Slashdot Mirror
NVIDIA GeForce GTX 780 Offers 2,304 Cores For $650
Vigile writes "When NVIDIA released the GTX Titan in February, it was the first consumer graphics card to use the GK110 GPU from NVIDIA that included 2,688 CUDA cores / shaders and an impressive 6GB of GDDR5 frame buffer. However, it also had a $1000 price tag that was the limiting specification for most gamers. With today's release of the GeForce GTX 780 they are hoping to utilize more of the GK110 silicon they are getting from TSMC while offering a lower cost version with performance within spitting range. The GTX 780 uses the same chip but disables a handful more compute units to bring the shader count down to 2,304 — still an impressive bump over the 1,536 of the GTX 680. The 384-bit memory bus remains though the frame buffer is cut in half to 3GB. Overall, the performance of the new card sits squarely between the GTX Titan ($1000) and AMD's Radeon HD 7970 GHz Edition ($439), just like its price. The question is, are PC gamers willing to shell out $220+ dollars MORE than the HD 7970 for somewhere in the range of 15-25% more performance?" As you might guess, there's similarly spec-laden coverage at lots of other sites, including Tom's, ExtremeTech, and TechReport. HotHardware, too.
Im a little surprised at people snivelling over 1000.00 video cards. I was in printing for 10 years back in "the day" and a supermac thunder card (with on board jpeg acceleration) was 2500.00
More Coors takes the pain of remembering away.
The cow says "Moo." The dog says "Woof." The Timothy says "Thanks, valued customer. We appreciate your input."
I must have crossed the border into adulthood somewhere back there because I would never pay that much for a performance uptick in a video game. I can get myself a nice new laptop for that cash, and it would be still be proficient at 90% of today's games.
"Here Lies Philip J. Fry, named for his uncle, to carry on his spirit"
Ten years more and it will be one core per pixel. That's insane.
Is anyone else getting real tired of companies purposely crippling their high end products in order to sell them for less money? It's like openly broadcasting that their cards cost way too much to begin with.
It's a question of tolerances. The chips that come out of the fab are not 100% perfect. The designs are amazingly complex, and they usually contain some defects in the manufacturing process. If they don't meet the high-end specs, maybe they can disable the broken cores, relabel it as a mid-range chip, and sell for less money. It allows the yield to be higher and it lowers the price for ALL of the products.
"Here Lies Philip J. Fry, named for his uncle, to carry on his spirit"
Implementation trumps architecture. There's a reason nobody who's interested in power efficiency, noise and/or heat uses AMD products.
They aren't necessarily purposefully crippling, though that sometimes happens. This allows them to sell chips that have some manufacturing defects; they turn off the bad shaders. The more chips from each wafer they can sell, the lower the price of each chip.
While this does happen it is also a way for them to increase the yield of the chips. These are huge chips they are building and the chances of bad cores on the die is rather high. So instead of junking the chips with bad cores since they cannot be sold as the ultimate high end, they create a cheaper product with cores disabled. Not all disabled cores will be bad but this does help them improve manufacturing efficiency. Also, they might have serious manufacturing issues producing these huge chips so it might cost them an arm and a leg to build them right now. So they set the price abnormally high to control the demand until they can iron out the manufacturing issues and improve the yield.
Since WoW is CPU-bound the answer is a clear, resounding "it depends".
Sounds more like a regional thing.
Here in upper/middle NJ, $1000 gets you a pretty bare-bones apartment in most towns. Maybe a 2-bedroom in worse area.
Unless you're splitting rent with someone at $1000 doesn't carry much weight here.
They seem to be using the terms Cores and Shaders interchangeably. is a Shader a core ?
We are Dead Stars looking back Up at the Sky
If this was a previous generation where AMD was actually still competitive, Titan would have been the high end part, and it would have cost $500 instead of $1000. The part known as GTX 780 would have been a slightly depopulated part capable of 90% the performance for a 20% savings or so and the rest of the line would have fallen under those two. Since AMD is no longer really a threat in the high-end GPU space, Nvidia can literally maintain the MSRPs of the old parts as if the new parts are merely higher performing extensions of the previous generation without any downward pricing pressure on anything.
Noise? Wouldn't that just depend on which manufacturer packages the AMD GPU?
Imagine a Beowulf cluster of these babies rendering images of Natalie Portman covered in hot grits!
In GPU terms, yes. The shaders and cores are very different between AMD and NVIDIA (that's why AMD can have 1536 and compete with a HD 7970 with 2048 shaders).
Anything more the $250, and it's likely you won't find any game out there that can't be played well with a $250 card.
