than fine.
Using 2x 4GB GTX 680s. I do lots of gaming and the ram was good for cycles. My only issue for now is there are plenty of features that are CPU only without a GPU implementation yet. Of course I really got myself into pickle with a scene that has a few problems. It was an older scene that I kept working on and maybe something went corrupted in it along the way. When I tried CPU rendering console shows “Out of SVM stack space.” about 3 or 4 times and then Blender insta-crashes. When I go back to GPU rendering, it might show one or two “Out of SVM stack space.” messages but then renders perfectly fine. Another scene has the opposite problem (not too much ram or any experimental features). CPU works (6 core, 12 threads at 4GHz is still plenty slower than GPU), but then GPU causes a driver crash randomly, sometimes.
Kinda wondering if I should have gone the multi-processor route.
GPU: faster, sometimes more unstable (windows 7 with 680 anyway), less ram, less features.
CPU: slow as all get out, but 64GB of ram =D, more cycles features.
When I was using a 460, GPU rendering seemed a lot more stable. Only had 2GB ram though.
Sorry, my wallet doesn’t allow me because it’s too expensive 
This! Especially Strand and SSS! I need that so much.
up up up up
You won´t benefit from it anway, if you do not have at least a Fermi-onward semi-high end card. Just use the CPU.
That’s a good card for sure. I would recommend going with EVGA though. While I like MSI (and they have one of the fastest 570’s on the market), EVGA offers a life time warranty on any model with an AR at the end of it (as opposed to KR, which is 3 year warranty). Not only is EVGA a company based in California (if im not mistaken) they are much easier to get ahold of if anything goes wrong, where as MSI is a taiwanese company with very little support to be had.
Actually, I have a Zotac GTX560 from last year. But, I have a plan to upgrade it to 600 series. I heard that 600 series are slower than 500 series. So I asked here to know if 600 series are way faster than 500 series… If so, I’m going to upgrade immediately. But actually, no progress at all (except TITAN, but IT’S TOO DAMN EXPENSIVE).
I live in Asia, so MSI should be a good choice though…
up up up up
Hey since when blender internal got rendered preview viewport?? amazing!
My lovely feature in cycles, strand and subsurface are not developed well right?
GPU Strand Rendering is enabled now, but it can’t handle heavy scenes.
Subsurface on GPU is still black if I render it.
Yes, that’s my experience. I’ve got a GT630M, and it’s more or less the same speed as my i7 3830 quad core (8 logical) @2.4 GHz.
Must admit it’s a bit of a shame that GPU progress is so slow, but also understandable. I picked up a couple of Titans last month, rendering with them has been beautiful as the 6GB of VRAM can actually accommodate all my scenes now (was using 4x GTX 580 1.5GB before, memory was a major bottleneck). Really looking forward to the day when all the strand/particle work is available for GPU computation, hell I’d love to help implement these features but it’ll be years before my programming is up to scratch.
Try downloading a buildbot version, it wont be as stable as a normal release but it has gpu enabled hair rendering.
Hope that the GPU Strand rendering progress is not slow, because I have a project which has many strands (over 100K) and human which must use subsurface scattering.