Nvidia Kepler - Geforce GTX680 Thread - Now with reviews

Oh, and this...


G7QLn.png


HiGJC.png
 
Nvidia's Kepler to get compute cousin, says analyst

MOUNTAIN VIEW, Calif.--Nvidia Corp. will likely announce a follow-up to its recently launched Kepler platform to better address upcoming compute products and the high performance computing (HPC) space later this year, according to an analyst.

David Kanter of Real World Technologies said the newly launched Kepler platform was specialized for high-end graphics, whereas more general purpose workloads would necessitate a differentiated version of the chip in order for Nvidia to remain competitive in the HPC space.

“It will be a derivative of Kepler, to re-use as much of the engineering effort as possible, but with several significant changes,” he said, hinting that a Kepler cousin could be announced as soon as May.

When it comes to different workload requirements, Kanter said it was clear that a graphics centric chip would not require much in the way of cross-system communication, whereas scientific computing did need it for algorithms to be efficient.

“For purely graphical use, the pixel that’s on the bottom left corner of your screen doesn’t care what the pixel in the middle of the screen is doing at all,” he explained, noting this wasn’t the case when trying to accelerate flow calculations or other more complex HPC data.

Thus, said Kanter, the hardware for each specific purpose -- gaming or scientific-- would have to be slightly different.

“Fermi had great resources for communicating between different parts of the application, but Kepler doesn’t have nearly as much capability to communicate between various levels of the system,” said Kanter, explaining that this would require a two-fold approach from Nvidia.

“I don’t believe Nvidia is abandoning HPC at all, it’s clear to me that the firm will continue on that path, because the area is really heating up, especially with AMD’s new offerings in the space and Intel on track to release Knight’s Corner at the end of this year. Nvidia can’t be trying to enter this market with one hand tied behind its back,” he said.

On the other hand, said Kanter, Nvidia is also acutely aware that it had not done quite as well as its rival AMD in the last round of the graphics battle, namely because it had spent a lot of resources in terms of power and area on things that were more useful for compute but not useful for graphics.

Hence, Fermi was much more attractive in the HPC space than in the niche high-end gaming segment.

“What Nvidia decided to do with Kepler was to scale it back, to really focus on graphics,” said Kanter, noting that the Kepler follow-up would likely be a bit more separate and distinct to fine tune it for general purpose.

Compute specific GPU requirements
Compute products, said Kanter, require error-correcting code ECC memory to protect data, as well as double precision for more floating point accuracy, important for things that aren't graphics.

“AMD had a much more graphics optimized chip in the last generation and so, rather than have one design to rule them all, Nvidia will likely split Kepler into two families and reuse a lot of the same elements but have slightly different features for the separate workloads,” he said.

The approach is somewhat similar to what Intel did with its server and client version of SandyBridge, though the firm effectively used the same core for both but simply added a lot more cache and memory bandwidth to the server version, which also had twice as many cores, more PCI express and QPI. “That’s the ideal thing to do,” said Kanter, though he said Nvidia’s plan would be to build similar cores but not quite the same.

“Nvidia has to scale it up to do the compute side, so it will probably be a much bigger chip,” he said.
This variant on Kepler should make its official launch at Nvidia’s graphics technology conference (GTC) in May, with products likely to appear around the Q4 timeframe.

Kanter said the real issue was not necessarily whether Nvidia could design such a product, but whether its manufacturing partner TSMC could actually produce it with sufficient yields, especially as the 28-nm process was still so new.

Yields, said Kanter, typically only improved with time, and this, he said, is also the reason that Intel never started out on a new process technology with a large die.

By coming out first with the consumer graphics part, which is smaller at around 30-mm squared, Nvidia can wait for TSMC’s yields to mature before coming out with Kepler’s larger compute cousin for which it will likely need a larger die size of around 500-mm squared range.

When asked, Nvidia said it had not spoken to Kanter about its compute solutions for Kepler but wouldn’t comment further.
Click to expand...
But as we all know GK110 is already in production according to dr_rus ;)
 
Hooray! I grabbed one about 30 minutes ago, I was about to post it here before I was pulled away from my desk.

I have until April 3rd to return my 3GB 580 to Amazon so hopefully it'll arrive by the end of this week and I won't have to go without a video card for a few days.

And as I hit submit it looks like it's out of stock again on Newegg :O
 
I'm so glad it was Newegg and not Amazon that got them in stock. If it was Amazon I would've ordered without hesitation - free Prime shipping! - ignoring the fact that I as a Washington resident would have to pay nearly $50 in sales tax!

Newegg - no sales tax. Woo!
 
My 2 680's left eVGA a few hours ago and are on their way home. Since UPS delivers late, I will be getting them around 5pm Wednesday night. Already planning to have my rig opened and waiting for the new cards. Just not sure if I will go straight into SLI or test each card on it's own first.
 
Dammit I hate being impatient.

Local site has plenty in stock but they also jacked up the price a bit. However, buying a card with a stock cooler? I'd rather be whipped to death. Then you could usually rely on Arctic for a third party cooler but oh-no, they didn't bother making it compatible with the stacked power connectors. Then we have word of upcoming 4GB cards, yay! but since I'm living in Denmark it will either be a) very limited supply or no supply b) damn expensive, which is unfortunate when you want to do SLI and need an AX1200 as well. Don't even want 2GB models.

