Slashdot Mirror
NVIDIA Unveils Dual-GPU Powered GeForce GTX 690
MojoKid writes "Today at the GeForce LAN taking place in Shanghai, NVIDIA's CEO Jen Hsun Huang unveiled the company's upcoming dual-GPU powered, flagship graphics card, the GeForce GTX 690. The GeForce GTX 690 will feature a pair of fully-functional GK104 "Kepler" GPUs. If you recall, the GK104 is the chip powering the GeForce GTX 680, which debuted just last month. On the upcoming GeForce GTX 690, each of the GK104 GPUs will also be paired to its own 2GB of memory (4GB total) via a 256-bit interface, resulting in what is essentially GeForce GTX 680 SLI on a single card. The GPUs on the GTX 690 will be linked to each other via a PCI Express 3.0 switch from PLX, with a full 16 lanes of electrical connectivity between each GPU and the PEG slot. Previous dual-GPU powered cards from NVIDIA relied on the company's own NF200, but that chip lacks support for PCI Express 3.0, so NVIDIA opted for a third party solution this time around."
The top right connector is different; any idea why this is? I also have cables that look like that, and in a moment of lazy weakness and a lack of initial comment would love it if someone cleared that up for us?
CS majors know the time/space tradeoff, but they never get taught the 3rd, crucial, tradeoff of the set: comprehension!
But can it mine bitcoins?
"To prevent this day from getting any worse, I'll just read ERROR as GOOD THING" 1GJU8xLuDKDxEs4KLf8fAGyptoDsqvEsBT
That connector is digital only. The extra pins are for analog signal.
Help stamp out iliturcy.
It is pretty much impossible right now to get a GTX 680 unless one wants to get gouged due to the short supply.
When will nVidia get enough chips out so my searches aren't forever out of stock?
Mine sweeper is going to look great on this thing!
They're all dual-link (at least the connectors are - that doesn't guarantee the hardware behind them is). Single-link connectors have two blocks of nine pins on each side, and the middle block of nine pins is only on dual-link connectors. The top connector is dual-link DVI-D, while the others are dual-link DVI-I. A DVI-D port will not support a VGA adapter.
Does anyone know if this new card will be capable of taking advantage of double precision under CUDA as is the case with some of their other high end Tesla boards?
Except Nvidia has had SLI based multi gpu boards since at least the 8000 series, whereas 3dfx hit the limits of their Voodoo architecture, and required external wall power by the time Voodoo5 came out, and for all the extra hassle, you had a card that performed about as well as a GeForce256, but which also took a spot on your power strip. That's why 3dfx died, not because of SLI boards.
Mine bit coins, run Crysis, and ignite your weed.
But can it feel love?
This is a slashvertisement, nothing revolutionary is being reported here. Theyve been making double cards like this since at least the GX2 steps in the line, and maybe before.
while(1) attack(People.Sandy);
That's what I'm waiting for as well. Nvidia got pretty lucky with GK104. Most speculation is that it was intended to be the GTX 660 and GK110 was supposed to be the 680. However, GK104 was faster than AMD's fastest offering so why not sell it as a 680. The specs for GK110 "Big Fermi" are pretty intimidating and worth waiting for. I was also dissatisfied with 2 GB of memory for GK104, there are 4 gb cards coming out but they're around 800 bucks. GK110 will come with 4 gb standard.
I do have to hand it to Nvidia. The power requirements for the current 680 are very low and performance is quite impressive but GK 110 is going to be a monster...
Not likely, it's the third generation of their Fermi architecture and has the lowest power draw compared with the last 2 generations. I have a 480 and a 580 that are both still performing spectacularly.
Ummmmm yeah... no.
http://www.amd.com/us/products/desktop/graphics/ati-radeon-hd-5000/hd-5970/Pages/ati-radeon-hd-5970-overview.aspx
As a game developer, I can tell you that the only thing that significantly affects frame rate in a GPU-bound game is GFLOPs. And as the owner of a 3-year old PC with a stock power supply, I'm most interested in the "x40" cards, because those are the highest card you can install in a machine with a stock 350W power supply.
