[wireframe]

Do you feel like a next-gen card must have 512MB of onboard video memory? Are the card manufacturers just dragging their feet and milking by keeping it at 256MB even for the top of the range? Share your view on this and tell everyone why you think 512MB should be implemented as a new top of the line standard or why 256MB is still enough.

[digitalwanderer]

It depends. If the 512Mb & 256Mb models have the same speed memory I'd get the 512Mb, but I've seen too many cards loaded up with slow memory just for the PR bragging points.

[wireframe]

Because I am awful at these types of things, I should clarify what I am asking. I am not looking for an argument of semantics. I am asking what YOU consider "next-gen". Not what the manufacturers are dictating to you.

For example, although I think the 7800 GTX is a step up, I don't consider it a good one for me because I expect my next card to have 512MB of memory. First of all, it has already been demonstrated that the NV40/R420 generation can benefit from this increased memory capacity. Secondly, I don't want to be stuck at 256MB-land forever. I want the cards out there so developers will make something for them. If you put those two together, you will see that they/we are running out of space (or you would get zero performance benefit). I think it's high time we all moved on to 512MB so when we pay through our nose for a "next-gen" card we can actually support next-gen graphics.

And to answer the first question: Of course I am assuming similar/or better speeds here. I am not looking for a "spot the loophole" argument. Ceteris Paribus...all other things equal...just MORE of it. Just like we moved from 128MB to 256MB. More and faster.

[Moloch]

Right now I think 256 is enough with pci-express.
Hl2 may benefit from 512MB, but it doesn't NEED 512MB, and pci-express helps frame rate a bit looking at techreport's benching.
Maybe for the super 800 doller cards, but you don't need 512MB for any game, even at 1600x1200 4x aa.
Maybe if have a dell 24" it makes enough difference to where it's unplayable with 256MB card, but atm with 1600x1200 being the "standard", no.
But for next gen games, I should hope so.

[wireframe]

Originally Posted by radeonic2

Right now I think 256 is enough with pci-express.
Hl2 may benefit from 512MB, but it doesn't NEED 512MB, and pci-express helps frame rate a bit looking at techreport's benching.
Maybe for the super 800 doller cards, but you don't need 512MB for any game, even at 1600x1200 4x aa.
Maybe if have a dell 24" it makes enough difference to where it's unplayable with 256MB card, but atm with 1600x1200 being the "standard", no.
But for next gen games, I should hope so.

So what exactly is your position? You say "right now" and "next-gen"...this thread is about next-gen...as suggested by the title. Either you think 256MB is good enough: we stay with the textures and resolutions (AA) we have today and play around with shaders or you think we need more concurrent assets and shaders aren't gonna solve all our problems.

The question is obviously open to everyone's interpretation, but for discussion I would say that at 1600*1200 and AA we are already running out of space with the "crappy" assets we have today.

[Moloch]

Originally Posted by wireframe

So what exactly is your position? You say "right now" and "next-gen"...this thread is about next-gen...as suggested by the title. Either you think 256MB is good enough: we stay with the textures and resolutions (AA) we have today and play around with shaders or you think we need more concurrent assets and shaders aren't gonna solve all our problems.

The question is obviously open to everyone's interpretation, but for discussion I would say that at 1600*1200 and AA we are already running out of space with the "crappy" assets we have today.

For games right now, it's not needed IMO, but for the next gen games, it will be.
unreal 3 ships in 06 right?
I think the R520/ G70 refresh should have 512MB.
For right now, I think we should focus on shaders and lighting.
Far Cry to me is what I expect to see from games released in 2005, and better.
It has good textures, good shaders and lighting.
Doom 3 is the most visually impressive game to date imo, but it also is so dark it hides alot of the crappy textures, but for the type of game they were making, it's great.

[HighTest]

Personally I believe that we'll start to see another trend.

Cards with 256MB or so buffer with TurboCache technology ala PCI-E 6200 Turbo Edition.

It's far easier for users to add memory to their motherboard and then using PCI-E and a caching method receive very good performance. Seriously, look at the benchies of the 6200 in regard of it's core speed, number of pipelines, etc. Impressive if you consider it in comparison to it's features (don't go and compare it to a XT850XT PE, take it in perspective).

