|
|
AMD announces ATI Radeon HD 3870 X2
Author: Tim SmalleyPublished: 28th January 2008
AMD's Radeon HD 3870 X2 is quite an elegant design considering the fact it's a pair of GPUs on a single card.
For those that were expecting a review of the HD 3870 X2 on bit-tech this morning, we're sorry that it's not quite finished yet - as we're waiting on a few answers from AMD on some problems we encountered during our testing.
AMD describes the card as "an elegant yet aggressive design", as it incorporates two GPUs connected through an integrated CrossFire chip on a single PCB.
Like the rest of the Radeon HD 3000 series, the Radeon HD 3870 X2 supports DirectX 10.1, CrossFireX (when drivers become available), ATI PowerPlay and also AMD's Unified Video Decoder.
AMD claims the card delivers a two fold increase in performance-per-watt over the Radeon HD 2900 XT - this is thanks to the combination of the 55nm manufacturing process and ATI's PowerPlay technology.
The GPUs are collectively known as R680, but it's actually a pair of RV670 chips with different clocks to the Radeon HD 3870. The core clock is set at 825MHz - compared to 775MHz on the single 3870; the memory, on the other hand, is clocked at 1.8GHz - down from 2.25GHz on the 3870.
A lot of this card's success is going to depend on AMD's support for CrossFire---a lot of which we discussed in our comprehensive review of the Radeon HD 3870, and we'll see how it fares when all of our results are complete. For now though, you can discuss AMD's re-entry into the high-end in the forums.
Related articles on bit-tech.net
Comments (84)
peace
so, what will end up faster? 3870x2 or 2x 3870?
And the other good news is that its NOT crossfire either. (Works as a single slot card). There may be two GPU's on one card, but its not like the Nvidia solution (7950 GX2 card) which was a bug ridden SLI piece of rubbish.
Looks like ATI has a hit on its hands.
Wonder what UK pricing will be. About time AMD/ATI had some good news.
ocuk got it listed for exactly the price of 2x 3870. (300£)
this is a good question.
Anand is always better than the trash that tom's and Kyle at HardOCP put out. That said, can't wait for the Bit article on it.
what driver problems are these all the reviews i have read stsate there has been no driver problems and it out performs the ultras
This is either an assumption or you've been misinformed; I've seen many deals for 8800GT @ £150, which puts it at exactly the same price-point as the new 3870x2.
2x 8800GT requires you to use a nvidia mainboard though, while with 3870x2 your free to use good mainboards.
Lol, a little OT but oh so true.
Heh, now everyone who has bought into marketing hype and purchased a PSU >800w will actually start to be able to use it to >50% of it's capacity.
1) Does ATI's "stretch" mode (ATI equivalent of Nvidia's Span) work with this?
2) When using both outputs are both GPUs working? (I ask because on my 7950 GX2 if I used 2 monitors the second GPU did nothing )
but im still waiting for a single GPU best card to upgrade to
(conditions: at least 1.6x the performance of 8800GTX, only require me to put in £200 after selling my 8800GTX)
Yep it as a built in SLI bridge chip between the gpus, so it appears to the O.S. a single 3875. Question, as it appears as a single card is the cards power still reliant on games providing SLI profiles like two discrete cards
is Bittech's review going to have comparison between 8800GTX?
4 of these would be awesome!
Though I am not sure if any motherboards would have space
this is a glue together with slower ram. Sorry to say but the 9800GX2 will probably be faster.
depends... the pictures floating around on the web show a dualpcb card as the 9800gx2, which would very likely mean lowered clocks due to heat.
Now if AMD would get their ass in gear and start releasing top notch processors again to bring us back to the athlon 64 days.
could we get some Sli tests in there too? If possible Ultras in SLI, its rare to see a test have that setup so would be interesting to see results.
nice, good to see a comparison against the complete current lineup of both, but you could add something experimental, trying to crossfire a 3870x2 with a 3870 for trfire... for quadfire (i.e. 2x 3870x2) you prolly still got an nda.
Anyway, review is live: http://www.bit-tech.net/hardware/2008/01/29/amd_ati_radeon_hd_3870_x2/1
amd really needs an all new gpu architecture where they wont have to fix the drivers to make their products work properly with each game.
anyway, i'm seriously considering to buy one still... but i'll prolly just get a 3870 in 2 days (woot payday soon) and then give 3870x2 a month for the drivers to come out in trifire flavour and for fullcover wc blocks to be avalaible.
what? aluminum does not dissipate heat better than copper. does it?
