Slashdot Mirror
NVIDIA GeForce GTX 690 Benchmarked
MojoKid writes "NVIDIA has lifted the embargo on benchmarks and additional details of their GeForce GTX 690 card today. According to a few folks at NVIDIA, company CEO Jen Hsun Huang told the team to spare no expense and build the best graphics card they possibly could, using all of the tools at their disposal. As a result, in addition to a pair of NVIDIA GK104 GPUs and 4GB of GDDR5 RAM, the GeForce GTX 690 features laser-etched lighting, a magnesium fan housing, a plated aluminum frame, along with a dual vapor chamber cooler with ducted airflow channels and a tuned axial fan. The sum total of all of these design enhancements results in not only NVIDIA's fastest graphics card to date, but also one of its quietest. In the performance benchmarks, NVIDIA's new dual-GPU powerhouse is easily the fastest graphics card money can buy right now, but of course it's also the most expensive." The GeForce GTX 690 has been reviewed lots of different places today, Tom's Hardware and AnandTech to name a few.
Finally I can play minecraft the way it was meant to be played!
Tomshardware is showing GTX beating ATI by 50 - 200% in every benchmark. Anandtech shows the opposite with ATI still winning under the same games? Anyone else notice this?
Does Toms Hardware or Anandtech get paybacks from either company for biased remarks?
http://saveie6.com/
The GTX 680 and 690 have turned out to be pretty spectacular. The most impressive aspect is the relatively low power consumption for a high performance card.
I'm still waiting for the GK110-based "Big Fermi" due out Q3. Considering how well the 680 and 690 have performed the Gk110 will be a monster, probably power hungry but still a monster. Nvidia really hit gold with their latest generation, it is speculated that the current 680 was intended to be the 660 until it outperformed AMD's top offering. Can't wait to get my hands on a 4gb GK110.
Actually high performance computing has created more demand. Nvidia GPU's are being used in massive supercomputers using OpenCL and CUDA. "AMD GPU's support OpenCL." There are a many more people who are interested in the latest and greatest GPU than you may think, specifically on a news for nerds site. So yeah, sweat article and thanks for the heads up about the new benches MojoKid.
argh... *Sweet
My bad.
Whatever happened to the Slashvertisement tag?
I got an AMD 6870 over a year ago ($150), and it's played everything I've thrown at it just fine with maxed graphics. Skyrim, Witcher 2, etc play without any stutter and look wonderful. All on an AMD 965 (3.4 Ghz X4) CPU from the year prior.
I'm just trying to figure out what I'm missing by not spending 5x that price.
According to Semiaccurate there's a mask design flaw in the GK104, which has caused poor yields. Less than 10,000 GTX 680s shipped worldwide, even though it's been released a month ago.
http://semiaccurate.com/2012/05/01/why-cant-nvidia-supply-keplergk104gtx680/
Yeah, but the "I can simulate the 14000-body problem in real time" fad is alive and kicking.
well it depends on what you want things for
basically I don't really see much difference in the graphics openGL/DX11 side of things but this was very interesting to me :
http://www.anandtech.com/show/5805/nvidia-geforce-gtx-690-review-ultra-expensive-ultra-rare-ultra-fast/15
regards
John Jones
I'm looking forward to seeing a few Blender benchmarks with this little puppy. I mean, wow, a guy could probably run a high-end render farm on a couple of domestic electric circuits, now.
I always harp about this but in a couple of years there will probably be a game that requires that much power. However by that time there will be a $150 card that can run it.
Did you know 80 to 90% of the moderators on slashdot wouldn't recognize a troll even if one dragged them under a bridge.
The only game my Radeon couldn't handle at top graphics was FFXIV. And the only person I've ever met that got XIV to benchmark at max graphics was running on a brand new $2000 rig with SLI. 95% of us won't need a card this powerful, but I bet more than 5% of us sort of want one anyway.
