AMD's Dual GPU Monster, The Radeon HD 3870 X2

MojoKid writes "AMD officially launched their new high-end flagship graphics card today and this one has a pair of graphics processors on a single PCB. The Radeon HD 3870 X2 was codenamed R680 throughout its development. Although that codename implies the card is powered by a new GPU, it is not. The Radeon HD 3870 X2 is instead powered by a pair of RV670 GPUs linked together on a single PCB by a PCI Express fan-out switch. In essence, the Radeon HD 3870 X2 is "CrossFire on a card" but with a small boost in clock speed for each GPU as well. As the benchmarks and testing show, the Radeon HD 3870 X2 is one of the fastest single cards around right now. NVIDIA is rumored to be readying a dual GPU single card beast as well."

Comments

But does it run Linux?

[Ed Avis]

No mention from the article summary of whether this is supported by ATI's recent decision to release driver source code. If you buy this card can you use it with free software?

(Extra points if anyone pedantically takes the subject line and suggests targetting gcc to run the Linux kernel on your GPU... but you know what I mean...)

Re:But does it run Linux?

[habig]

No mention from the article summary of whether this is supported by ATI's recent decision to release driver source code. If you buy this card can you use it with free software?

While AMD has done a good thing and released a lot of documentation for their cards, it has not been source code, and has not yet included the necessary bits for acceleration (either 2D or 3D). That said, I'm watching what I'm typing right now courtesy of the surprisingly functional radeonhd driver being developed by the SUSE folks for Xorg from this documentation release. While lacking acceleration, it's already more stable and lacks the numerous show-stopper bugs present in ATI's fglrx binary blob.

Dunno yet if this latest greatest chunk of silicon is supported, but being open source and actively developed, I'm sure that support will arrive sooner rather than later.

Re:But does it run Linux?

[GuidoW]

Actually, what did they really release? I remember some time ago, there was a lot of excitement right here on /. about ati releasing the first part of the documentation, which was basically a list with names and addresses of registers but little or no actual explanations. (Although I guess if you have programmed graphics drivers before, you'd be able to guess a lot from the names...)

The point is, it was said that that these particular docs were only barely sufficient to implement basic things like mode-setting and 2D-support and maybe TV-Out, but certainly not 3D-acceleration. There was a promise by ati to release even more documentation in the future to allow these things, but so far, I haven't seen anything. I did some googling to find out if maybe I've missed something, but that turned up very little. Even the X.Org wiki didn't help much.

So, does anyone here know a bit more? What's the real status of the released docs? Is there enough to do a real implementation with all the little things like RandR, dual head support, TV-Out and 3D-support, or is ati just stringing us along, pretending to be one of the good guys?

With AMD Open Source Linux Drivers

[lord_rob the only on]

It's time to change my aging Athlon 900 MHz then :-).

Multiprocessing everywhere!

[AceJohnny]

Can't make it faster? Make more. Another multiprocessing application. Can I haz multiprocessor network card plz?

When can I have a quantum graphics card that displays all possible pictures at the same time ?

Re:Multiprocessing everywhere!

[mwvdlee]

Here's your "all possible pictures at the same time" (using additive mixing), and it doesn't even require you to buy a new graphics card:














Cool éh?

Re:Multiprocessing everywhere!

[kvezach]

Bah, that probability distribution is just wrong! Or you overflowed all your pixels.

Re:Multiprocessing everywhere!

[MulluskO]

Man is finite, and therefore cannot make infinite observations.

Don't bother

[BirdDoggy]

Wait for the nVidia version. Based on their latest offerings, it'll probably be faster and have more stable drivers.

Seriously? Yawn.

[esconsult1]

Am I the only one underwhelmed by almost every new graphics card announcement these days?

Graphic cards have long since been really fast for 99.9999% of cases. Even gaming. These companies must be doing this for pissing contests, the few people who do super high end graphics work, or a few crazy pimply faced gamers with monitor tans

Re:Seriously? Yawn.

[Cerberus7]

Actually, graphics power isn't fast enough yet, and it will likely never be fast enough. With high-resolution monitors (1920x1200, and such), graphics cards don't yet have the ability to push that kind of resolution at good framerates (~60fps) on modern games. 20-ish FPS on Crysis at 1920x1200 is barely adequate. This tug-of-war that goes on between the software and hardware is going to continue nearly forever.