Saved up a heap of moniez for this momentous occasion :(
 
Alright guys, my Asus P6T mobo/i7-920(stock)/ 460 sli'd seems to be getting a bit long in the tooth. Would grabbing an 860 make a substantial difference or would I need to upgrade my mobo/CPU as well? I'd love to get a 680 and learn to over clock my CPU if I could avoid having to upgrade my mobo as well. Any thoughts/ideas?
 
Sethos said:
Dammit I hate being impatient.

Local site has plenty in stock but they also jacked up the price a bit. However, buying a card with a stock cooler? I'd rather be whipped to death. Then you could usually rely on Arctic for a third party cooler but oh-no, they didn't bother making it compatible with the stacked power connectors. Then we have word of upcoming 4GB cards, yay! but since I'm living in Denmark it will either be a) very limited supply or no supply b) damn expensive, which is unfortunate when you want to do SLI and need an AX1200 as well. Don't even want 2GB models.

Saved up a heap of moniez for this momentous occasion :(
Click to expand...

You absolutely don't need a 1200W PSU. These cards pull less then 200W each so unless you are running 4 680s with a SBe processor overclocked with a ton of fans and lighting and a foot heater you don't need 1200W
 
Now this is a slight bit OT, but related, ish.

Thanks in part to this thread keeping me informed about the release of the Kepler/Geforce GTX680 won me...well a non Nvidia card, but hey.

http://www.hardwareheaven.com/news.php?newsid=3901

Of course, my bet didn't get me a GTX680 as XFX no longer make Nvidia cards, but hey, not bad for a result of a little debate.
 
scogoth said:
You absolutely don't need a 1200W PSU. These cards pull less then 200W each so unless you are running 4 680s with a SBe processor overclocked with a ton of fans and lighting and a foot heater you don't need 1200W
Click to expand...

I know but I only have a 750 now, which is enough but will be running at the edge of efficiency, and no way in hell will I be spending over an hour redoing cables just for a pitiful 850-1000W upgrade. Might as well go big, get the AX 1200 and then I have room for another 680 down the line or it can carry on over to my future build.
 
Sethos said:
I know but I only have a 750 now, which is enough but will be running at the edge of efficiency, and no way in hell will I be spending over an hour redoing cables just for a pitiful 850-1000W upgrade. Might as well go big, get the AX 1200 and then I have room for another 680 down the line or it can carry on over to my future build.
Click to expand...

FYI, I don't think the AX1200 is even the best PSU, anymore. There have been lots of contenders since. I recommend checking http://www.jonnyguru.com/ for reviews prior to taking the plunge.
 
sk3tch said:
FYI, I don't think the AX1200 is even the best PSU, anymore. There have been lots of contenders since. I recommend checking http://www.jonnyguru.com/ for reviews prior to taking the plunge.
Click to expand...

From what I've read, I'd disagree. The AX1200 is still one of the top dogs. Plenty of PSUs out there in the same bracket that can match it but there's so many that fall below it. Almost went for a good deal on the Cooler Master Silent Pro Gold 1200W but after reading a few reviews that was a big no-no.
 
I have to say, I'm happy I waited. The card is chewing up, pretty much everything I've thrown at it, on maxed out settings @ 2560x1440. It's quiet and it runs cool. I'm impressed. Wish the 4gig cards were available at launch, but so far, 2 gig seems adequate.
 
Fixed1979 said:
Meh, if it's anywhere close it's worth it.
Click to expand...

It's almost $50 more. Not worth it.

EDIT: Looking at reviews, like that Jonnyguru site, the AX1200 scores higher than the Seasonic. Sounds like a fantastic PSU for sure but let's not pretend the AX1200 is a slouch by any measure. Had it been cheaper than the AX, I might have jumped.

EDIT2: Found a retailer that had it a bit cheaper than the AX1200 but out of stock with no delivery date on new stock. Sent them an email.
 
scogoth said:
I'd go seasonic 1000W platinum. More then enough for anything you want to do, better efficiency and amazing PSU.
Click to expand...
This is the PSU that I have and it is indeed amazing. I mean the AX1200 is a really good PSU and I also considered upgrading to that, but eventually went with the Seasonic due to its Platinum rating and the reviews I read on it. Zero coil whine, a hybrid mode that knows when to kick in the PSU fan and when not to (or you could just leave it on using the normal mode). Read about some that were DOA under newegg user reviews but for me it's been a flawless experience *knock on wood

How much more is the Seasonic vs AX1200?

Edit: guess its more expensive then the AX1200? I'd probably go with the corsair in that case.
 
You must log in or register to reply here.

Similar threads

Nvidia Releases GeForce Now App for the Steam Deck
44
4K
jshackles replied
Nvidia Community Stream - 18th August (12pm PT // 3pm ET // 8pm UK) 'Major Geforce NOW news'
1
170
tronied replied
Anyone here use Geforce Now (Ultimate?)
48
429
Three replied
Geforce Now getting 5080-class GPUs
11
372
peish replied
GeForce RTX 5060/5060 Ti review thread
2 3
145
12K
ThisIsMyDog replied
Top Bottom