Occasionally living proof of the Ballmer peak.
who is going to pay $1000 for a piece of hardware with a halflife of maybe one year? this card is really worth about $400 at most.. and the 680 should be $200. what games actually take advantage of this? there are hardly any pc games worth playing nowadays :\. It's too bad too, because I LIKE new graphics hardware. it's always fun to play with, but at $1000 I can't justify it.
Yes, except the GTX 6xx series is slower at CUDA processing than its predecessors. This is a gaming product. Nvidia did this on purpose, sacrificing some compute speed in favour of rendering performance and power efficiency. They also artificially limit double-precision FP speed on consumer boards, to steer professional users toward the Quadro.
As a result of this hobbling, GPU computing hobbyists tend to gravitate toward the Radeon, which has outperformed the GeForce in OpenCL for a few years now, in both performance-per-dollar and performance-per-watt.
-Billco, Fnarg.com
There is zero actual evidence that there is going to be a "GK110" this year, or that if there is it will be a high end part (bigger numbers in their internal code names don't always mean higher end parts).
I see people all in a lather about the supposed amazin' graphic card that is up and coming, and lots of furious rumors, but nothing in the way of any proof. I also can see some fairly good arguments as to why nVidia would NOT be releasing a higher end card later on (excluding things like Teslas and Quadros, which are higher end in a manner of speaking).
Speaking of Teslas and Quadros, that may be all that it is: A version of the hardware with a redesigned shader setup to give higher FP64 speed. As it stands the card is quite slow at FP64 calculations compared to FP32. It could be 50% of the speed, in theory, but is more like 1/16th. Basically it seems to be missing the necessary logic to link the 32-bit shaders together to do 64-bit calculations for all but a fraction of the shaders. Maybe to protect their high end market, maybe to keep size and heat down (since it does take additional logic). Whatever the case a Tesla/Quadro version with that in place would have much improved FP64 speed, and thus compute performance for certain things, but be no increase to gaming at all.
So I think maybe people need to settle down a bit and stop getting so excited about a product that may not even exist or be what they think, and may not launch when they think even if it is. Chill out, see what happens. Don't get this idea that nVidia has something way MOAR BETTAR that is Coming Soon(tm). You don't know that, and may be setting yourself up for a big disappointment.
Oh that's right: Video games. You know, the thing it was made for.
The GTX series are nVidia's gaming cards. They are made for high performance when you wanna play 3D games. They aren't made for compute performance. That is not to say they cannot handle compute stuff, just that it isn't what they are primary designed for. So the kind of compute stuff they are the best at will be more related to what games want.
Their compute products with be the Teslas. They are made for heavy hitting compute performance of all kinds. If you are after purely GPGPU stuff, they are what you want.
nVidia seems to be separating their designs for the two to an extent. Still common over all design, but concentrating on making the desktop GPUs more efficient, at the expensive of high end computer features (like Integer and FP64 power), and the workstation/compute cards good at everything, even if they need beefier power and are louder.
I'm ok with that. I buy a GeForce to play games, not to do high end GPGPU stuff. We buy Teslas at work for that.
Also, there's a shitload of other things out there GPGPU wise that are FP32, and the 680 really is killer at that. Does a great job accelerating video encoding and the like.
To be fair, FFXIV runs horribly on every system because it's CPU bound. It doesn't use multi-cores efficiently, and the game itself is pretty terrible.
Other than FFXIV, most MMORPG games are designed around 10 year old hardware because the more people that can play it, the more people they can fleece. Exceptions are the new star wars MMORPG which requires more CPU power than typical games, but you can get away with 5 year old graphics cards.
The pattern often seen in games, at least since the advent of Intel "onboard video" is to make it "work" on the oldest thing that many people have:
http://store.steampowered.com/stats/
http://store.steampowered.com/hwsurvey/videocard/
Only in the last two months have Intel "GPU"'s even been on the board, and even that, the 3000 series brings it up to around 10%. So if onboard video is starting to be "Good enough", you'll start seeing dedicated "budget" GPU's disappear. Only a handful of games actually work "well" on these GPU's, the vast majority of games still only work on onboard GPU's with all the settings set to the most crippled mode.
