OTOY Forums

I know it's a dual card settup, but it's 4gb. 2gb are completly going to waste. I bought it because it was 4 gb. Why can't Octane use ALL 4gb?? I feel like I spent $1,000 and only get half the use out of the card. Really frustrated. What can I do??? UGH, hope there is a fix for this soon, Thanks.

Randall.

This is not an issue with Octane. That 4GB gets divided by 2 because the GTX690 is a dual GPU card. To be clear: Only 2GB is available per GPU.

A GTX 780Ti w/ 3GB VRAM or a GTX Titan w/ 6GB VRAM would have been better choices.

- issue is probably with all GPU computing
- in essence octane pushes the same scene to every card and then it renders in a sort of split screen
- anyway that was obvious in the first version of multi gpu support where there were discrepancies between split screens, which works differently now, but probably the same issue

- think of having one GPU as a separate computer in a network, showing results on one screen, it would be too slow to access both computers memory

- basically it is because of the PCIe speed, it's the network in this case, if the transfer speed get the same speed at which your computer RAM operates, you could use you computers RAM then - and CPU would be also used without that speed limitation -

- I am guessing when the computer architecture changes so the PCI speeds become as fast as the memory speeds then there wouldn't be any memory or CPU+GPU restrictions as they are now.. it's up to the mainboard manufacturers and what is worth to them - basically because most of the GPUs are used for gaming most manufacturers aim it towards that, and that is good enough for them, for us not so..

- the amount of time it takes from when you first press render on octane - until it shows up on the screen is the time it takes it to load the scene onto the GPU.. that's not that fast, and gamers don't care, maybe it's too expensive to manufacture faster PCI buses, so it could take a few generations, it's always the bottlenecks, system runs as fast as it's slowest component

So, if it was a single gpu with 4gb of ram, Octane would recognize all 4gb? Great, so I have a $1000 paperweight. I am always getting low memory warnings. My scenes are not that populated, but I do use alot of hi res textures. So, what's the point of a dual gpu? you only get 2gbs of use right? Even in gaming.

Why can't the GPU ram be additive? Is there ANY way to get 4gb in Ocatne with this card?

To my mind the best option for a "fix" to this is region rendering split, where one gpu does 50% of the scene and the other gpu handles the other half.

However I'm sure it's not as simple as this, and possibly the two sides would converge slightly differently & some post work would be required?

Interesting but currently Octane does not support this. Just sucks, you see a card and it says 4gb, so I think, cool, I'll have 4gb of gpu ram available. NOT. My bad for not doing my home weork I guess. I just "ASSumed" Octane would use ALL 4gb. Lesson learned I guess.

Yes before dropping 1K on a GFX card it would help to do a bit more research. A simple forum search would have yielded plenty of recent discussions over the benefits of the GTX690 vs GTX Titan vs GTX 780 vs GTX 780Ti.

If you purchased the card retail why not return it even if it means taking a hit from the restocking fee?

The GTX 590 & 690 are effectively 2 cards in 1 - so 4GB of VRAM gets split into 2GB of VRAM per card. This is covered in the Octane FAQ http://render.otoy.com/forum/viewtopic.php?f=23&t=21372. A 690 is a very fast rendering solution - certainly not a "paperweight".

Paul

It's not Octane issue. It's devious marketing policy of these GPUs. As in real this 690 is just "two 680 GPUs in one box", each of these two has 2GB of own memory. Each GPU core can only work with its own memory, it can not use another's GPU memory, it is how NVidia GPUs work. So the memory of two (or any amount) different GPUs is used in parallel and must be thought as parallel memory chunks of each GPU.

It is the same as when you buy two (or four) one core GPUs (e.g. 680s with 2GB each) - you do not get 4 or 8 GB of VRAM in this case, but still the same 2GB of each GPU's memory.

But when you buy the same two 680s just "in one case" as "690" - you get fooled by NVidia marketing policy which states that "it has 4GB" which is not true, as NVidia GPUs can use this memory only in parallel, using its own 2GB chunk each. The same as if you buy just two 680s...

Here is nothing to do with Octane (or any other CUDA software). Here is something to do with novadays marketing people...