Any work arounds to Nvidias RAM monopoly?

[Rhodesy]

It seems to me that octane is restricted quite considerably as a production renderer for large scenes, not neccessarily by the speed of the GPU but by the amount of RAM you have on your GPU - which is obvious. Nvidia clearly have a large gaming market which currently doesnt demand large amounts of RAM and a scientific pro market that has the nicely speced cards but an unjustifiable price tag. Just looking at the 6GB 2070 which has a price tag of £3,500 for the same speed card as a GTX 480 (as far as i can see), just more RAM. You could buy 8 GTX 480s for that sort of money or a couple of top spec workstations!!

So it seems there is a big gap in the middle that many of us fall into, where we need more RAM but not a chance we can afford to spank that sort of money to get it. So we're pretty stuck trying to optimise our scenes which cuts creativity, eats time, incredibly frustrating and compromises the end result. I can't see nvidia bridging that gap anytime soon as they have just bumped up the price bracket of their Pro cards by quite a bit, and maybe if we wait another 18months for the next gen gamer card we might get 2GB RAM woohoo! seems to be a bit of a non starter for this type of tech which is sad. I've read previously that system ram cant be used as the access time is too slow but what about RAM disks? I dont know much about them but perhaps the likes of cubix would be wise in trying to crack that nut as that would be the ultimate coup.

Random post I know!

[radiance]

Hi,

We're focussing our efforts in minimizing GPU usage, that's all we can do until nvidia releases consumer cards with 3 or 6GB models.

Radiance

[tungee]

What about a petition to Nvidia?

[havensole]

The problem is that the Tesla's are designed for gpgpu purposes and many of the scientist using them can afford to drop that kind of money. The quadros would really be what you should look at, though you run into the same kind of problem there. Though they wont be as expensive as a tesla.

[radiance]

we need to try to get the OEM's to make them, eg EVGA, BFG etc...
unfortunately they won't respond to a petition of a few thousand people imo.
there needs to be a big market, to risk the research and production of the cards.

hooking up the memory should be doable as the chip can adress large amounts of memory.
i just think we need to have a bit of patience, and we've got some ideas that will make use of gpu memory much smaller in next releases.

Radiance

[Rhodesy]

I agree that the petition might fall on deaf ears at nvidia but could possibly be worth a shot if it was to the OEMs rather than nvidia as they could create a market all on thier own. However they probably dont want to tread on nvidias toes by undercutting them. Are the teslas the same as the GTX 480 but just with more RAM and possibly slightly different drivers? As for the market size I guess it just depends on the cost of the final card - im sure there are plenty of gamers out there who just want the most expensive, pimped up kit on the market so would add to our cause. If they could do a 3GB version along with any squeezing Radience n co can do at their end then we could be getting somewhere - or find some magic way to share memory between GPUs but im pretty sure thats impossible! The problem is at the moment, as everyone is aware, you are only as capable as your smallest card regardless or horsepower so it seems frustrating if you are in a position to invest in a few consumer cards but know A, that you will always be limited to what you can produce with those cards and B, when you do upgrade those cards will come redundant - unless you are rendering a smaller scene but if its that small you probably wont need the extra cards.

[radiance]

I agree that the petition might fall on deaf ears at nvidia but could possibly be worth a shot if it was to the OEMs rather than nvidia as they could create a market all on thier own. However they probably dont want to tread on nvidias toes by undercutting them. Are the teslas the same as the GTX 480 but just with more RAM and possibly slightly different drivers? As for the market size I guess it just depends on the cost of the final card - im sure there are plenty of gamers out there who just want the most expensive, pimped up kit on the market so would add to our cause. If they could do a 3GB version along with any squeezing Radience n co can do at their end then we could be getting somewhere - or find some magic way to share memory between GPUs but im pretty sure thats impossible! The problem is at the moment, as everyone is aware, you are only as capable as your smallest card regardless or horsepower so it seems frustrating if you are in a position to invest in a few consumer cards but know A, that you will always be limited to what you can produce with those cards and B, when you do upgrade those cards will come redundant - unless you are rendering a smaller scene but if its that small you probably wont need the extra cards.

I think that with the new cards coming out with 3GB ram you will be able to do some serious scenes.
Sticking 4096x4096 image textures on tiny objects won't help.
creative use of by then more flexible procedurals together with more optimizations on our end (and instances) will allow you to make some pretty complex scenes in a few months.
I think the time of 'convergence' of capability between what you can produce now with octane vs a cpu based engine will probably happen before the end of this year.

Radiance

[ilac]

I think that with the new cards coming out with 3GB ram you will be able to do some serious scenes.

Hi radiance,

This is the first I've heard of the 3GB video cards! That's great news.
Did your source give you any indication of when these are likely to be available? I was planning on buying a new card but if these cards are only a couple months or so away then it might be worth waiting a bit longer!

Thanks

[Rhodesy]

Yeah lets hope there are some 'affordable' 3GB cards on the market this year. Prudent use of high res textures is also a must and saving a few different res versions of your tex library might be time well spent. However with the interactive experimental qualities for camera placement with octane you might end up closer to objects you thought were going to be low res but are actually right in your foreground.

I wonder if there could be a way for octane to resample your textures for the final large render kind of like MIP sampling? obviously you couldnt do this interactively but if you set up your camera and scene in the smaller preview window using medium to low res tex versions that sit in a folder so they all fit in RAM then octane processes the high res versions you have in another folder - or same folder when it came to final output and optimised each texture through its relative distance to the camera and filled the RAM you had available. This would only work for stills but that is when you need the high res tex anyway - animations are only going to be outputting smaller frame sizes so no need for high res tex. This would take up processing time upfront but if its possible that would be a great trade off for me.

[GeorgoSK]

Tesla isn't only more ram, that ram is ECC, as for driver, you're also paying for development(funny not ? ) and almost hotline support, chips are better chosen from waffer,....lot of small things, to make them run stable and error-free in supercomputers. They obviously weren't intended for us
Quadro a bit more yes...but it isn't cheaper the last time I checked...
I think nVidia isn't planning any new release this year either, to help sell their fermis, just look how long it took them to put them on market !
OEM could be our chance...but it's lottery

I too want to wait, because no matter what you buy now, it will be useless with next generation (mainly because smaller ram--) bottleneck), but we can wait for a long time, and result is really unsure.

There must be some way of using the system ram in the future, either through OpenCL or Caustic's OpenRL, but I am no developer..