But if you're into HPC, this is not the card for you. You want a 685/695 which won't be released until 1st quarter next year.
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
I don't care for Anad's benches much because they seem to like synthetic compute benchmarks. That is really all kinds of not useful information for a game card. I want to see in game benchmarks. If any compute stuff is going to be benchmarked, let's have it be an actual program doing something useful (like Sony Vegas, which uses GPUs to accelerate a lot of what it does).
Personally I'm a HardOCP fan when it comes to benchmarks. Not only are they all about game benchmarks, but they are big on actual gameplay benchmarks. As in they go and play the game, they don't run a canned benchmark file. This does mean that it isn't a perfect, "each card sees the precisely equal frames" situation, but it is far more realistic to the task they are actually asked to do, and it all averages out over a play session. I find that their claims match up well with what I experience when I buy a card.
http://hardocp.com/article/2012/05/03/nvidia_geforce_gtx_690_dual_gpu_video_card_review is there 690 benchmark. It's a selection of newer games, generally played with triple head (the game displayed across three monitors at once) on a 690, 2 680s SLI'd and two 7970s CF'd.
So FP64 performance went from 1/8th FP32 performance in the 500 series to 1/24th FP32 in the 600 series? I, for one, would love it if using doubles in OpenGL 4.x didn't suck so much. I write visualization software with planet-sized ranges and having that extra precision on the card would be quite nice.
-SaNo
You mean BitCoin mining don't you? That is the only thing you can consider "faster" with OpenCL, while actually using it for a real purpose.
"When life gives you lemons, don't make lemonade. Make life take the lemons back!" -- Cave Johnson
It's all about the monitor. Are you running at least 1080P? HDTV has done a horrid thing and effectively keeping displays at or close to HD res there are very view monitors that will do more than 1920*1080 some pack some more vertical pixels in at 1920*1200 with a 16x10 ratio. Anyways it's rather easy for a few year old card to run maxed out if you only running an average display 1366x768 was the most predominant per http://www.w3schools.com/browsers/browsers_resolution_higher.asp and that's really not a lot of pixels.
No sir I dont like it.
when getting a new graphics card every few years was like a new console launch to me. I get that they're being used by statisticians but seriously, what do gamers do with these? The only game that comes close to taxing a $150 graphics card is Crysis 2, and even that's not doing much... Maybe nvidia could put more effort into making these easy to program for so we'd get better games cheaper? I think we've hit the limit on graphic quality, if only because it's too much work to do the art assets...
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
I've been watching my UPS power load meter since I upgraded from a GTX 560 to a GTX 680. I'd estimate the 680 uses a bit less than half the power of the 560 when idle. At peak usage the 680 uses more, but only by a hair.
I was never happy with the 560 in general. The 3D performance was surprisingly glitchy at 1080p. Even though I wasn't too keen on trying NVIDIA again after that, I gotta admit they won me back with the 680.
There's no -1 for "I don't get it."
I know some people are very passionate with their hobbies but even if I were a billionaire I just couldn't justify spending around $1000 just for a few more frames per second. Just seems very strange.
It sucks. Now that we have graphics cards capable of effectively pushing >3 million pixels effectively, we don't have the high end CRT monitors that have that kind of resolution any more.
no thanks, summer's coming.
Incorrect.
If you're into HPC, you want an Intel MIC. Or if you're REALLY into HPC you already have a few. All the love is currently leaving CUDA for the Knights Ferry|Bridge MIC cards. They kick the hell out of the CUDA and OpenCL cards.
Though they only barely give the Tilera cards a run for their money on a few apps. (Then again... as few apps out there that can run without floating point, who cares??)
Have you tried OpenCL on AMD GPUs recently? OpenCL on AMD GPUs does not work. The drivers are buggy as hell.
AMD pays lip service to GPGPU.
One question though: If I can play Skyrim with all settings to max at 1920x1200 with a GTX 560, what is SLI of two GTX 690's needed for?
