Asus shows off dual Nvidia Fermi card
Two GF100 GPUs make the card an absolute beast.
Each of the two GeForce GTX 480 GPUs (GF100-375-A3) on this massive Asus card have 480 stream processors, and are connected to 1,536 MB of GDDR5 memory across a 384-bit wide memory interface.Nvidia's NF200 (A3 version) chip is also used.
Powering each core are four chunky power phases, with one driver, two chokes and six MOSFETs per phase, with one equally sized phase for the memory too. The card appears to be 30cm in length if we extrapolate from the fact that about half the card is 15cm. It requires three 8-pin PCI-E power connectors.
Next to these is a circular gizmo - we're guessing it's something RoG specific - maybe an OC or overvolt tool. Republic of Gamer hardware often débuts with a new or revised performance feature.
There is no indication of the cooling yet as we've only managed to obtain PCB shots, however given a normal GeForce GTX 480 is hot, expect the apparatus involved to be considerable. Given that there's only one fan connector and that if we assume Asus will use the knowledge gained from its Ares card - which promoted the use of a single, large fan, maybe Asus will try a variation on this method again.
Asus even expects two in a case for quad SLI, or at least maybe paired with a third GTX 480 with the SLI connector in the top corner.
Hopefully Nvidia will be happy one of its partners is chancing such a ludicrously powerful card - but then that's not always been the case with these kind of products. Last year at Computex when Asus showed off its original Mars with dual GTX 285s - one upping Nvidia's own GTX 295 in the process - we were told a top Nvidian was so livid with Asus that he stormed onto the booth telling them to pull the card from show! If Nvidia is making a dual GF104 card, then we wonder how Asus one upping them once again will go down...
Click to enlarge
Previous Article
71 Comments
Discuss in the forums Replyhttp://forums.bit-tech.net/showthread.php?t=191021
I would assume it's a diagnostic port of some kind, something for an engineer's test board, but something that won't make it onto the final version.
for a $1000 I rather buy a car as a HD 5870 2GB ($500) can power a 6 screen Eyefinity set up quit nice.
It's cheaper to just get a gtx 480 SLI set up instead of a $1200 single card that want perform any better than a GTX 480 sli
3 GF 104s on one board would have been better as 3 way GTX 460 sli is not supported.
I think that's stupid that an Nvidia upper would be mad about that, if they sell more units why should he be all up in arms about it?
lolz @ drilling holes in the card jsheff...too funny...
the clover honey in his pink teacup doesn't taste quite as sweet anymore :D
Looks like Asus are putting a cheeky middle finger up to Nvidia with this card. Well that and the 'because we can attitude they seem to have.
The entusiant in me loves it, the practical person in my thinks they are mad.
They should release a 4 Femri Core card with 3192MB ram, consisting of 3 logic boards:
Top board: Power. Two Femri cores, half the ram, and a GF100 chip
Middle board: The other two cores and half of the ram.
Bottom board: Power rectification - in the footsteps of the V5-6000, this needs an external, surge protected connector for power. I'm not sure if they would put the entire 500W power supply on this board (shouldn't be a huge issue, as we do have 1U 500W units), or just a DC-DC unit. In any case.
Of course, I'd say the very top slot should have 3-4 HDMI or Displayport - since we want flexibility here.
The only big issue is that I don't see much of a market for $3,500 6-slot {gaming} cards.
If this thing isn't water-cooled, it's going to heat up your room in this hot summer faster than the heater can say "who stole my job."
Talk about excessive heat, excessive wattage, and excessive noise from the 100,000rpm fans.
Right on Asus!
I completely agree this would have to be water cooled or else like someone else said the cooler alone would take up the remaining slots lol. Probably have an AC unit for it =p yay more power draw lol.
Would you have the kindness to explaing jsheff joke to me.. i dont undestand it :|
Yes, it is THAT superfluous. Except maybe for 3DMark junkies...
nope probably 220CFM Deltas =p
Every bit-tech graphics review comes with a picture of the card with no cooler. With the gtx480, in the process of removing the cooler they managed to strip one of the screws. Time was late, everyone wanted to get home, so out came the trusty dremel to cut the screw off so the critical shot could be taken. *Slip*. Oops. Before the graphics card has even been launched to the general public, before the first review numbers have been made available, bit-tech have destroyed their sample of what was at that time the most exclusive graphics card in the world by drilling a neat hole in the PCB.
All that's left of it in the labs is the cooler - EVGA took back the board so that they could salvage what they could from it and make it into a new product.
http://www.gadgettastic.com/images/fan_case1.jpg
+
http://www.webbaviation.co.uk/industrial/ferrybridge-power-station.jpg
+
http://toolmonger.com/wp-content/uploads/2008/07/DM_MilwaukeeEarmuffs.jpg
That is what I suppose the logic behind this card is...
Your forgetting that the Mars/ROG cards are for people who don't care about money.
