Fusion 360 and GPU questions

[O.Tan]

Hi,

With all the new GPUs out in the market, I got curious. 

1. Even though this question is model dependant, how do we know how much GPU memory do we even need? 

2. Since iGPUs like the Intel Iris uses up to 1.5GB system RAM (is that the Max memory it can use or user is able to assign it?) I guess if a computer has 8GB memory, Fusion + iGPU will take around 5-6GB system RAM? 

Also, I found this slide http://www.slideshare.net/IntelSoftware/high-end-modeling-imaging-with-intel-iris-pro-graphics

The results admittedly is quite interesting

[brianrepp]

Hey Omar - I pinged one of our graphics experts who should be able to chime in with some insights when he is back online.

[chengyun.yang]

Hi,

With the current OS and GPU driver, the video memory is actually virtualized by OS. That is, even you only have 1 GB dedicated video memory, in theory the application may use up to 4GB(or even more) video memory by using some of the system memory for the GPU usage, depending on the OS and driver implementation. So from the application point of view, functionalitywise ideally it should not really care much about how much video memory there is actually in the card. The major difference is on the performance and how well the GPU driver implements the virtualized video memory behavior. The dedicated GPU memory is obviously much more efficient than the system memory and sometimes if the driver is not well implemented, you can find the application runs into some crashes caused by the "running out of video memory" issues. In Fusion, to deal with this kind of problems, we actually try to monitor the consumption of dedicated video memory. If the consumed dedicated video memory is close to the limit, then we will warn the user about that and then degrade some heavy visual effects automatically to reduce the video memory consumption. 

Does this answer your query to some extent?

Thanks

Chengyun

[O.Tan]

ah, thanks for the response. So I guess the allocation of GPU memory usage is dependant on how big or complex the model is? And it'll function like a RAM, where once the large assembly is closed and the user opened a small assembly, it'll purge the large assembly off from its graphics memory?

[chengyun.yang]

From the user's perspective, you can say basically it works that way. Internally we could use some memory pool so when you are done with the large model, those memory will be released to the memory pool. And next time, when you request the video memory, it will come from the memory pool. 

[O.Tan]

aaah, so the error of not enough memory comes when the particular model is so large/complex that it uses up all the memory pool right?

[chengyun.yang]

That is probably right. Besides the size of the model, the screen resolution and various visual effects are also the major factors that affect the video memory consumption. So you have a high resolution screen(e.g., 4K) and turn on the anti-aliasing effect at the same time, then the video memory consumption could be pretty significant. 

[O.Tan]

That clear things up. Thanks!