Me, I'll be waiting for the card that can do Crysis set to 1920x1200, all the goodies on, and 50-60fps. Until then, my 7900GT SLI setup is going to have to be enough.

Re:Seriously? Yawn.

[Man in Spandex]

Honestly, I doubt you play FPS games because the difference between the 24-32fps range and the 50-60's is way noticeable. Forget the theoretical technicalities of human eyes capabilities for one second because I'm sure when the FPS of a game reaches the 30's, there are other factors that make it sluggish and all that together give us the perception that the difference between 30's and 60's is an important difference.

Re:Seriously? Yawn.

[stewbacca]

Play the game and enjoy it for the best settings you can get. I downloaded the Crysis demo last night for my 20" iMac booted into WinXP (2.33ghz c2d, 2gb ram, 256mb X1600 video card, hardly an ideal gaming platform, eh?). I read that I wouldn't be able to play it on very good settings, so I took the default settings for my native resolution and played through the entire demo level with no slowdowns. It looked great.

The real problem here is people feeling like they are missing out because of the higher settings they can't play. Just play the game! Quit fueling the ridiculous ten-year-old trend of spending more on graphic cards than the computers themselves! If the game were unplayable at the medium setting, then yeah, I'd say the complaint is valid.

Re:R680

[MightyYar]

You must be mis-pronouncing it - it's R-Upside-Down-Nine-Vertical-Infinity-Circle. That's how the engineers all refer to it internally.

Pretty cool if you ask me.

Driver dependent performance

[Overzeetop]

Ultimately though, the real long-term value of the Radeon HD 3870 X2 will be determined by AMD's driver team. That doesn't really bode well, given the clusterfuck that the CCC drivers tend to be.

Re:Driver dependent performance

[chromozone]

ATI/AMD's drivers can make you cry. But their Crossfire already scales much better than Nvidia's SLI which is a comparative disaster. Most games use Nvidia's cards/drivers for development so Nvidia cards hit the ground running more often. As manky as ATI drivers can be, when they say they will be getting better they tell the truth. ATI drivers tend to show substantial improvements after a cards release.

Re:Driver dependent performance

[ashpool7]

Hm, well if that's the case, then nobody should run out and buy this card.

WRT Crossfire... I had a friend who wanted to buy Intel because they're "the fastest." Hence, he was stuck with ATi for video cards. Except the latest driver bugged Crossfire and he spent a couple hours uninstalling the driver to reinstall the older one. Doesn't that sound like fun?

nVidia's drivers aren't better because they're used for development, they're better because nVidia knows "IT'S ALL ABOUT THE DRIVERS, STUPID". ATi still has not learned this lesson. They need to just die and maybe somebody who knows a thing or two about driver development will pick up their charred remains.

Does it come with...

[Walles]

... specs or open source drivers?

I haven't heard anything about any specs for 3d operations being released from AMD. I know they were talking about it, but what happened then? Did they release anything while I wasn't looking?

Re:Does it come with...

[Andy Dodd]

They probably are pulling a Matrox. Release partial specs, promise to release more, rake in $$$$$$$$$$ from gullible members of the Open Source community, fail to deliver on promises. Great short-term strategy but only works once before said community stops trusting you, especially those who were dumb enough to go for your promises like I was back in 1999.

Ever since I made the mistake of buying a Matrox G200 (Partial specs - more complete than what ATI has released so far as I understand it, and a promise for full specs which were never delivered) I never make buying decisions based on promised specification/driver releases - only what works NOW, whether binary or not. (Hence I've been happily buying NVidia for 6-7 years now.)

Re:Does it come with...

[Kjella]

I haven't heard anything about any specs for 3d operations being released from AMD. I know they were talking about it, but what happened then? Did they release anything while I wasn't looking? They released another 900 pages of 2D docs around Christmas, 2D/3D acceleration is still coming "soon" but given their current pace it'll take a while to get full 3D acceleration. So far my experience with the nVidia closed source drivers have been rock stable, I have some funny issues getting the second screen of my dual screen setup working but it never crashed on me.