Great, yet another reason I wish I was Canadian.
NJ has a high cost of living. Though it varies: if you want to leave in Western NJ near the PA border it's a little cheaper. Or if you move way down to South NJ it's a little cheaper still. Though hadn't seen a nice apt for $500.
But unfortunately there are a lot more jobs in northern NJ, and it's not worth driving 1+ hours each way let alone 2.
Same in most major US cities. But people have to have a place to live but they don't have to play video games. If its too much, don't buy it
my karma will be here long after I'm gone
AMD/Nvidia/inbreadkitty comes out with faster, more expensive video card, News at 10. This happens yearly, why is this on /.? I understand being on the hardware websites, but not here
my karma will be here long after I'm gone
Until I was asked to write a few tech. articles on bitcoin and other virtual currencies last year, I didn't really pay a lot of attention to them. But I've learned that high end ATI video cards are pretty much the "engines" required for any respectable bitcoin/litecoin mining rig to work successfully.
(As a rule, nVidia cards have been ignored as "not as good of performers as ATI" for this specific use -- though I wonder how this GTX 780 would do?)
People building these mining rigs generally cram 3 - 4 of the cards on one motherboard, and run several identically configured machines at a time -- meaning a pretty hefty investment in video boards. It makes me wonder if this isn't really a significant reason for the sales of the more costly models, as opposed to the audience you'd assume was buying them -- 3D gamers?
This is where you don't understand that maybe 1% of PC gamers will buy this card. Of course these are the same people that will drop lots of money on bling like LED lights. Most gamers will spend much less and still have a PC that beats your 8 year old console as it hasn't been/can't be upgraded when it comes to hardware.
Well, there's spam egg sausage and spam, that's not got much spam in it.
Yes, but people living in higher cost of living centers generally have higher relative incomes as well. So it is less of a proportional hit on the budget than elsewhere.
As long as Nvidia keeps crippling double-precision performance on their (non-Tesla) cards, I'll keep buying AMD.
One of the highlights of the GTX Titan was that the card did double-precision floating point at full speed, just like one of Nvidia's Tesla products. That's no longer the case here - the GTX 780 performs double-precision at 1/24 of normal rate, just like a standard desktop GPU.
Welp, it took about 5 years.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
I'm sure you meant "except their APU platform"
"When life gives you lemons, don't make lemonade. Make life take the lemons back!" -- Cave Johnson
Hell, I could pay both my rent and electricity bill for two fucking months for the price of one GTX Titan card.
Then you're not the target market.
Some people buy $250,000 cars, some people buy $1,000 video cards, some people pay $2500 for Mac with the same hardware as a $1000 Windows PC. To most of us, they're just a curiosity.
Does AMD support VDPAU these days? Because VA-API support is mighty poor in my experience. Broke down and bought a card when I had a perfectly find integrated one for my TV box because I can't get VA-API to work with mplayer. There is a source version that supports it, I couldn't get it to compile cleanly, though.
I also had a ton of trouble with the "legacy" vs new ATI drivers (the computer was low end, but only a few months old when this nonsense happened). Not sure what caused that split, but it was hell to get working on Ubuntu. Left a bad taste in my mouth all around. Think they resolved it, but it was just yet another barrier between me and what is normally a flawless experience.
md5sum
d41d8cd98f00b204e9800998ecf8427e
Well, kind of... or not really... not in the traditional sense. I can't speak for AMD but in Nvidia architectures those 2000+ "cores" are clumped into very wide (32+) groups which all share a single instruction decoder and all the parts that go around that. Are they really individual "cores" if they all have to execute the exact the exact same instruction in lockstep?
The chips that come out of the fab are not 100% perfect.
While this may be true for these graphics cores, I don't think it's necessarily true for Intel's CPU chips. I think they have their design so refined that their yield is close to 100% for all but the highest density cores.
Otherwise they simply would not be able to offer multi core chips. Maybe someone in the know could comment on this.
Take the cheese to sickbay, the doctor should see it as soon as possible - B'Elanna Torres, "Learning Curve"
FWIW, they did the same thing on the PS3. Disabling one of the SPU cores to get higher yields. Even on machines where all the SPU's passed, they still had to disable one though to "standardize" performance... so sometimes functional chips are actually crippled to meet demand for lower specs but it's more often a factor to attain higher yield. Intel did "crippling" on functional Pentiums at first to meet Celeron demand (before actually making a new die for celerons). It's a little bit of both to be honest.
