I've been using the Nvidia 970 which has proven to be a very unreliable GPU in substance painter 2. Something to do with 500mb of it's 4gb ram being absolutely useless. Anyway, I digress - Can anyone here recommend a better GPU with 4gb ram or more for £250 or lower? I've picked out a few myself but I want to hear what you guys say first before I buy something.
Edit - I should make it clear that the 970 is a fantastic card but when trying to view/work in 4k using substance painter, it constantly crashes for me.
Thanks.
Replies
http://www.videocardbenchmark.net/high_end_gpus.html
I'm curious if the increased memory bandwidth of the next generation cards (GDDR5x and HBM) will help with Painters speed.
Download GPU-Z and check if your VRAM goes to its limits. 3GB is a whole lot for a painting tool that operates on average on maybe a couple 4k textures and a 50k polygon model, even if you write and read them in real time, 3GB should be plenty. Those numbers are a bad joke to your GPU. Reading and writing a 4k texture every step should occupy somewhere around 200VRAM. But either way, GPU-Z will find out.
Also update your drivers.
Also, the 500MB thing is an issue for painting as it's access speed is super slow compared to the other 3.5GB. Once again, Allegorithmic has discussed this in quite a few places so I'm not going to repeat it.
They also have mentioned that the newer Nvidia drivers don't always solve performance issues, and in some cases, can make performance worse.
Everywhere I read about the 970 in substance painter, there is an issue listed with newer drivers (which are not recommended right now) or some other issue with it's Vram. Im going to try last years drivers (apparently they work better) and see if this issue im having persists, if it does, goodbye 970.
I also agree with ZacD, wait for the new line if you can.
VRAM can btw be rather cheaply aquired with AMD cards, where you gotta pay very big with Nvidia to get over 4GB, thats something to consider but careful with cuda requirements.
I believe I have found a work around for my situation. The issue lies with the alloted render time windows 7 allows the GPU inside games, applications etc (this is by default 2 seconds). Due to nvidia's poorly designed drivers (of late), the 970 often struggling to keep up with this time and when it misses the 2 second limit, windows immediately crashes the drivers and any programs using you GPU (Nvidia's fault not Microsoft's).
So how did I fix this? There are several ways, I edited some registry to "fix" this issue but I have just discovered there is a program on the official Nvidia forums that can do this work for you. This is only a temporary work around until Nvidia fix the issue.
Issue > http://nvidia.custhelp.com/app/answers/detail/a_id/3633
workaround > http://www.wagnardmobile.com/forums/viewtopic.php?f=8&t=8
So wait, calculating for more than 2 seconds makes the API terminate the application ?
Edit - I set the time to 30 seconds, which is more than enough, probably to much but it's a safe number.
@Shrike - I usually run GPU-Z while I'm painting. At the middle/end of an asset I'm usually hovering between 3.2 - 4gb
Just to add, i have a 970 with the second latest drivers and painter 2 runs fine
- You should avoid the 970 if you want to work at 4K with Substance Painter (1.x or 2.x). We did a lot of profiling on this GPU (I have one myself here at work) and it usually kills the performances as soon as you reach the last 500mb of VRam (which can happen quite fast on big projects). Usually, every computation done with those 500mb is 10 times slower than normal. That include the rendering of our 3D viewport too. I have an 980 Ti at home which, as you can imagine, performs much better.
- From what I have been able to see during these last month, AMD drivers broke more often than Nvidia, causing many rendering issues or even crash of our software. Nvidia drivers are not perfects, but generally more stable (at least for now). We work with both types of GPUs here, and even Intel integrated GPUs. We also do some support so we have some metrics on the subject.
- TDR happens indeed because the operating system assume that a computation needs to be done in a certain amount of seconds, and if not reset the driver because it thinks the program is blocked. However it's wrong to assume this is true for all applications. While for a real-time rendering application this is logical, it is not in our case. Sometimes you can't break a computation in batch and avoid reaching the 2 seconds limit. It even change depending of the GPU computation power. Breaking a computation in sub-computations to avoid a TDR would even give very bad performances in our case. That's why we usually recommend to increase the time limit of the TDR to something higher than the default value.
If you guys want more information you should read this page : https://support.allegorithmic.com/documentation/display/SPDOC/Performances
I hope that helps.