This should be a good time to market too, unlike the original Mars card that came out just before the 5*** series.
It's not going to be. Each card is about $1,500. Even in high density situations, they're impractical. Especially considering that 7 full slots of these has a maximum theoretical draw is over 3.5kW. Cooling that would be an atrocity; not to mention powering it... while it's an interesting concept for a render farm, but just prohibitive.
AS for this dual chip, it harkens back to Voodoo's cards.
The cheapest GTX 480 is $100 more than the cheapest 5870 I can find, performance is within 5% of each other in almost every case, it's noisier and hotter.
Yes, it does CUDA and has PhysX, so the GTX 480 isn't a bad card, it just makes no sense to most people to buy one.
If you watercool of course.
If you are a surround user (or are thinking about it) the GTX480 and GTX470 cards are far from a waste of time.
Both the GTX480 and GTX470 scale better than the HD5870 when running in multi GPU scenarios (former card especially so). The extra memory could also be useful when running 5760x1080 and beyond.
I agree though the GTX465 was and is a complete dog. The only sacing grace it had was ATI also launched a dog in the shape of the HD5830.
Where does it say that it's going to be 7 slots? All the pictures I've seen show a one-board GPU.
Yes, this will probably end up as a premium board, like the Ares (which comes in an attache case LOL) but it's still a fun idea.
I'd say that it should come with a waterblock, but not for the reasons you think (2xGTX480 can be cooled if you throw enough time into it).
Other high end cards (such as the Ares, or even the Matrix 5870) are bloody fast but they have aircoolers which can get quite noisy. If you've got a large amount of money to spend on a premium product, you'd probably have a large amount to spend on cooling it as well. The problem is that Asus have obviously used custom PCBs, so noone makes waterblocks, thus making watercooling it hard and difficult.
For a card that costs that much, you'd want it to at least be coolable.
480 cheapest pre installed water cooled card is alot of money sadly
For £1000 you could watercool two normal GTX480s, or even three HD5870s (albeit for a bit over £1000)
Reading comprehension isn't your strong suite, huh? I said 7 slots OF THESE. Meaning populating each of 7 slots with one of these cards - which would probably create an instant water boiler.
...just going down the shop now to buy my lottery ticket lol!!!
How would that be helpful in any way? You can only use up to 4 gpus in quad-SLI, which means two cards.
There's absolutely no point using any more than two of these cards (because they wouldn't be used), which would draw a maximum of around 1200W. I read your post exactly, I just don't see how you plan to fit 7 of these in a system when there will certainly be at least a 2 slot cooler and where you'll only be able to use two of the cards.
Secondly, if someone made you waterblocks for them, you could wedge these into single slots.
While a 14 GPU (and presumably 12 core) render farm is titillating to think of, the 4,000W it'd draw ain't.
Or the £10k+ price tag?
It would be *SO MUCH* cheaper to simply use either more than one computer or just use pcie splitters and normal GTX480s.
I can't think of any possible consumer or prosumer uses for this, purely because it would be massively low value for money in comparison to the other options.
ATI has a better solution with Eyefinity unless you are talking of 3D as well.
True but what I am saying is that they could have refined GF100 as they did with GF104 to strip out the features not essential to drawing 3D graphics to produce a better graphics card. What they did instead was rush GTX 480 to market as they were way behind schedule anyway.
Yep but you can boost the clocks up a lot higher on GF104 as well, you can get performance greater than a GTX 470. I don't think it would be possible to challenge GTX 480 though with a GF104 chip.
nVidia uses market segmentation for the highest profits possible. IMO, the reason why the GTX460 is clocked so low at stock (they could have clocked it at least 75MHz higher imo) is that having it clocked higher would push the 460 closer to the more profitable 470.
The 480 is faster and more expensive than the 5870. The Ares uses 2x 5870s, the Ares costs 1200£, so this will probably be closer to 1500£ if it ever hits the market (which I don't really believe in).
So yeah. The only real time when something like that would be appropriate is where you need as much processing power as possible inside a really confined area. Considering just a few racks of SAN (Storage Across the Network) can run into the millions, 250k per rack of insane processing power isn't much.
But if you're doing something like that, I assume you can just work directly with NVIDIA to create custom femri blade servers...
That said, I really wouldn't want to hear what the fans sound like when air is all you have to cool 48 femri cores.
4gpus max per system which means 2 total of this card as it has 2 gpu on it you can't but 7 in sli even if you wanted to
4 x 480 would be cheaper and faster
Before anyone else goes waffling on about some imaginary GPU limit, read this - http://en.wikipedia.org/wiki/CUDA
You aren't using 7 of these cards for gaming. You're using them to crunch data in specific programs. Say a folding rig. So yes, while 4 x 480 would be faster, I fail to see how 4 GPUs would be faster than 14.
How long is this card going to be?
It's an nVidia limitation.