OTOY Forums

I know Radiance said accessing RAM instead of the GPU video memory was far slower. BUT, would it be possible to imagine you could choose which one to use?

Using RAM would allow users to use more geometry with higher resolution textures. Of course rendering would be slower but that would just make Octane an affordable alternative among other unbiased renderers. Considering the power of GPUs, I assume rendering would be slightly equivalent to using CPU. So having the choice would make sense to me. Would other people be interested or are users mostly interested in the gain of speed?

I'm interested in both large scenes and speed, but if I look at the complexity of the scenes that I have been able to render with only 512 MB of VRAM used at half capacity, I'm sure that graphic cards with 1.5GB, 2GB or 4GB will allow very comfortable work !

I belive 4gb will be just fine in most cases.

the problem is that it's just too slow.
it can take thousands of cycles while the GPU just sits there waiting for something to arrive over the PCI-e bus...

there is'nt much i can do about this, let's all hope GPU's will soon be available cheaply with far more ram.

Radiance

radiance wrote:the problem is that it's just too slow.
it can take thousands of cycles while the GPU just sits there waiting for something to arrive over the PCI-e bus...

there is'nt much i can do about this, let's all hope GPU's will soon be available cheaply with far more ram.

Radiance

OK now i'm curious how fast does the PCI-Express bus need to be before its practical to use system RAM? PCI Express 3.0 will give you 16 GB/s is that enough?(i'm guessing probably not)

I don't foresee the consumer cards getting 6GB of VRAM anytime soon, you'll need to get Quadro/Tesla for those and their extremely expensive(worth it if your a big company). 12 GB Quadro should be possible next year.

How much slower is it Radiance? Are we talking hours or minutes?

Cheers

Chris wrote:How much slower is it Radiance? Are we talking hours or minutes?

Cheers

It will probably be 2-3x as slow as using a CPU based renderer.

Radiance

I believe more GPU Ram will just be a matter of not too much time. Right now I'll be fine with 1GB of VRam, and when I know how to make use of and max out Octane (the Beta isn't even out and we already scream for more?), I'll worry about getting that 16GB GPU

One of the issues that you need to take into account is not just the speed of transfer between memory and the chip in question,
but also the access time.

Eg, the time it takes to send a message requesting a certain memory location's data until the first packet of data is actually returned.
This is already a problem that requires carefull programming on the GPU,
when requesting some data from the GPUs memory.

Once you start doing this for data that needs to go over a chain of 10 different devices, (even the CPU needs to coordinate access to it on the other end),
the actual lag for the whole transaction becomes too high, and the GPU will be waiting for data 99% of the time.

if this was not the case, GPUs would not have lots of VRAM, and would already be using the host system's RAM for games, etc...

Radiance

2gb is ideal 3gb is just right between cost and flexibility... just wait for the 3gb gtx 480 to arrive... ideal hardware for us mere mortals....