Both ATI and nVIDIA could then focus on features/processing performance and save costs on memory by not going too insane on the amount loaded on the board.

[Chalnoth]

Next-gen will be determined by technology, not by memory sizes. A GeForce 6600 GT 128MB, for instance, is still a generation beyond the GeForce FX 5900 Ultra 256MB.

With PCI express and the usage of technologies like turbocache, though, there isn't a whole lot of reason for huge amounts of onboard memory, not until monitors catch up, at least.

[ANova]

I personally think the 512 MB 6800 Ultras and X800 XLs were a bit premature. It won't be until next year that we start to see games that will be able to take any kind of advantage from that amount of memory. I'll agree with Chalnoth and Hightest in that PCIE and turbocache is a better and cheaper way to go as well, though I wouldn't argue with a 512 MB R580.

[Skinner]

A big yes, its time move to larger textures and alltough 512 is overkill right now, it allows devs to finally support bigger textures.

[MasterBaiter]

Originally Posted by Skinner

A big yes, its time move to larger textures and alltough 512 is overkill right now, it allows devs to finally support bigger textures.

I'm pretty much in agreeance with that. Not so much for todays games, but definitely for tommorows. Can't program for, and create higher quality textures & effects if the existing hardware doesn't support it...

I think the turbo-cache thing could be an interesting way to leverage memory, but I haven't bothered to look into it. Perhaps when we see a move to 65 micron shrink we'll start seeing large frame-buffers as well ( & dual-core gpus to boot...) :P

Right now I"m pretty happy running @ 1600x1200 with everything cranked up to 11, but I look @ the games themselves now as more of a bottleneck then the system itself. :?

[poopypoo]

Although "all other things" (price included???) being equal, sure, i wouldn't argue with more memory, it seems like the 6600GT is good proof that it's more important to have a good design with good clocks than a given amount of memory. That said, by waiting a year or so to buy my next-generation card, I'll allow myself the opportunity to see what the next batch of "big" games requires before deciding. I generally buy cards to suit the games i want to play right now, not "next-gen" stuff to try and predict the future, and price is a bigger factor to me than frame rates at 1600x1200 w/ultra-mega-omega-high texture set.

[Albuquerque]

The thought that comes into my mind in "next gen" games is huge floating point textures (HDR anyone?), lots of displacement / normal maps, very high resolutions, and high levels of AA both SSAA and MSAA. This all sounds like an all-you-can-eat buffet of both memory capacity and bandwidth.

The true "next gen" cards are going to need 512mb of ram in my personal opinion. That doesn't that everyone will run out and grab them, or that they'll even be affordable, but it's going to come to pass sooner rather than later.

[Anon Lamer]

BF2 requires over 300 meg video ram at higher settings. Turbo Cache = bandwidth of about 6.4 Gig /sec, video ram = bandwidth of about 35-40 Gig sec = no contest.

[Albuquerque]

Well now wait on that whole turbo cache = lower speed and therefore = sux0r thought process.

One of the things that I know Carmack has mentioned and PowerVR implemented in their hardware was an ordered texture memory paging scheme. An example given was "over 300mb of video ram usage", but does it really need ALL 300 megs of video ram to complete that scene? Think before you answer "yes", because it likely doesn't. Lots of those textures are likely not being fully displayed, there are lots of places where only 1/4 of a texture is being displayed...

A texture paging system would allow for non-local textures to be tranferred in block-incremets rather than entire-texture-at-a-time pieces. This would allow you to load just the pieces of the texture that you need into local video ram, rather than the whole texture. Further, even if you do overflow the local video ram and must start going to system ram, you're transferring over the PCIe bus only the pieces you need and (again) not entire textures. A 32-bit 4096x4096 texture is going to be a slow transfer, but if you could read that texture in 512x512 blocks and you needed only 780 x 340 pixels, you would've used only 1/32nd of the original bandwidth (two 512x512 chunks versus one huge 4096x4096 chunk)

PVR already has hardware and drivers that do this; it's likely that ATI and/or NV could begin using this as well to drastically reduce memory bandwidth utilization (both local and over the PCIe bus) by implementing such a thing.