2 3870's in crossfire.
maybe... but only if the two 8800 gt's on a single card dont end up as a failure.. which is still unknown. and if they would have something ready for an immediate counter they would have announced it by now to prevent 3870x2 sales from happening.
peace
EK allready has a 3870 X2 block on Pre-order!
4 months away, gah
Aluminium gets rid of heat quicker than copper (i.e. dissipate, radiate, etc), while copper holds onto heat much longer (i.e. better conductor, higher specific heat capacity)
supposed to read 5th february.
just checked ek weshop, there it says 31.01.2008, so 5th feb makes sense. there will apparantly be three versions, copper with acetal top, copper with acrylic top and nickel plated copper with acrylic top.
"05/02/08" oh lol, silly brits and their different way of writing dates. I'm used to MM/DD/YY instead of DD/MM/YY
Good Luck to ATI/AMD!
Silly? Seems far more logical to me as it starts with the smallest value (day), then the next (month) and so on...(year). Still, I guess if there isn't a "u" to remove anywhere, just shuffle it up a bit.
But then again I'm a silly Brit ;)
Hardware Canucks conducted a review of the same card and it ended up trumping nVidia's cards in the majority of their tests.
"ATI Radeon HD 3870 X2 512MB" appears in each graph on Page 7 - Crysis.
Anandtech benchmarked a cutscene... not sure about Tech Report though. Our tests were done manually inside the game.
Oh and while I'm at it... not sure why they're running FLYBYS?!
waiting for 9800GX2
seems like gameplay results are same as HardOCP, where the more than 1 year old G80 core is still better. this dual GPU on one PCB card is good for benchmarking (as other reviews showed) but bad for gameplay, where most of the time, with worse minimum FPS than even GTX.
im now not sure about 9800GX2 is a good idea, just 2 G92 stuck together, may be more shader processing power, but will the drivers keep up? will nVidia's driver offer better minimum FPS figures? (IMHO, minimum FPS is what we should be looking at, where it represents the shutter moments, and if it's better than 20FPS, it'd be fine)
http://i33.photobucket.com/albums/d96/noway_1961/38703dmark.png
Not bad for on stock settings with immature drivers i'd say, also from a first person visual perspective.
the images on the whole, and in game are nice and crisp.
qx6700@ 3.2ghz
HIS 3870x2 all stock
2gb crucial ballistix 6400
asus commando
my last card was a bfg 8800gtx oc2 version, I have to say already I prefer the 3870x2, saying that my experience with the gtx was not a happy one.
may i ask why?
run more games, im interested in whether it will have compatibility issues with games. afterall, it runs on principle of crossfire (aka, software managing GPU power, not like a single 8800Ultra, pure power)
Hi if your asking why, my experience with a gtx was not a happy one. The card was rma'd 6 times in five months. This is just my experience.
I know many people who have had great experiences with gtx's.
My theory is the retailer stocking my particular gtx got a faulty batch, [chipset maybe]the problem was if I left my pc on 24/7 the card was great. if I powered off over night
the card would not post at all from a cold boot, like I said this happened 6 times in total, at the end of it all the gtx cost me close to £400.
Fortunately for me the retailer refunded me, even though on the last one I was outside their refund period.
So I was a bit nervous about going nvidia again. [btw] the issues described above were not related to my psu or any other factor] the cards were dead.
This 3870 [back on topic] ive played crysis and cod4. crysis I played on defualt high spec settings, I ran the crysis benchmark tool. In game I didnt experience any glitching at all as reported in many reviews, though in saying that I didnt play it for very long.
heres some results from the benchtest with the hotfixed 8.1 cat drivers..
Completed All Tests
<><><><><><><><><><><><><>>--SUMMARY--<<><><><><><><><><><><><><>
1/30/2008 5:13:45 PM - XP
Run #1- DX9 800x600 AA=No AA, 32 bit test, Quality: High ~~ Overall Average FPS: 57.08
Run #2- DX9 1024x768 AA=No AA, 32 bit test, Quality: High ~~ Overall Average FPS: 53.345
Run #3- DX9 1024x768 AA=2x, 32 bit test, Quality: High ~~ Overall Average FPS: 37.56
The resolutions are on low res settings due to running on a 17" monitor, ill hook it up to a 24" later and get a better idea.
Im not certain, but I think those averages are not bad considering the drivers are basically beta drivers.
