[Dave Baumann]

In our 512MB Radeon X800 XL review we ran several Doom 3 tests because id Software have long since spoken of Doom 3's "Ultra" mode requiring more than 500MB of video memory to operate effectively. However, our first look at a 512MB board we could see very little performance difference between a 256MB and 512MB board with either our own Doom 3: Resurrection of Evil test or the standard, pre-supplied demo1 test in Doom 3. We've queried this with id, asking how our performance numbers reconciles with their comments on the Ultra mode's memory requirements and id Programmer Brian Harris replied with the following:

"I'm not an expert on this but as far as I know, this is what's going on:

"A single scene doesn't use up all 512MB of video RAM. Having extra RAM will decrease hitches when moving from one area to another because the textures from the new area will already be on the video card. If a single scene is using more memory (textures + vertex data) than is available on the video card, then will you see significant frame rate drops because it has to upload and dump the textures multiple times as it's rendering. This happens in very few cases (we spent a lot of time minimizing it) but I believe it still happens in the "Turkey Baster Room" in Alpha Labs 2 (the room with the giant laser thing)."

The implications of this are that the visibility of the effects of the Ultra mode is going to be very dependant on how it was tested and the demo used. For instance, if the timedemo is fairly small, and/or limited to a single area, and/or not in an area where heavy texture swapping occurs (Brian does say that id have tried to minimise the effects and it now only occurs in a few places) then there is likely going to be very little performance difference between the Ultra mode and Normal mode in that demo as the textures that apply specifically to the areas that the timedemo occurs in may well fit within the local RAM available to 256MB graphics boards, thus there is little performance difference visible between 256MB and 512MB boards. If, however, a more sprawling demo is used, that does hit some areas of the game where some texture swapping does occur, then there may be less swapping occurring in boards with larger RAM quantities available, resulting in a higher performance.

We've updated the Radeon X800 XL 512MB review with a much larger demo recording from Alpha Labs 1 (which happens to be the area Brian is talking about, not Alpha Labs 2) to see if there were any relative performance differences between the 256MB and 512MB boards in Doom 3's Ultra Quality mode.

[Reverend]

This is not surprising (I'm actually surprised this made its way to a news post at B3D). I thought this was a known fact.

Think back to the times of the original GeForce and Quake3. And a certain Quake3 demo named "Quaver" made by me and how that demo became really popular with hardware review sites back then (for, like, about a year). Anyone remember why I made/distributed that demo?

[Joe DeFuria]

Thanks for the update, Dave. Nice to have a theory officially investigated and an open-ended question addressed.

Originally Posted by Reverend

Anyone remember why I made/distributed that demo?

Nope....I don't even remeber it being "popular" for that matter.

[Neeyik]

Revs talking about texture compression; I remember the timedemo but I have to confess I never used it much in reviewing - as I do now, I used my own demos back then.

As to the information concerning the amount of local memory taken up a D3 scene, I certainly do not think it is common knowledge at all; even if it were, it makes for good journalism to address queries made by many people (on other websites, as well as B3D) as to why there is such a difference in testing the 512MB cards.

[Jawed]

I think this experience intensifies the argument for including an evaluation of framerate minima in a benchmark. I know it's much harder to do... erm...

Jawed

[Neeyik]

Frame rate minima are too reliant on factors other than the graphics card to be useful in a review - well, declaring just the absolute min is no use; one really needs to look how frequently the frame rate is down that end. Better off doing standard deviation to be honest.

[BetrayerX]

Too small a differece to really care....both performance and IQ wise.

[digitalwanderer]

Should I be shocked that Id exagerated/lied about the need/usefulness about 512Mb boards?

[Tim Murray]

Originally Posted by digitalwanderer

Should I be shocked that Id exagerated/lied about the need/usefulness about 512Mb boards?

Lied? Not really, considering they apparently spent quite a bit of time trying to minimize the effect. I'm curious to see if RoE is pretty similar in terms of the usefulness of 512MB or if the optimizations weren't done to the same extent in that.

[Richard]

Originally Posted by digitalwanderer

Should I be shocked that Id exagerated/lied about the need/usefulness about 512Mb boards?

Here's the original id statement for reference:

In Ultra quality, we load each texture; diffuse, specular, normal map at full resolution with no compression. In a typical DOOM 3 level, this can hover around a whopping 500MB of texture data. This will run on current hardware but obviously we cannot fit 500MB of texture data onto a 256MB card and the amount of texture data referenced in a give scene per frame ( 60 times a second ) can easily be 50MB+. This can cause some choppiness as a lot of memory bandwidth is being consumed. It does however look fantastic and it is certainly playable on high end systems but due to the hitching that can occur we chose to require a 512MB Video card before setting this automatically.