According to what I see on Wikipedia, NVIDIA apparently pulled a fast one this generation and re-branded some 500 series cards as the PCIe 2.0 x16 versions, while all the cards with impressive performance are PCIe 3.0 x16. The impressive ones get ~2x higher GFLOPs/W.
Old PCs like mine can't use PCIe 3.0, so this means the GF116-based GT 640 that gives 415 GFLOPs at 75W is still the fastest card that you're likely to find in a 2-3 year old PC with a game enthusiast who updates his card every generation. Compare that to the GT215-based GT 240 from 2009 which gets 386 GFLOPs at 69W, and you can see that there is ZERO reason to upgrade this generation, unless you also plan on upgrading your motherboard.
So yes, you can get a GK107-based GT 640 with 730 GFLOPs at 75W, but you have to upgrade your machine and get a PCIe 3.0 x16 motherboard. Boo, NVIDIA. BOO.
The last company to get all "multiple core happy" and "SLI On A Board" happy was 3dfx. Who NVidia bought out when they... oh yeah, crash and burned.
Whoops.
I'm pretty sure that both ATI and Nvidia(or one of their OEM partners at the time) have kicked out a 'logically speaking, this is two cards in SLI/Crossfire; but on one card!!!' product for basically every generation since the introduction of the respective GPU-linking features.
The hard part is the fancy tricks that make cooperation between two separate GPUs work at all. Once the vendors decided that they did, in fact, want that to work, the rest is constrained largely by the fact that people willing to pay $1k for a graphics card aren't all that common(especially now that motherboards with 4 PCIe 16x slots are quite reasonably available, so even one's excessive desires can be satisfied with cheaper, more common, single chip cards).
Well, it's also the fastest graphics card on the market... that's enough nerd porn to warrant an article. And it sets a new record as the first kilobucks card too.
Live today, because you never know what tomorrow brings
Can someone explain to me why general purpose CPU-memory interfaces don't have this kind of bandwidth to keep the newer 6 and 8 core monsters well fed with data and code to crunch?
I do not fail; I succeed at finding out what does not work.
...beowulf cluster of these!
Emulated on GPUs...
Sheesh, evil *and* a jerk. -- Jade
Unless he cooked it over an nVidia GPU, I don't see how that's relevant!
... just tell me how much it would cost for 4 of these with the SLI bridge thingie so I can make WoW run faster.
Support microSD: in a post 9/11 world, it is unwise to carry your data on media that you cannot comfortably swallow.
now I can play my xbox360 ports (that would run pretty decent with a geforce 8 series) at 180fps instead of 120, let me just shit myself
here nvidia, have 1000 bucks!
People in HPC buy these things in 10,000 lots. Now that you can put 4 of them in one server, that's going to happen more and more. It's not all about the videogames any more.
Help stamp out iliturcy.
The run length on PCIe 3.0 is quite limited (about a foot, I believe, though it's not given). There is an ePCIe spec. There are external devices that will do external PCIe 2.0 and are used for external card chassis or to host SSD storage with run length to two meters. While it's theoretically possible to do a laptop dock with one of these inside it, I don't see that happening anytime soon because there's not enough market for it. As the frequencies increase the distance a workable signal can propagate is reduced (very roughly).
For a while there was some talk about there not ever being a PCIe 4.0 spec because the run length would be down to mere centimeters - not enough even to get out to add-in cards. I see now that they've found a way - or at least think they have.
Help stamp out iliturcy.
Then this card is so advanced you don't need to be testing on it at all. If you want to target high-end gamers then get yourself a rig with PCIe 3.0 slots. Or at least one to develop on.
Help stamp out iliturcy.
About $4000, and $3000 more for the box to put it in.
Help stamp out iliturcy.
It isn't like they are doing this on purpose. The 680 is just a card that a lot of people want. The thing is, there's only so fast they can have them produced. TSMC is their sole supplier, and they only have one 28nm production line up and running. That line is still having some troubles (TSMC has been a bit over ambitious with its half-node plans and has had trouble at the beginning with them) so total yields aren't what they might like.