Drivers are something for the here and now, they don't have any sort of long term implications like say what document format you use. The day ATI delivers, I can buy a new card and switch instantly. They can promise to release specs all they want, I'll promise to buy one when they do. ATI may find that promises are cheap both in the giving and the getting. I'm still afraid that all the good stuff will be stuck in the legal department forever.

Re:This just in: New technology faster than old.

[Kamots]

Interesting thing is what happens when you stop looking at synthetic benchmarks... and start looking at real gameplay.

Take a read through hardocp's review for an example.

As to why AMD released? Well, my understanding is that NVidia is looking to release thier own 2-GPU card (9800 GX2) in Feb/March. Given the benchmarks of the current cards, I can't see the 3870 X2 holding up well... so... beat 'em to market. Although when you factor price in, I'd imagine it'll still be competitive; just not anywhere near the fastest.

What I'm waiting to see come out from AMD is the R700 cards... especially if it convinces nvidia to finally release thier true next-gen cards as well (not merely the continued tweaking/shrinking of the G80 architecture). Then we can all have something to look forward to :)

Re:Anyone remember . . .

[TheRaven64]

The Rage 128 Pro was never close to the top of the line for a graphics accelerator (and doesn't really qualify as a GPU since it doesn't do transform or lighting calculations in hardware). It was around 50% faster than the Rage 128, which was about as fast as a VooDoo 2 (although it also did 2D). You had been able to buy Obsidian cards with two VooDoo 2 chips for years before the Maxx was released, and the later VooDoo chips were all designed around putting several on a single board.

Next up...

[jez9999]

Work is in the pipeline for a board which can house all your computer's necessary components, including a multiple core CPU that can handle graphics AND processing all-in-one! It will be the mother of all boards.

Crossfire on a Card.

[seeker_1us]

So we have come full circle to the Voodoo 5 then?

Not at all

[Sycraft-fu]

Many things you are wrong with there. The first is framerate. If you can't tell the difference between 24 and 60 FPS, well you probably have something wrong. It is pretty obvious on computer graphics due to the lack of motion blur present in film, and even on a film/video source you can see it. 24 FPS is not the maximum amount of frames a person can perceive, rather it is just an acceptable amount when used with film.

So one goal in graphics is to be able to push a consistently high frame rate, probably somewhere in the 75fps range as that is the area when people stop being able to perceive flicker. However, while the final output frequency will be fixed to something like that due to how display devices work, it would be useful to have a card that could render much faster. What you'd do is have the card render multiple sub frames and combine them in an accumulation buffer before outputting them to screen. That would give nice, accurate, motion blur and thus improve the fluidity of the image. So in reality we might want a card that can consistently render a few hundred frames per second, even though it doesn't display that many.

There's also latency to consider. If you are rendering at 24fps that means you have a little over 40 milliseconds between frames. So if you see something happen on the screen and react, the computer won't get around to displaying the results of your reaction for 40 msec. Maybe that doesn't sound like a long time, but that has gone past the threshold where delays are perceptible. You notice when something is delayed that long.

In terms of resolution, it is a similar thing. 1920x1200 is nice and all, and is about as high as monitors go these days, but let's not pretend it is all that high rez. For a 24" monitor (which is what you generally get it on) that works out to about 100PPI. Well print media is generally 300DPI or more, so we are still a long way off there. I don't know how high rez monitors need to be numbers wise, but they need to be a lot higher to reach the point of a person not being able to perceive the individual pixels which is the useful limit.

Also pixel oversampling is useful just like frame oversampling. You render multiple subpixels and combine them in to a single final display pixel. It is called anti-aliasing and it is very desirable. Unfortunately, it does take more power to do since you do have to do more rendering work, even when you use tricks to do it (and it really looks the best when does as straight super-sampling, no tricks).

So it isn't just gamers playing the ePenis game, there's real reasons to want a whole lot more graphics power. Until we have displays that are so high rez you can't see individual pixels, and we have cards that can produce high frame rates at full resolution with motion blur and FSAA, well then we haven't gotten to where we need to be. Until you can't tell it apart form reality, there's still room for improvement.