Asus shows off mammoth dual GTX 285 graphics card
PROVING THAT it has never been a company to stick to toeing the tedious line of conformity with reference designs, Asus has just shown us a new monster graphics card that it calls the Mars. It features two GeForce GTX 285 GPUs in SLI configuration, and it also looks like it could double-up as a blunt weapon in Cluedo.
In terms of design, it’s not dissimilar to a GeForce GTX 295. There are two PCBs sandwiched together inside one package, although each of these features 16 memory chips making for a total of 4GB of 1.2GHz (2.4GHz effective) GDDR3 memory. In short, this isn’t a card for anyone running a 32-bit operating system. Asus has also designed its own dual-slot cooler for the card, although it works in a similar way to the reference Nvidia GTX 295 cooler, with a single fan cooling both PCBs.
Interestingly, Asus has decided to spurn Nvidia’s own nForce 200 bridge chip to enable the two PCBs to talk to each other in SLI mode, and has instead used an anonymous third-party chip. One of the interesting innovations here is that, despite the new design, the card will still work with standard Nvidia drivers. The card will be recognised as a GeForce GTX 295, but will still be able to access the features of the GTX 285 GPUs.
As a point of comparison, the GTX 285 GPU has the same tally of 240 stream processors as the GTX 295, but it also has much higher clock speeds, which should make the Mars significantly quicker. The Mars’ core is clocked at 648MHz, with 1,476MHz stream processors, compared with 576MHz core and 1,242MHz stream processors in a standard GTX 295.
As well as this, the GeForce GXT 285 has a wider 512-bit memory interface, compared with the 448-bit interface on the GTX 295 GPUs. Plus, the GTX 285 has 32 render outputs (ROPs), compared with 28 on the GTX 295 GPUs.
Despite these differences, though, the compatibility with standard Nvidia drivers means that the card can be set up in Quad SLI configuration with a second Mars card. In fact, to demonstrate the point, Asus sent us this photo of two cards running in Quad SLI mode on a pretty extreme looking test rig in their lab.
Asus couldn’t confirm whether the Mars is going to reach retailers, or how much it might cost, although the “Limited Edition 1/1000” badge on the top corner leads us to believe that the card will at least make it out in a limited run. Plus, Asus also told us that the colour scheme for the card was decided last month, so it looks as though some serious design decisions are being made.
If you want to see the card in more detail, Techpowerup has also published some under-the-hood shots of the Mars today. µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
could the anonymous chip be lucid's hydra?
It Was Mere Year Ago Mr. T made Those Words In Reference to Vista Ultimate 64 here in theINQ.Msr Kingsley Proved That Statement true. HOW MUCH TRUER TODAY.
Ultee' Slides Down Porcelian Like Fudge Covere Ice Cream, Yummm. 64 BIT ONLY. YEA.
Finally Idea of 32X Pci-e Slot Has PotentialUse & Those 32X Standards Have Been Established.
Remeber 32 Bit IS 32 Bit, While Labeled 64 Bit windows Is Actually iA64 or 40 Bit wrapped in 48 bit Stream Packet, from Outside. So 32 bit to 40 Bit IS Solid, yet 16X to 32X IS Tremendous UpSwing, Same with Pci-e 2.0 Being Strong Step.
'Member in February ''6 When PreNatal Ultee' Tried to Blurble Out WARNING Message, Vista Ani't Right with Out some Majot HardWare Up grades, Yet To Be Known, At That time..
Well I Got thrown Out of here & There, yet Basis Of confusion seems In Mind Of Few "Experts", Whom make living writing on Hardware/Software,Back then.
Time Has Shown Light To Be Slope that Logrythmeticaly Has Had To Increase Quality of Raw Design & NOW, well You Guessed: Mr. T approved: 25May''9.
RIP,Big Fellow.
I PITY FOOL WHOM DOSN'T USE 7 R/C 64!!!
Was just thinking exactly the same thing as Ed....why the secrecy over a bridge chip? If this is the first graphics product to use the lucidlogix Hydra engine SOC then that would be really awesome.
Don't tell Charlie!
- I thought this couldn't be done!
is the floating fan above the CPU...how they do that?
Do my eyes deceive me or are they using TWO power supplies to power that test rig?
I don't even what to think about the power/cooling bills one would have with two of those cards on a core i7 rig...
can it play Crysis?
trying to fit SLI and such into a PC case. They high-end graphic cards, since last year, have become too big, too hot and too power hungry to be comfortably fit into a PC case.
It is about time that graphics will move to an external case, kind of a "graphic sub-station", with a separate power supply and a more effective cooling. For normal day-to-day operation, use a simple graphic chip inside the board. When you want to play games, flip on the graphic sub-station.
A greener solution, more flexible upgrades, better overall engineering.
I agree it's overkill and fitting one card into a mid size tower is hard enough. This setup would need a full tower and atx board with 1000W PS. A small power sub station might be order hehe.
I know I have a Lian Li mid tower case and a GTX 285 in it. That one card rocks anything including Crysis and Farcry 2 at 1920 x 1200 and I see no need for more.
The article says : "In short, this isn’t a card for anyone running a 32-bit operating system."
Why the hell not ? Because it has 4GB of RAM on it although each GPU only address 2GB ? That remark is rather clueless I think. Much like vondrashek's replies.
32 bit Windows xp/Vista can only adress a maximum of 4Gb of memory. Video memory is added to this. So this video would max out the OS with no more memory room.
