Memory capacity need on future high-end graphics boards.
- Thread starter MistaPi
- Start date
Yes, but I would like to know as to why.
- sysram is not a option for (large) textures that is accessed frequently?
- sysram not a option for vertex buffer for high-end gfx cards?
- higher modes of AA an FP precision increases framebuffer size.
All of the above? Something else? Please elaborate.
- sysram is not a option for (large) textures that is accessed frequently?
- sysram not a option for vertex buffer for high-end gfx cards?
- higher modes of AA an FP precision increases framebuffer size.
All of the above? Something else? Please elaborate.
Dave B(TotalVR)
Regular
Well I think with PCIe, now is the time to really start deciding what there is that can be stored in local memory. Perhaps a number of very large textures you can determine are accessed very infrequently, but predictably, so perhaps store them in local memory and precache them near the time to render.
Dunno, could be very very scene dependant as to what could be chucked off-card.
Dunno, could be very very scene dependant as to what could be chucked off-card.
We had pretty much the same situation when AGP appeared. Back then, they also wanted to put stuff in system RAM until they realized that this solution is so much slower and more complicated that it's worthless. The same thing will happen with PCIe.
It is a lot faster than AGP, at least theoretically, but the bandwidth requirements are also higher. So I say this will end the same way as with AGP.
MasterBaiter
Regular
I'm with Dave B on this one. PCIe should open some doors, but the speed of system memory is still a step back from what is being used on video cards. I guess this is mildly comparable to cpu's using L1, L2, and L3 (and even L4) cache in a way. :?
-edit- Perhaps there could be an option for future motherboards to have a memory slot designed specifically for additional graphics memory.
-edit- Perhaps there could be an option for future motherboards to have a memory slot designed specifically for additional graphics memory.
why not just implement some sort of streaming for the single player games? (like lots of console games)
-reduced load times, less total memory required...
Halo PC has some of the biggest levels I've ever walked through yet the load times are roughly 3 seconds... and then there's only the 1 second for loading the new section of the level as you go.
-reduced load times, less total memory required...
Halo PC has some of the biggest levels I've ever walked through yet the load times are roughly 3 seconds... and then there's only the 1 second for loading the new section of the level as you go.
I don't think PCI-E will have any impact on the amount of memory on high-end cards. Sure it is useful in the low-end, especially with Longhorn and virtualized resources on the horizon.
Low bandwidth and high latency compared to local memory on high-end cards means it's only suitable for a few uses, e.g. vertex buffers. But those aren't that big anyway.
Of course getting a few GB/s additional bandwidth "for free" is a nice thing, but the "heavy stuff" (framebuffer and texture accesses) you want to happen locally.
Low bandwidth and high latency compared to local memory on high-end cards means it's only suitable for a few uses, e.g. vertex buffers. But those aren't that big anyway.
Of course getting a few GB/s additional bandwidth "for free" is a nice thing, but the "heavy stuff" (framebuffer and texture accesses) you want to happen locally.
Probably not, you need the best electrical connections you can get for the sorts of speeds that memory chips utilize. Plus the amount of people that would even consider upgrading videomemory isn't that great either.MasterBaiter said:
On the subject of PCI-E's added bandwidth, that is probably going to equate to more available bandwidth to swap huge textures from main memory.
How about...
**DIM LIGHTS..... HORROR MOVIE MUSIC.... WINDS WHISTLING AWAY.... BOLTS OF LIGHTNING STRIKE***
...Embedded RAM?
**DIM LIGHTS..... HORROR MOVIE MUSIC.... WINDS WHISTLING AWAY.... BOLTS OF LIGHTNING STRIKE***
...Embedded RAM?
Dave B(TotalVR)
Regular
Well dont forget that an increased use of procedural textures will reduce the required texture footprint.
Once you can get themn to look like what you want them to look like, Im sure they will handle LOD, aliasing and blurring up close much better (as in no problem) than a standard texture map.
I think at the very least developers should be considering a hybrid situation, like procedurally generated detail texture or normal map. Particularly normal maps because low detail normal maps really do seem pointless to me.
Once you can get themn to look like what you want them to look like, Im sure they will handle LOD, aliasing and blurring up close much better (as in no problem) than a standard texture map.
I think at the very least developers should be considering a hybrid situation, like procedurally generated detail texture or normal map. Particularly normal maps because low detail normal maps really do seem pointless to me.
squarewithin
Newcomer
If anything, they use it on top of conventional texturing to hide such things as artifacts that show up from texture tiling that couldn't be removed with a low pass filter exactly, or such things as addition details when reaching the top end of texture magnification.
I know of several movies and (and Riven) that have used it for such.