Kepler Nvidia GeForce GTX 780

[Olivon]

but based on feedback from manufacturers GTX 680 easily exceed the HD 7 970 is no problem

http://translate.google.fr/translate...m%2F18768.html

[Andrew LB]

[Tim]

http://translate.google.fr/translate...m%2F18768.html

Aha!

GK104 released the name really is FeForce GTX, 680 GTX 670 of Ti only NVIDIA confuse the opponent's tool.

Wait, what?

[boxleitnerb]

The name 670 Ti was thrown out there by Nvidia to confuse the competition. Come on, that was easy

[SimBy]

Why would that confuse competition?

[boxleitnerb]

Ask the author, I'm just translating.

[WeaponX]

When unicorns are involved things get confusing.

[prava]

Wouldnt it make more sense to have put 7950 in that graph instead of 7870? As that seems (die size) wise where there going to be trading blows?

Nope. That comparison is decent because it compares the full fledged chip in each instance. If you were to compare a 7950 it wouldn't be as efficient as a 7970, at least surface wise. As it is, they compared every top bin, which is the think to do when comparing chip efficiency based on transistor count / die space.

[GoldenTiger]

When unicorns are involved things get confusing.

Yeah, I know man... for example, why is this unicorn attacking things when it's supposed to be making video cards?



Was the mouse attacking the factory? Was it trying to steal a Kepler sample?? I'm really confused!

[Tim]

The name 670 Ti was thrown out there by Nvidia to confuse the competition. Come on, that was easy

Gah, of course, I was merely quoting the ridiculousness of the translation.

[SubZero.it]

So..there is nothing sure even if the launch "should" be close

[Dimitriman]

If Nvidia doesn't make a big Kepler then this round is shaping up to be much like the last where Nvidia leads by 10-15 pct on the flagship. The difference is that unlike Thermi, now they have a very efficient chip much like Amd.
What I am still left guessing is how they achieved such great efficiency, something must have had give. My guess is that GK104 will be inferior in GPGPU computation than Fermi and will lose a lot of its features that are oriented towards the professional segment and stream computing. This chip will probably be geared only for gaming which is likely how they managed to save so much die space and still make a powerful chip.

[zalbard]

If Nvidia doesn't make a big Kepler then this round is shaping up to be much like the last where Nvidia leads by 10-15 pct on the flagship. The difference is that unlike Thermi, now they have a very efficient chip much like Amd.
What I am still left guessing is how they achieved such great efficiency, something must have had give. My guess is that GK104 will be inferior in GPGPU computation than Fermi and will lose a lot of its features that are oriented towards the professional segment and stream computing. This chip will probably be geared only for gaming which is likely how they managed to save so much die space and still make a powerful chip.

I don't think they would've done that. Unless they are specifically making a completely different chip for their professional cards...

[highoctane]

New uarch ushers in new/different standards in power and performance for Nvidia, similar to how Intel's conroe core2 based units changed allot of dynamics from the P4 uarch.

Nvidia is very heavily invested in gpgpu, there's now way they will sacrifice in that area especially now with AMD and Intel taking the whole gpgpu more seriously.

[Lanek]

Does really peoples buy a GTX580 for professional computing ? or do they buy Tesla quadro ?

[Kuroimaho]

Same die, that's what matters.

[kuroikenshi]

^^^ the difference??... I am no expert on it, I've only used Quadro cards with my friends doing rendering on Rhino. So I'll let someone else who's well versed on this answer your questions.

[Darakian]

Does really peoples buy a GTX580 for professional computing ? or do they buy Tesla quadro ?

They go tesla or quadro for full FP64 support. The geforce line is gimped in FP64 calculations.

[highoctane]

The biggest issue I see when trying to run a consumer grade video card with professional software is consumer hardware usually lacks the software developers certification/validation which is needed to enable specific hardware accelerated features.

So while the consumer card should be able to do everything the professional cards do without the certification/validation you can't take advantage of some or all of the hardware accelerated features possible for a software package.

[UrbanSmooth]

Do you have links to some data that says VRAM use is over 2GB in Surround, barring isolated examples like "game X with texture pack Y at 8X MSAA". Point there being I can probably find settings where 3GB are exceeded, or 4GB for that matter.

I've done a lot of surround gaming at 57X10 and 60X10 (most common res for surround, 1080PX3) with 1.5GB, 2GB, and 3GB. For the most part, 1.5GB works at 4X MSAA. I don't think I've ever seen problems at 2GB or 3GB.

Not to mention for all of 2011 pretty much everyone and their brother was saying "2GB is the thing to have for surround/eyefinity".

If 680 ends up at 2GB, that won't be a factor in my buying decisions.

http://www.xtremesystems.org/forums/...=1#post5068055

It will in mine. I'll stay with Nvidia, but, wait for the non-reference cards with 3, 4GB+ VRAM. (See GTX 580 3GB)

[Pontos]

The biggest issue I see when trying to run a consumer grade video card with professional software is consumer hardware usually lacks the software developers certification/validation which is needed to enable specific hardware accelerated features.

So while the consumer card should be able to do everything the professional cards do without the certification/validation you can't take advantage of some or all of the hardware accelerated features possible for a software package.

Couldn't a simple DeviceID spoofing solve that issue?

[highoctane]

Couldn't a simple DeviceID spoofing solve that issue?

It used to be easy with rivatuner but eventually you simply cave to buying the hardware you need to avoid the headache and hassle since you need to have validated/certified hardware as well as specific revision certified/validated drivers. I'm sure if you're determined enough you can jump through some hoops to make things function but at some point it's not worth the effort.

I'm guessing now with the professional market much more developed than earlier days they seem to be locking out features and limiting consumer grade hardware from what I've seen. Up until around Nvidia's 2xx series it was rather easy but with newer software revisions and newer hardware generations they seem to be forcing a separation of the consumer market and professional markets even though the hardware is fundamentally the same.

[snoid_zero]

Kepler family will be at 6:00 am Pacific Time on March 22, local time at 21:00 on the 22nd lifting of the ban

Several others also pointed to this same date. So this maybe it.

[DilTech]

That's my reasoning as well. Until we see actual GK104 benchmarks though it's hard to be certain.

The problem this presents for some people is that if GK104 is only a smidgen ahead of the 7970 that we wind up with two overpriced and unimpressive cards because AMD doesn't have anything ready to regain the top spot (dual GPU monstrosities are irrelevant in this case). So Nvidia rides out GK104 until AMD can release an 8000-series card at which point they finally dump the flagship GK110/112/100/whatever it's called.



And that's what I don't understand. The GTX580 has been out since November of 2010. Plenty of 580 owners are ready to upgrade (myself included) but I'm not going to do it for a little gnat-fart GK104 when everyone knows the big leap will be the GK100/110. GK104 is seeming more and more like a stop-gap card to keep Nvidia from hemorrhaging any more customers to AMD.

Not really a stop-gap, more like a "That's seriously all you got? Say hello to my little friend... His big brother is going to sit at home and laugh at ya".

[STEvil]

GK104 is seeming more and more like a stop-gap card to keep Nvidia from hemorrhaging any more customers to AMD.

Not really a stop-gap, more like a "That's seriously all you got? Say hello to my little friend... His big brother is going to sit at home and laugh at ya".

Either way is an inefficient way to hold onto customers.