For example if you have 2Gb of RAM and a 512Mb Video card your computer would report 2.5Gb of memory
hope this helps
Although I'm excited by the thought that this card may have the lucidlogix hydra chip I can't help but think it has been released too late in the cycle...with 40nm cards supporting DX11, GDDR5 and new architectures just around the corner who would spend the top dollar on a card like this so soon to be usurped?
Might be one of these:
http://www.plxtech.com/products/expresslane/pex8647.asp
It is the same PCIe 2.0 switch used on the 4870X2, and the blanked section on the pictures for the GTX285 at TPU is about the right size. It has 48 lanes and three ports, so x16 to the NB and 16x to each GPU. I'm not aware of any other manufacturer that makes PCIe 2.0 switches with enough lanes, aside from the NF200 from nVidia, which probably has cost or performance (or both) reasons not to prefer it over the PLX switch.
I that case why the secrecy over a bridge chip ? that doesn't make sense....also this card appears to have been develpoed quite secretly...has come from nowhere!
Hey what is big issue with that secret chip anyways? I mean 1 card please Asus. Now I smell driver issues in the horizon for this card. But still I want one! Darn 4GB of RAM! I prefer the 2GB flavor please Thank You.
I have just checked by stats and it only shows the 4 GB of RAM that I have on the motherboard, not the additional 256 MB of graphic card RAM. Still I am sure that 4GB on a graphic card will not be fully addressed with 32 bit OS.
Mad Winemaker
The big issue with the chip is that if it is the Lucidlogix chip then it gives linear scaling of multiple gpu's so this card would give you X2 the performance of a 285 rather than the X1.6 or whatever sli normaly gives....If your not familiar with it got to www.lucidlogix.com
Would give a very large increase in performance compared to the current 295 which uses two lesser gpu's and has the standard sli to hinder it.
That was whole point of the one poster who is right. This setup has so much video RAM you need to use a 64bit OS to be able to address it all.
If you had Vista 64 and this setup.
4Gb Video Ram
Say 6Gb System Memeory
Total 10Gb
Vista 64 would see and address all of it.
A 32 bit OS would check when it it 4Gb and probably report 3.5Gb total.
normally i don't get drawn into these sort of conversations, but the shear ignorance being displayed here is hurting my brain.
right. simple version, without crayons.
IF you have 4GB of system ram, and a 32bit OS, the 32bit OS can only see a maximum of four gig. and that has to include all system devices, buses, address range for video ram etc.
so, what ever amount is missing from the 4Gig total, is the amount allocated to video ram, and everything else in the system. then whatever is left is allocated to the address range of system ram.
example, i have 4gb in my laptop, minus the 512MB address space taken by the GFX card ram, and the approx. 540MB eaten by the system devices, 4Mb cache on the CPU yadda yadda yadda, i have just under 3Gb usable sytem ram.
So, if you're GFX card has 4GB ram, a 32bit OS would not be able to allocate ANY system ram at all, as the entire address range has been swallowed up by the GFX card.
got it? further questions on the subject will be ignored.
i've never seen GFX card ram added to or counted as additional to the pool of system ram apart from some strange Linux voodoo.
not typical windows behavior.
To begin with, this card has 4GB of RAM but it is an SLI configuration so the memory is split between the GPUs and this means it appears to the system as 2GB.
Second, Regulus said :
"32 bit Windows xp/Vista can only adress a maximum of 4Gb of memory. Video memory is added to this. So this video would max out the OS with no more memory room.
For example if you have 2Gb of RAM and a 512Mb Video card your computer would report 2.5Gb of memory
hope this helps"
It does not help because it is wrong. I have 2GB of system memory and 1GB of video memory on PCI-X card. The system tells me that I have 2GB of memory (about dialog of the explorer) and the task manager says that I have 1.6GB available out of that 2GB.
Obviously some of you people need to read up on virtual memory and how it is mapped to the physical address space by the OS.
I checked out the memory mapping of my video card and it was rather interesting and I have seen this pattern on two systems with very different PCI-X video cards now. You can go to the control panel of XP and open the device manager and look at the resources used by the video card(s). On the two I looked at it used the following mapping from system memory :
C0000000 - CFFFFFFF = 256MB
FA000000 - FBFFFFFF = 32MB
FD000000 - FDFFFFFF = 16MB
That totals 304MB and that is for a card with 1GB of memory - a GTX285. My other system with a PCI-X video card and 256MB of video memory uses the same memory mapping and I would bet that this card will also. Again : it uses 304MB out of system memory and NOT the 1GB or what ever amount of video memory is on the card.
It might not be DX11 compatible, but its an impressive piece of equipment nonetheless. Anyway, there are no DX11 titles around, and a monster card like that can handle pretty much everything for (a couple of) years to come. If they can do this with the GTX285 chip, they can also do it with the RV790 chip. It would be interesting to see Asus releasing this card, based on Nvidia's chip, and calling it Venus with a green hood (instead of that blue casing) and the RV790 based card called Mars with a red hood. That would be a pretty neat and a showing of the best Nvidia and AMD has, in a single card.
Is everyone forgetting about PAE? The magically thing that allows a 32Bit CPU to address 64GB of memory? Its been around for quite some time now. That would solve the problem with using the card on a 32Bit OS.
Yes, Bob, in principle that is true. In practice M$ would include such support in Windows v42 and charge extra for it - home and lower 1/3 of subversions would not include it. Linux (and I say this as one who started back with Slackware 1.0 and has never left) might take this approach - but who really cares? You can't play Crysis on Linux.
Segmented memory architectures suck. That's why 32 bit & 4GB, and 64 bit and 2^48-or-so B have been successful.