We do love our big numbers, but there are limits to what our eyes can perceive in FPS. What does this mean for real world applications like video encoding and password cracking? How long do we anticipate having to wait for tech like this to get affordable? Also, how does this compare to the nVidia Tesla, the current gold standard in password cracking?
I saw only one reference to nVidia Tesla (and no references to password cracking or video encoding) in those reviews (@Tech Report), and it might be damning:
Use my userscript to add story images to Slashdot. There's no going back.
GPU manufacturers have a tendency to use the word "core" to mean "one ALU in the middle of a vector unit". It's not really very different in principle from saying that an AVX unit is 8 cores, though, so you have to be careful with comparisons.
If you look at the AMD architecture for each compute unit, it's not so different from the cores you see on the CPU side, so it's much more fair to call the 7970 a 32 core chip. The way that a work item in OpenCL, say, or a shader instance in OpenGL maps down to one of those lanes is as much an artifact of the toolchain as of the architecture.
Pretty much seems like the entire point of this site. News for nerds...did we as nerds stop caring about hardware suddenly? I was not included in this memo.
AMD sold tri-core processors for a while - most if not all of those were just quadcores with one core either non-functional or intentionally crippled. Pretty smart move.
Seriously, for some people, gaming is their hobby and that kind of money is not that much when you talk what people spend on hobbies. My coworker just bought himself like a $2000 turbo for his car, to replace (or augment, I'm not sure) the one that's already there. He has no need for it, but he likes playing with his car.
Now that you, and most others, don't want to spend that kind of money is understandable and not problematic. There's a reason why companies have a lineup of stuff and why the high end stuff is just for those with plenty of money. It also doesn't scale linearly since the higher end something is, the less units get sold, and so the more the fixed costs influence the unit cost.
However don't hate on it. That you don't wish to spend that kind of money doesn't mean that nobody should. Also you should be glad people do: The expensive parts fund the cheap parts. They can recover more R&D costs on these units, letting them sell lower end parts for less, since lower end parts are the same tech, just less of it.
Alternate description: "Nvidia lowers the cost of standard desktop GPUs by not including features for high-speed high-accuracy functions that serve no purpose in gaming".
Where have you been since 1997? Since then you've been able to consistently spend roughly $200 per video card and play the latest games at acceptable settings. That's why people cast a sneer at a $1,000 card let alone a $650 card that in of itself can build a really fast computer.
If you wanna get rich, you know that payback is a bitch
If you want higher resolutions and frame rates, you need more powerful GPUs to handle it. For example moving to 2560x1600 or to 120fps doubles the pixel requirement over 1920x1080@60fps. So whatever amount of power you needed to achieve 1080p60, double that for either of those targets. 4k will require a quadrupling, and 120fps 4k would require 8x the power.
All this is assuming you are getting 60fps in the first place. Now maybe you are fine with trading off lower frame rates, or lower resolutions, that's all up to you. If 720p30 is your target, you can get away with a whole lot less power. However that doesn't mean that nobody wants to target higher resolutions or frame rates.
There are also other visual quality settings to consider, like anti-aliasing and so on that can require more power. Depending on what you are targeting with that, you can need a lot of power.
Personally I really find frame rates much below 60 pretty annoying in most games. I really like the feeling of fluidity you get. 120 fps is even better, but the monitor I normally use doesn't handle that. Well maintaining that 60fps at a 2.5k resolution is not a trivial feat. I don't think a $250 graphics card would do that for most games.
Their problem is that the cost of implementing large-die processors is getting extremely expensive compared to how it used to be. We used to see previous-generation processes used for high-end cores because the maturity overcame the extra cost of the large die. But now that large dies are prohibitive (and assuming prices cannot grow), the graphics makers have no way to improve performance until the new 20nm process is released.
Nvidia has an out because of their vase supercomputing following with Tesla and Cuda, so they can afford to make an outlandish GPU like GK110 and charge $3500 for it! This gives them a path to offer a "new" top-end card with more performance, but since it's powerful enough to *almost* cost them a Titan or Tesla sale, they still want to charge a premium. In the end of the day the ONLY chip Nvidia makes that is mass-market affordable is GK104, so that's why AMD has no response to the GK110.
AMD is plenty competitive. They revamped their drivers and improved silicon/clocks to make the HD 7970 GE the fastest graphics card on the market, and priced it lower than the GTX 680. The GTX 780 (And 770 to be released soon) are a direct response to that performance acheivement.
Man is the animal that laughs.
And occasionally whores for Karma.