High quality uses compression ( DXT1,3,5 ) for specular and diffuse and no compression for normal maps. This looks very very close to Ultra quality but the compression does cause some loss.

[Dave Baumann]

Originally Posted by Jawed

I think this experience intensifies the argument for including an evaluation of framerate minima in a benchmark. I know it's much harder to do... erm...

Doom 3 timedemos don't output min and max frame rates unfortunalty.

[Reverend]

Originally Posted by Joe DeFuria

Nope....I don't even remeber it being "popular" for that matter.

Of course, "popular" can be a subjective term. I do remember quite a number of sites used it back then. A google revealed quite a bit of results.

No big deal anyway.

[Reverend]

Originally Posted by Neeyik

Revs talking about texture compression; I remember the timedemo but I have to confess I never used it much in reviewing - as I do now, I used my own demos back then.

No, it wasn't about texture compression per se but about texture space requirements. The reason I made that demo was to show the benefits of 64MB vs 32MB boards. The level that "Quaver" was based on was one of the Quake3 levels that uses a lot of textures.

I used a 32MB GF and a 64MB Quadro to show the difference IIRC.

[Neeyik]

IIRC (which isn't overly reliable at the moment), with S3TC enabled the textures fit onboard 32MB cards; the benefit 64MB gave was that one didn't have to use texture compression to avoid thrashing in your demo. IIRC again, it was NV's visually iffy compression that helped people view the need of greater local memory over the use of S3TC at that time.

[Reverend]

Originally Posted by Neeyik

IIRC (which isn't overly reliable at the moment <snip>

Likewise here. Crazy times then for me

[Randell]

I certainly remember Quaver being a standard on many sites. I also remember a Rev UT demo being a standard (Thunder?) but misused as it was the small 2player map and really emphaised the CPU dependance.

[Thowllly]

Originally Posted by digitalwanderer

Should I be shocked that Id exagerated/lied about the need/usefulness about 512Mb boards?

Should I be shocked that you exaggerated/lied about what Id said?

Nothing personal, and I see Mordenkainen already addressed it above, but that meme has annoyed me since before D3 was even released. What Id said was that

- they didn't enable ultra mode for 256MB cards by default, because it could cause some stuttering.

but this got immediately warped in peoples heads into

- ultra mode requires a 512MB video card!

And for some reason that sort of thing annoys me. There are a lot of them, so I'm annoyed a lot of the time :P


OT: Another example was when Toms hardware first made a big deal about the fact that if you ran an Athlon (t-bird) without a heatsink you would ruin your cpu and possibly your MB. This immediately got warped into "If your cpu fan stops, your PC will burst into flames" (I have, unintentionally, disproved that statement twice)

[hesido]

I believe texture compression should always be used whether your gfx card is 512mb or 1gig or whatever in future. Otherwise, precious bandwith is being stolen for sth that is hardly noticable in action games. It could be turned off for research purposes in a scientific study, but nah, I want my texture compressed anyway while playing games. Instead of uncompressed textures, I want the textures double the size in pixels for my double amount of memory. How about that. That really steps up the quality.

And btw, I really find it disgusting that they are planning geforce 6200's with 512mb memory. What useless feature is that? This is simply deceiving the public, making people pay extra for a feature that is absolutely no use, while re-instating the false notion that more memory is always better. I don't want to number the amount people that will go for a 512 mb 6200 over a 128mb 6600gt.

[hikasl]

Originally Posted by hesido

I believe texture compression should always be used whether your gfx card is 512mb or 1gig or whatever in future. Otherwise, precious bandwith is being stolen for sth that is hardly noticable in action games. It could be turned off for research purposes in a scientific study, but nah, I want my texture compressed anyway while playing games. Instead of uncompressed textures, I want the textures double the size in pixels for my double amount of memory. How about that. That really steps up the quality.

I heartily agree. I always liked that IL-2 allowed you to turn on S3TC compression. Speaking of which, wouldn't IL-2 be a useful game for testing the value of 512 MB on a video card over a 256 MB card? The maps are often large enough, that's for sure. The last time I ran the game with "Landscape Detail" set to perfect, the Windows Task Manager told me I was using over 700 MB of System RAM, as opposed to less than 400 MB with "Landscape Detail" set to merely "Excellent". Wouldn't a lot of flight sims for that matter?[/quote]