Then the real problem is just that everyone wants a piece. TSMC has a lot of customers who want 28nm chips. So a single customer can only get so many wafers per day. They aren't going to snub another company to try and fill nVidia's demand, they have to think long term and that means keeping everyone happy.
However demand for these things isn't infinite. It isn't like people are buying them, tossing them in a hole, and then buying more. As more people get their fill, it'll stabilize. How long I can't say.
This new card isn't likely to affect things much, because there won't be many of them made or sold. It is going to cost $900-1000. There aren't many people who will spend that kind of scratch on a gaming video card. It'll be a low production run.
If you ask me wall outlets were a very good idea. GPU's are the number 1 reason we have to upgrade our power supplies. And the necessity for power requirements to be correct means that bringing your own power supply can be the source if a plethora of bugs and crash's. Consistent power and precise currents with power hungry 3.5 billion transistor microchip's is a necessity. Pairing the power supply with the board means resolving a very real problem most end customers don't know exists.
I disagree and think it's quite valid to need a new internal power supply when the hardware requirement consumes more power... a seperate power supply means another point of failure and a pain in the ass for nothing really. I wouldn't however buy a new power supply because it doesn't have enough leads or the right type of leads (that do the same thing, but with just a different plug)
On a side note anonymous coward postings deserve score-ability (without which most people never see these posts. However valid they might be.)
Welcome to slashdot, get an account, it's free.
120 characters ought to be enough for anyone
There are cheap solutions that aren't pretty...
Like this is example you could use to hook it up to a laptop...
http://www.hwtools.net/adapter/PE4L-EC000A.html
120 characters ought to be enough for anyone
http://www.newegg.com/Product/Product.aspx?Item=N82E16814133347
$3998 for a card. The 690 is just the first non-workstation kilobuck card.
Tian-He 1A has 7,168 Teslas, and is the fastest supercomputer using GPUs. Titan (formally Jaguar) will have 18,000 GPUs. Amazon probably has quite a few.
The very top HPC projects may buy 10,000 lots, but most don't.
You must be stupid, as ACs can get rated/modded up.
4-digit UID my ass. Maybe 7.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
Is that why we have Thunderbolt?
Sent from my ASR33 using ASCII
At least for the present generation, I'm pretty sure that all the Tesla boards are 1 GPU per card. Nvidia supports, and sellers offer, arrangements with a fair number of cards per node(including external enclosures for expansion of systems that can't accommodate all those cards internally, 4 cards in a 1U connected by a special PCIe cable seems to have replaced the previous toaster-shaped 'deskside' chassis, with 'deskside' now handled by motherboards with loads of PCIe x16 slots); but the two-GPUs/one-card arrangements only seem to crop up on the gaming side.
I don't know why exactly this is, whether routing traces for 6 GB of RAM, per GPU, is bad enough with only one on the board, whether it is a thermal thing, or whether customer demand is such that it is cheaper to produce only single-GPU cards, and multi-slot chassis for the heavy users, rather than multiple flavors of card with various chip populations...
The article shows three DVI connectors, but are they single-link or dual-link?
-- Ed Avis ed@membled.com
About $4000, and $3000 more for the box to put it in.
$3500, you forgot the power supply
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Amen. I have a GT240 specifically because it's low-power. I'm not installing any graphics card so power-hungry it needs its own magical power connector. I did once and it was a mistake.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Except the GK104 is double-precision crippled, making them not-very-useful for HPC.
Excuse me, wtf r u doin?
Well..., I actually want the power slurping beast! 300 watts "400 with a hearty overclock" is fine with me for the biggest baddest card out there and that's what I'm expecting with GK110. I guess what I was trying to get at is that If the current 680 is any indication of performance per watt then GK110 is gonna be a whole lotta woah...
*Whom*
Tubal-Cain smokes the white owl.
Man on Dog? Oh wait, he dropped out of the race, right?
Tubal-Cain smokes the white owl.
That's what GPUs are used for these days.