Im running 8.1 drivers with a hotfix addon.
http://i33.photobucket.com/albums/d96/noway_1961/atigpuz.png
I have'nt played many other games yet because there is an issue with my rom drive. Ill swap the drive over and get some results up later.
heres a screen from crysis..
http://i33.photobucket.com/albums/d96/noway_1961/Crysis2008-01-2921-51-08-18-1.jpg
Im looking forward to cf being enabled with up coming drivers which will technically be quadfire...
need another card though..
bustamove: sorry to hear about your bad experience with the GTXs - it could have been any number of factors that were triggering the cards to die. I've had similar undiagnosable situations and that's when it sucks the most because you know you're not getting what you should be getting. :(
I should have said before it was a bfg oc2 version gtx. I do believe there has been problems with that particular model. [could be the factory overclock maybe]
When I put my old x1950xt in the gtx's place I got post immediately.
as a comparisin, and as you rightly say Tim [for a single card] the gtx was getting 14750 in 3dmark 06 with the same set up as above. card was at stock settings 626/2000mhz.
which actually closely matches ultra performance.
For what its worth and I have no way of proving it. the image quality in general seems more defined and crisp with this 3870x2 than the gtx I had.
The only reason I draw a comparisin to the gtx is because I very recently had one in the system Im using the 3870x2 in.
that is precisely why i am againest multi-GPU card/setups.
those 3Dmark 06 scores means little verses gameplay, as long as one is happy with their gameplay experience, there's no reason compare scores. and this is why i asked about other games, looking at Bittech's Crysis minimum FPS, i wanted to know whether you get a good gameplay experience. some game may experience glitches due to the fact that driver isn't well optimised (which comes back to the disadvantage of multi-GPU)
Are you running XP and if so could you tell me if span mode is available on this card. (I think ATI calls it "stretch" - basically the system sees two monitors, at the same resolution, as one larger monitor.)
@Tim Smalley / Bit-Tech staff: Can you tell me what happens when you connect two monitors. I know that the 7950 GX2 ran in either SLI mode or single GPU mode. If I wanted two monitors it shut down the second GPU. What does the 3870 X2 do in the same situation?
very smooth gameplay.
Is anyone here aware of a benchtest for cod4? is there a command or something?
I really am impressed with this card so far. @ Tyinstar, im going to hook it up to my 24" monitor in the morning, my rig is watercooled and its a bit of a pain moving the pc around.
not sure about the stretch thing but ill check it out.
Im really happy i bought this card now!
Can i quote something someone said on another forum? I would be interested to see what you guys say about it.
quote; Both gpus have 512mb each. It has 1 gb total, but seeing how it's technically CF on a card, it has to store the same info in both gpu's memory banks, resulting in it functioning like it was only 512mb.
also, quote; A LOT less latency than regular cross-fire.
You see, regular cross-fire has to do a majority of it's communication through the chipset. Meanwhile this card has it's own splitter and does it's communicating all on card. As such, there's less latency all the way across the board.
ALL current multi-gpu set-ups have to duplicate data for each gpu in ram. They have to, because both gpus need the same textures and shader information as they both work together to render the scene. Essentially they all run as if they had the same amount of ram as the lowest card does. That's just how things have to work currently, and why I believe the R700 will have to have an external memory controller for the MCM package.
To sum it up, this card is like a lower latency CF set up. unquote;
well I kind of see what the guy is getting at, but it doesnt seem logical. i.e. how, at stock settings does it perform very similar to a single card cf set up?
any thoughts on this?
I don't think there's really a latency issue with standard CrossFire - if there was, we'd definitely know about it (you'd notice it when you were playing games). Of course, the fact that the data doesn't have to go via the CrossFire interconnect/chipset is a bonus... but since the GPUs are rendering three frames ahead anyway, it's not going to make any difference IMO.
are you saying that single card crossfire only uses 512mb as well as the x2 version?
You can probably tell that Im not very technically literate. but I dont understand why it performs on a par or better in most tests Ive seen that 2 single cards in cf mode?
thats the part im getting lost at.
Prior to buying the card, I did consider buying two singles but the possibilty of having two x2's in crossfire in the future clinched the choice I made.
thanks for your reply to the question.
When the CPU sends commands (from the game) to the graphics card, it is sent to both cards in a CrossFire/SLI configuration - this data is stored in both memory pool as one GPU can't access the other's memory pool directly (there's no physical connection).
The reason the X2 might perform slightly better in some situations is because of the higher core clock. There is a 50MHz increase on the core (825MHz vs 775MHz), but the memory loses 450MHz (effective, 1800MHz vs 2250MHz)... In memory bandwidth limited scenarios, the memory clock is the reason why the 3870 X2 can be slower than the 3870 CrossFire config.
Hope this helps to explain it a bit mo