Nope. The NVIDIA GK110 has in it's design 16 clusters of 192 CUDA cores. And on the Titan, any two of those clusters may be defective and are turned off giving you a total of 2688 CUDA cores. In my opinion, that's a nice way of increasing their usable yield. On the GTX 780, they can have 4 defective clusters giving a total of 2304 CUDA cores. So chips that they would otherwise trash can still be used in a nice high end card. Isn't redundancy in design nice?
You're paying for the luxury of what appears to be the world's first high-end video card with a built-in speaker. Nvidia finally reached the point where the polygons their products could produce exceeded the nominal human capacity to perceive them, so now they've added the ability to hear the extra polygons you can't see, as ultra-soothing HD Brown noise! The only side effect is that it reduces your available gaming time by increasing the number of bathroom breaks you need to take.
cutting silicon with laser to cripple it does NOT lower the cost.
Who logs in to gdm? Not I, said the duck.
1) Drink Coors
2) Drink more Coors to forget you are drinking Coors
3) PROFIT!!!
At $650, the GTX 780 costs less than one monthly rent payment for me.
Uhm... no. They're cutting double-precision floating point performance through the driver! Hardware is otherwise the same.
and still have a PC that beats your 8 year old console as it hasn't been/can't be upgraded when it comes to hardware.
Reply to This Share
This is something I don't get. Why does the existence of both PC's and console have to be a "competition" that "just one" wins. There's room for both.
SOmetimes I think PC gamers prefer playing benchmarks for bragging rights than actually playing games.
Then you are essentially a bit like a wheel chair bound person arguing with experienced runners that you don't need >100 euro running shoes, 20 euro ones from supermarket are good enough. They are for you because you don't run.
Run battlefield 3 at 4k with 3d glasses on "standard 3d card" of your choice. Run it again on titan. This isn't audiophile stuff. There is a very real and easily quantifiable difference known as "frames per second".
If I had mod points left this would get all that I could give
AMD did this back with their Phenom X3's. There was such a high demand for the low cost triple core processors that they disabled a core on some four core processors and sold them as X3s till they ramped up production. I was lucky enough to get one of the four cored processors and was able to re-enable the fourth core through my bios.
It's not earthshaking news but it's an announcement that a significant number of Slashdot readers will be interested in. Looks worthy to me.
Intel doesn't currently disable cores in their CPUs to create lower-grade chips. They do speed grading however, and they may do cache disabling - if they do it would be because some of the cache fails testing. They did more of that sort of thing in the past; your FPU-less 486SX was likely to be a 486DX die with a defective FPU. AMD does disable CPU cores. That's where the 3-core Phenoms came from, and where the 6-core FX comes from now. I believe the 4 core FX is normally a separate design but some may be 8-core die with half the chip turned off; the end user gets the same performance either way. As others have already pointed out, this kind of thing has been standard business practice in graphics for many years. The down-spec chips may have out-and-out defective portions, or they may fail to meet their power and heat specifications if everything is turned on. Or they may be perfectly good chips that AMD or NVidia sells as lower-end chips because they don't have enough demand for the expensive ones.
They had both a three-core Phemon and a three-core A6 APU. Neither is a current design though it may still be possible to buy them. AMD currently sell six-core processors, which are really 8-core designs with one core pair turned off. Their current designs use pairs of CPU cores with some shared resources so the number of cores will always be an even number; I don't think it's likely that they will ever release a part that disabled only one of the two cores in a pair.
Does AMD support VDPAU these days? Because VA-API support is mighty poor in my experience.
Yes, actually! A month or two ago, AMD released VDPAU support for their open source driver.
Bizarrely, the closed source driver is still XvBA only.
Broke down and bought a card when I had a perfectly find integrated one for my TV box because I can't get VA-API to work with mplayer.
I did actually get VA-API / XvBA working on my AMD system, but it could only do h264 and MPEG2. You could forget xvid, forget advanced GPU deinterlacing, etc. Since it was a weak E-350 box, that left it able to play the highest bitrate bluray rips, but not broadcast TV (MPEG2, 1080i). Replaced it with an Intel Atom / Nvidia ION2 box.
This is precisely the kind of story that was considered slashdot's raison d'être in 1997... you know, news for nerds?
HSJ$$*&#^!#+++ATH0
NO CARRIER
As long as Nvidia keeps crippling double-precision performance on their (non-Tesla) cards, I'll keep buying AMD.
they aren't the only ones: http://youilabs.com/blog/mobile-gpu-floating-point-accuracy-variances/ (although this is targeted mainly at mobile gpus, I suppose the same or very similar can be said about desktop GPUs)
An eight year old console will play the latest games better than an eight year old PC.