Blender GFX Card

[Superguy4ever]

Well, i had a question since u guys sound like you know a lot better about blender than me (im still learning it). Anyways, when rendering complicated models (ex of a really big home with all the details (down to the smallest curve on a flower vase) or when rendering animations, does GPU core speed influence powerformance? Like i said im on a really tight budget and so i wanted to know, will getting a gtx 1060 (going $50 over budget) influence my performance much or not really. I could either get a gigabyte gtx 970 xtreme gaming (very good overclocking card with a 10+2 digital power delivery system) and get the advantage of having over 1600 Cuda cores but lose out on the extra 2GB of ddr5 memory (and slightly faster memory) or i could get an msi gtx 1060 for an extra $50 and get the advantage of having an extra 2GB of the ddr5 and get the 8000Mhz memory and get the much faster GPU core but lose out on about 500 Cuda cores. If u can help me choose i would be very grateful.

[Clanscorpia]

Just because something has high clocks doesnt mean it will perform good (FX line) but generally for a modern day GPU the higher the clock speed the better it will perform if you compare to a GPU of the same chip with lower clocks

[BrinkGG]

5 minutes ago, Superguy4ever said:

Well, i had a question since u guys sound like you know a lot better about blender than me (im still learning it). Anyways, when rendering complicated models (ex of a really big home with all the details (down to the smallest curve on a flower vase) or when rendering animations, does GPU core speed influence powerformance? Like i said im on a really tight budget and so i wanted to know, will getting a gtx 1060 (going $50 over budget) influence my performance much or not really. I could either get a gigabyte gtx 970 xtreme gaming (very good overclocking card with a 10+2 digital power delivery system) and get the advantage of having over 1600 Cuda cores but lose out on the extra 2GB of ddr5 memory (and slightly faster memory) or i could get an msi gtx 1060 for an extra $50 and get the advantage of having an extra 2GB of the ddr5 and get the 8000Mhz memory and get the much faster GPU core but lose out on about 500 Cuda cores. If u can help me choose i would be very grateful.

Just get the best that fits your budget. Just make sure GPU rendering is actually enabled. 

https://www.blender.org/manual/render/cycles/gpu_rendering.html

[coconuts4eva]

Good question. I would stay away from a super low end card, but any card 750 Ti or better will really speed up the process compared to a quad core CPU. 

 

For blender the rendering seems to be better optimized for Cuda instead of openCL. openCL is getting better.

 

Here is a video comparing a 750 Ti render time to a 12 core Xeon: 

 

 

And here is one where they through in two R9 390s. : 

 

Of course none of this matters if you are trying to render out a bigger set than your VRAM. The 1060 or R9 390 would probably be two affordable 8Gb cards.

[Superguy4ever]

1 hour ago, coconuts4eva said:

Good question. I would stay away from a super low end card, but any card 750 Ti or better will really speed up the process compared to a quad core CPU. 

 

For blender the rendering seems to be better optimized for Cuda instead of openCL. openCL is getting better.

 

Here is a video comparing a 750 Ti render time to a 12 core Xeon: 

 

 

And here is one where they through in two R9 390s. : 

 

Of course none of this matters if you are trying to render out a bigger set than your VRAM. The 1060 or R9 390 would probably be two affordable 8Gb cards.

But do you think its worth it to spend $50 on an extra 2GB of ddr5 VRAM (to make a total of 6GB). I mean, i heard cuda cores are the main factor related to render speed (more cuda cores = faster render). In that case, i would consider the gtx 970 a better deal (even tho it has lower VRAM) as it has about 400 more cuda cores which will equal to a faster render (plus its $50 cheaper). Also, i had a question (not really related to this topic but u sound smart). I am building a new PC for game coding with unreal engine 3 or unity, 3D rendering and some gaming. I could either get an 8 core haswell-e cpu (xeon e3-1231v3) for $250 or i could get a quad core skylake cpu (i5 6400) for $190 (possibly enabling me to get a gtx 1060 or maybe an r9 390 for the same budget but im not sure if i should sacrifice the 8 core cpu (with the advantage of future-proofing since certain computing programs get a good advantage out of those extra 4 cores) for a better gpu that could live or be more future proof. I am choosing the parts very carefully to make sure that the pc could live for up to 5 years (with an upgrade up to $400 when the time becomes right). Thank you in advance.

[coconuts4eva]

Here is the tradeoff. If you are working on a big project and don't have enough VRAM your project can crash. If you have more VRAM and less Cuda cores or openCL processors you will still be able to finish, it will just be a little slower.

Also in general Cuda cores = faster render, BUT that isn't true architecture to architecture. Your Cuda cores are similar to a CPU in that your particular cards Cuda cores may have a different number of instructions per second that it can handle depending on generation. You need to look at the TFlops rating on the card to find out just how good it is at parallel computation.

It can take a while to package and export games on Unity and Unreal. I've found that you can do this a good bit faster with more cores, but for the money probably not much difference with what you are talking about. Both will last the 5 years, but I'd say both would be like a snail by then. I wouldn't worry about beating the curve, just go for it. By the time 5 years passes you will either be interested enough that you will have upgraded or you will have just thrown in the towel. 

[AdamH96]

Just a heads up... currently in Blender the pascal architecture is not supported by the official release because of the new CUDA 8 Dev. kit (so rendering with the GPU will result in a crash). So to enable the 1060 you have to find a daily build to be able to run the new cards. So if you are trying to find a benchmark for the 1060 in blender remember those results can change when Blender implements the new CUDA kit for pascal.

[Superguy4ever]

 

13 hours ago, coconuts4eva said:

Here is the tradeoff. If you are working on a big project and don't have enough VRAM your project can crash. If you have more VRAM and less Cuda cores or openCL processors you will still be able to finish, it will just be a little slower.

Also in general Cuda cores = faster render, BUT that isn't true architecture to architecture. Your Cuda cores are similar to a CPU in that your particular cards Cuda cores may have a different number of instructions per second that it can handle depending on generation. You need to look at the TFlops rating on the card to find out just how good it is at parallel computation.

It can take a while to package and export games on Unity and Unreal. I've found that you can do this a good bit faster with more cores, but for the money probably not much difference with what you are talking about. Both will last the 5 years, but I'd say both would be like a snail by then. I wouldn't worry about beating the curve, just go for it. By the time 5 years passes you will either be interested enough that you will have upgraded or you will have just thrown in the towel. 

Well, i think i might get myself an r9 390 since i heard OpenCL compatibility with Blender is improving and i want the 8GBs of VRAM since i might get a 4K screen with the PC. Anyways, ill get a b150 mobo (for like $50), an i5 6400 (for $180) saving $70 over the 8-core xeon since i dont really need those extra 4 cores (ill mostly be using gpu accelerated programs so I'd rather spend more money on a better GPU). I just wanted to know, what features are missing on OpenCL compared to Cuda and how much do they affect rendering and performace. Thank you for all your help and have a nice day/night.