I'll wait for real benchmarks from a reliable source. But I hope the are up to snuff since Nvidia's driver quality has been pretty suspect lately.

Scott,
Nice review. Any plans on testing the 1GB 3870 in a CrossFireX setup to compare the scaling VS the 512MB 3870? I don't thing using three or four 1 GB 3870s is an economical solution to anything, but I am curious as to whether the scaling in memory constrained.

Here ya go...:))
http://forums.legitreviews.com/about15370.html

Damn, I expected the high resolution DX10 Bioshock to take advantage of 1GB of memory atleast.

Check out PCPER's review also. The World in Conflict results are jaw dropping. http://www.pcper.com/article.php?type=expert&aid=553&pid=2

Well, hmmmmm. I expected a little more than almost nothing when going from 512MB to something bigger, after seeing how the X2 cards seem to struggle with 2x512MB. Apparently, 512MB really is the sweet spot right now. Maybe not for some corner cases like modded out games (Oblivion + QTP3 + etc, for ex).

Scott, to try and expose the benefits (if any) of 1Gb video memory, have you tried any of these cards on a graphically intensive MMORPG like Lord of the Rings online?

Try setting texture detail to maxiumum, then enter a city like Bree at peak times, or better yet, the PVP area in the middle of an epic 50 v 50 battle. 100 different characters all clothed in a different set of textures. My bet is that even at 1024 x 768 the 512mb card will start tanking.

Hey Scott!

Typo in the Power consumption page., You said:

"The 1GB Radeon draws slightly more power under load than the 512MB GDDR4 version, but they're dead even when running at game—and, characteristically for a Radeon HD 3870, quite power-efficient."

When it should be "The 1GB Radeon draws slightly more power at idle than the 512MB GDDR4 version"

Good review btw. Nothing new really. We already saw this behavior before, with the 1 GB version of the HD 2900 XT.

Ooo… Copper… Shiny!!

Forget 1GB 3870s...in a month or so Radeon 4870 will leave it and the other NVidia cards in the dust.
25% faster than GX2 ?

Nice review.
I didn't think 1GB would help all that much at this point, and you completely confirmed that with this.
I wonder if this size of buffer would be more effective with CF-on-a-stick cards. Slice off a chunk of it strictly for coordination and memory management... squeeze in 3 GPU's... maybe then it'd have an impact. lol

Nothing unexpected as expected. The ridiculous thing though is that all game test are run with AA on not mentioning with ancient drivers. We all know that the AA performance is weak with new radeons so I think it doesn't have to be underlined anymore. I personally never use AA because it's benefits are minimal when using high resolutions. AA also blurs textures which isn't nice effect at all.

Memory is cheap, and it does add a bit of future-proofing.

Not really. By the time games really need 1 gig of memory, all of today's cards will be obsolete.

I have but one question. Okay, so I have two.

Why not use the Crysis benchmark? It can be a tiny bit I/O-bound during the first run, but as far as I've noticed, things become established in memory after that and all consecutive runs are well representative of performance.

Other one: planning to replace 3DMark06 with Vantage any time soon?

I'm still waiting for a 3650 X2...

Nice read. Must get a bit monotonous to do all that benchmarking when you already know where the bird will land, but it was a good article all the same.

EPIC FAIL

Even though I knew what your conclusion would be, I read this anyway... It's always entertaining to read your work Scott. Thank you!

Why don't they put 1gb of ram on a card that might actually use it like a 9800gtx? Might make 3 and 4 way sli halfway decent. Just stick a 64-bit vista only sticker on it.