I hit exactly *one* actual driver/APP bug in 2 years. All other times it was really a bug in my application or compute kernel. And we're testing against SDK 2.1 through 2.6 on catalyst from 10.3 up to 12.4 on cards ranging from a 4770 to a 7950.
So unless you provide something a bit more substantial than "waah", I call bullshit.
This is a 2560*1440 monitor for $320. The early ones had higher quality internals, and could actually run at 100hz at that resolution. They're shipped direct from Korea.
People saying they're running on maximum settings, without mentioning the pixel count are being disingenuous. The above monitor pushes over 3.5 million pixels. 1336*768 is about 1 million.
That's funny, because Newegg shows all of their Geforce 680 and 690 stock as being sold out. It seems that they are selling pretty well for something that nobody wants.
Actually high performance computing has created more demand. Nvidia GPU's are being used in massive supercomputers using OpenCL and CUDA. "AMD GPU's support OpenCL." There are a many more people who are interested in the latest and greatest GPU than you may think, specifically on a news for nerds site. So yeah, sweat article and thanks for the heads up about the new benches MojoKid.
They did not focus on the general purpose computing circuitry so that power and heat could be reduced. The main focus was on gaming. Seperate cards will be created specifically for computing.
I'm sorry but this is NOT insightful, because NOBODY is gonna use GEFORCE cards in supercomputers, as their FP is just too shitty. if you would have been talking about an AMD since that company is pushing their lines towards a single design (they are even switching their GPUs from VLIW to Vector to get more GPGPU performance) then yes, you would be correct. but this is Nvidia we are talking about here who have been consistently separating their GPGPU line (Tesla) and their gamer line (Geforce) and the two lines are NOT identical, nowhere even close.
So I'm sorry friend but this is NOT for supercomputers, this is for showing you have the largest ePeen on the benchmark boards or triple monitored Skyrim. If you wanted to argue they are testing designs on the gamers, kinda like how RHEL uses Fedora as a testbed? that might be possible, but nobody is gonna buy geforce cards (which the FP has gone down not up with these latest cards) to slap into supercomputers.
ACs don't waste your time replying, your posts are never seen by me.
Wake me up when you can actually buy one of those things. As for product which is 5 years in production it will be absolutely awesome. Like DNF surely was ;>
after looking at all the other charts for a single card it is indeed the fastest. interesting that for skyrim the crossfire places almost bottom and even blow the non crossfire solution.
for the compute section likely this is an architectural difference and demonstrating the same calculations that make AMD rock bitcoin mining too.
More like they only had three of them in the first place so now it looks like they are selling out everywhere when really there is just a huge supply problem.
I've poked around in XIV a bit lately and the patches in preparation for the "end of the world" and reboot to version 2.0 next fall have actually made the game a lot more bearable and, dare I say, fun than it was when it first came out. My best friend played it much more extensively, and she had a custom built rig from Square Enix she won in a contest a few years ago that choked and sputtered on XIV until we replaced the motherboard, processor, and video card. When even their own house built gaming systems won't run it without extensive upgrades....
Occasionally living proof of the Ballmer peak.
And yet it beats the crap out of ATI where it really counts: actual gaming. Good on you linking straight to page 15 of a technical review to find a single graph with which to troll, though. I'm sure you're very proud of yourself.
You mean @HOME projects, don't you? Nobody uses BitCoin.
I'm starting to think GNU is the problem with "GNU/Linux" these days.
Correction: this is the most expensive GAMING card you can buy. The price of a professional card can be up to US$10,000.
No, it's really much slower, particularly in double-precision:
Nvidia cripples GPGPU in Geforce GTX 680
Benchmark Results: Sandra 2012
NVIDIA GTX 680 Reviewed: A New Hope
Excuse me, wtf r u doin?
Uh, right. This is Newegg we're talking about, not some smalltime Ebay seller. They have hundreds of each model.
Well, when you test a gaming card by running GPGPU stuff on it, when nVidia specifically sells GPGPU cards, maybe you are running the wrong test.
APK likes to ask for responses to the same things over and over. Maybe he just likes the responses?