Usefulness of 512MB vram with X1900

[aaronbond]

put BF2 on Ultra for all in the video.con @ 12x10 w/ 2xAA and tell me you don't have +256 usage. I utilize that one german app to see usage. @ work and can't remember program name.

Also Chalnoth I'd like to know how you get your usage statistics? I have seen you report these b4 and wondered how you got them. Thanks. =D

 

[Belmontvedere]

I think 512MB will be alot more imporant going forward. Because X360 and PS3 both have 512MB of memory, it will pretty quickly become the "standard" for videogames.

 

[Pete]

Also Chalnoth I'd like to know how you get your usage statistics? I have seen you report these b4 and wondered how you got them. Thanks. =D

Let's see if I can save Chal some typing. AFAIK, n-sample MSAA framebuffer usage = front buffer + n(back buffer + stencil buffer). A buffer is simply pixels * bit depth.

Assuming 32bit depth for all buffers:

1280*1024*32b*(1B/8b) = 5,242,880B.
5242880B*(1KiB/1024B) = 5,120KiB.
5120kB*(1MiB/1024KiB) = 5MiB.

So, 6xMSAA @ 12x10: 5MiB + 6(5MiB + 5MiB) = 65MiB. 6x @ 16x12: 7.32MiB + 6(7.32MiB + 7.32MiB) ~= 95MiB.

(These #s are base 2. Xmas' are base 10, thus slightly larger. At least, I think they are, though his come in about two MBs higher than mine in either system.)

Edit: As Xmas has clarified, my #s may be base 2 but apparently my notation isn't. Corrected: kB -> KiB, MB -> MiB. My #s may also be off b/c I'm not taking into account downsampling, of which I'm clueless. In summation, .

I utilize that one german app to see usage. @ work and can't remember program name.

Yeah, IIRC Ail lists that app in his article. For those w/o .Net, ATI Tray Tools includes a similar app; maybe RivaTuner, too.

 

[Xmas]

(These #s are base 2. Xmas' are base 10, thus slightly larger. At least, I think they are, though his come in about two MBs higher than mine in either system.)

My numbers are MiB, thus base 2. But:

You might not want to downsample to the front buffer.

Although there are ways around this.

 

[Pete]

Thanks for the correction. In that case, my #s are actually MiB, b/c I divided by 1024. So, yours are higher b/c you're downsampling larger back buffers into a screen-size front buffer?

 

[Mabru]

Well FEAR stutters horribly on my 256MB X850 XT with texture settings on high, so there seems to already be a game or two that can make use of 512 MB.

The problem in FEAR is the main memory not the video one.

 

[aaronbond]

Let's see if I can save Chal some typing. AFAIK, n-sample MSAA framebuffer usage = front buffer + n(back buffer + stencil buffer). A buffer is simply pixels * bit depth.

Assuming 32bit depth for all buffers:

1280*1024*32b*(1B/8b) = 5,242,880B.
5242880B*(1kB/1024B) = 5,120kB.
5120kB*(1MB/1024kB) = 5MB.

So, 6xMSAA @ 12x10: 5MB + 6(5MB + 5MB) = 65MB. 6x @ 16x12: 7.32MB + 6(7.32MB + 7.32MB) ~= 95MB.

(These #s are base 2. Xmas' are base 10, thus slightly larger. At least, I think they are, though his come in about two MBs higher than mine in either system.)

Thanks alot Pete. I'll remember this formula in the future.

 

[Pete]

Sure thing, aaron. Just note the corrected binary prefixes.

 

[aaronbond]

Analysis this weekend w/ BF2 using Video Memory Watcher.

256MB 6800GT AGP
2GB DDR
All High
2xAA
@ 1280x960

playing Dalian Plant on 64 player server w/ about 10ppl.

.3MB available on board RAM
65MB available w/ aperture of 256MB in BIOS

So, correct me if I'm wrong, but in PCI-E it goes directly to main mem w/out having to worry about setting an aperture maximum within the BIOS?

If your calculations above apply Pete or Chalnoth why such a dramatic difference in my results from this? I'm not well versed in the ins/outs of graphics tech, but I'm guessing the mem is used for more than just texture caching, but why so much usage? Thanks.

Anyone have results from an actual 512MB card using Video Memory Watcher? I have yet to see this app used in a review. At first I thought it was a buggy app until I started seeing people in the forum talk about it. Thanks!

 

[Razor1]

mem is also used for shader caching and vertex positions awell.

 

[blackpawn]

the comments on needing the extra vidmem for vista are spot on i think. when you've got all the other windows competing for vidmem you're going to want a lot more vidmem. also consider the other applications in vista using 3d resources are not going to be written to be nearly as efficient as games are. we'll probably have IM clients and desktop widgets using 512x512 32bit textures when they could get away with 128x128 DXT5's etc.

have all that going on then try running your favorite game in windowed mode and it's gunna suck if you don't have enough vidmem for it all.

 

[Pete]

Well, the calculation we showed gives basically framebuffer usage for a given res and AA setting (~25MiB for 12x10 2xAA). I'm guessing the "dramatic difference" you're asking about is b/w your apparent usage (256MB card + 189MB sys) and the ~25MiB? Anything above ~25MiB is used by textures, vertices, etc. I'm not sure exactly how it breaks down, nor how accurate VMM is. Also, MP FPS do seem to use more VRAM than SP modes, for whatever reason (more player models?).

I don't know about PCIe and aperture.

Ail tested VRAM usage with VMM.

Edit: That's what I meant. Thanks, Ail and Chris.

 

[Ailuros]

I don't know about PCIe and aperture.

AFAIK drivers (irrelevant of IHV) assign as much system memory for texturing as there's GPU memory present (if that's what you meant).

 

[ChrisRay]

There really is no apature from my understanding of PCIE bus, the need for it is gone and as Ailuros said the drivers just assign the neccasary amount. It did cause a few problems for me back in the day when running Ruby on an AGP system but couldnt on a PCIE system without 3dc compression.