GTX 680, Turbo Cores, and Cuda Cores!

Subject: Graphics Cards | March 8, 2012 - 03:59 PM |
Tagged: nvidia, kepler, gtx 680, GDC

It seems that there have been a few leaks on NVIDIA's first Kepler based product.  Techpowerup and Extreme Tech are both reporting on leaks that apparently came from Cebit and some of NVIDIA's partners.  We now have a much better idea what the GTX 680 is all about.

View Full Size

Epic's Mark Rein is showing off his own GTX 680 which successfully ran their Samaritan Demo.  It is wrapped for his protection.  (Image courtesy of Extreme Tech)

The chip that powers the GTX 680 is the GK104, and it is oddly enough the more "midrange/enthusiast" offering.  It has a total of 1536 CUDA cores, runs at 703 MHz core and 1406 MHz hot clock, has a 256 bit memory bus pumping out 196 GB/sec, and has a new and interesting feature that is quite a bit like the Turbo core functionality we see from both AMD and Intel in their CPUs.  Apparently when a scene gets very complex, the chip is able to overclock itself up to 900 MHz core/1800 MHz hot clock.  It will stay there for either as long as the scene needs it, or the chip approaches its upper TDP limit.

These reports paint the GTX 680 as being about 10% faster than the HD 7970 in certain applications, but in others it is slower.  I figure that when reviews are finally released the two cards will have traded blows with each other over who has the fastest graphics card.  Let's call it a draw.

The GTX 680 should be unveiled in the next week or so, but initial reviews will not surface until later in the month.  Retail availability will be relegated until then, but with the issues that TSMC has had with their 28 nm process (it has been stopped since the middle of February) we have no idea how much product NVIDIA and its partners has.  Things could be scarce after the introduction for some time.

 

 

Source: NVIDIA
March 8, 2012 | 07:42 PM - Posted by nabokovfan87

Well good, Kepler is nothing to worry about.... not until the GK110 stuff.

The OC boost of a GPU worries me, that is a LOT more heat then a CPU puts out, and especially with 23xx stream processors, somewhere, someones system will get put down by it.

March 8, 2012 | 08:04 PM - Posted by pdjblum

Josh,

Do I understand it correctly that this is 10% faster than AMD's flagship, but is only Nvidia's midrange enthusiast card as the 7870 is AMD's midrange enthusiast card?

March 8, 2012 | 08:11 PM - Posted by Sonic4Spuds

That is what the article implys.

March 8, 2012 | 11:09 PM - Posted by Buyers

I'm also curious about the naming convention here. The GTX580 was the top of the line single gpu card in last/current gen, as was GTX480 in the generation before that. That leads me to believe that the GTX680 would also be the top end for this generation as well, not "midrange/enthusiast". How then are they going to be naming the high end and potential dual-gpu solution?

March 9, 2012 | 10:35 AM - Posted by Anonymous (not verified)

It just an excuse for Nv to price their GK104 at 500 USD or more... Naming it as a mid-range GTX 660 Ti would make alot of ppl complain with the release price of 500.

March 9, 2012 | 07:15 AM - Posted by Josh Walrath

NVIDIA looks to be taking a new route with this part. My best guess is that it is possibly a little smaller in die size than the HD 7970, but that does not really make it a mid-range card like the HD 7870. It is a much larger chip than that one!

Previously NVIDIA seemed content to make a really large GPU for its next generation of product. It seems in this gen they are going with a smaller one, and will work out all the design kinks and process issues that they feel that they were running into in the past. So with this they put out a still competitive part, get better yields, and are able to sell it for $500+ rather than put out a much larger chip with bad yields for the same price range.

Thankfully this "smaller" part is still very fast, and outpaces the previous generation quite nicely.

March 9, 2012 | 11:53 AM - Posted by AParsh335i (not verified)

"and it is oddly enough the more "midrange/enthusiast" offering."

I"m still confused what you are saying Josh. Like the other comment said, most people are reading this as the GTX 680 is not their flagship card for the Keplar release. By the way you wrote that sentence i quoted above i would assume that they are going to have a "GTX 690" or even a "GTX 780" to be their flagship. Based on the information provided so far, if it is true that they do have a flagship card above the GTX 680, and if the 680 is on par with the 7970, we are going to see something pretty nuts for the guys that have the $700ish to drop.

March 9, 2012 | 12:56 PM - Posted by Josh Walrath

Well, it isn't so much confusing, as it is different from how they have done it before. Basically they are first releasing a less complex/smaller chip... that is still faster than the previous generation of products. They will still charge $500+ for it, and as time goes by and prices start to go down due to better yields on the 28 nm process... they will likely release their larger GK110, and that will become the new high end of the Kepler generation. Or so that is what we assume they are doing.

For right now, this is their top end card. It will not always remain their top end card. Until that point comes though, this will be their $500+ offering.

March 13, 2012 | 01:06 PM - Posted by Anonymous (not verified)

"or even a "GTX 780""

This is exactly it I think and I'm surprised I haven't seen this guessed at more often.

We have to recall that the Fermi designs powered both the 400 series AND the 500 series. If the GK104 is enough to let nVidia claim "fastest" against the 7970 then my guess it we'll see that GPU released as the 680 now and the GK100 released as the GTX 780 when the time is right (read: when AMD counters with a newer card of their own).

They'll then cycle all of the GPU's upwards into the 700 series. The GK104 will magically become the 760 TI (or 760) and a downclocked, CUDA core disabled GK100 variant will become the GTX 770.

At least, this is my guess.

March 13, 2012 | 05:07 PM - Posted by Natai (not verified)

I'm thinking the same thing. Not sure why everyone seems to forget that both 400s and 500s were Fermi. I'm definitely expecting the 680 to fill the role the 480 did with Fermi. Once they get the kinks worked out, they'll release the GK100(110?) as the 780 and then start pushing more midrange 700 series cards. It's really not much different than how they launched Fermi, though everyone seems to think it is.

March 8, 2012 | 08:11 PM - Posted by Sonic4Spuds

This looks really nice. I don't think it will have any big heat issues, as NVIDIA seems to do well in making sure things do not kill your system.

-Sonic

March 9, 2012 | 12:14 AM - Posted by Irishgamer01

Its all about the price. Bang for buck. Or EURO.
So far AMD's are priced very high, so at the right price, nvida could have a winner. No matter what the slight speed advantage is.

If supply is low on the other hand, then we are are going to be in a duopoly scenario, where both sides decide to keep their higher prices and decide to milk their respective fan base for as much as they can.

Alas I think this is what is going to happen. So I will stick with my 2 580's for a while longer, and keep looking for a used third card.

March 9, 2012 | 01:09 AM - Posted by Tim (not verified)

X80 was a flagship number, so flagship will be 685/690? What will they call their dual GPU then? 695/699?

I was hoping for the 200% faster rumor to be true... guess i'll stick with my 7970

March 9, 2012 | 10:28 AM - Posted by Anonymous (not verified)

I'm confused. It is stated here that the card will be 10% faster than the 7970 in certain applications but slower in others. It's also stated that the card ran the samaritan demo successfully which previously required 3 x GTX580's. This seems contradictory to me as the 7970 is, lets say for arguments sake between 1.2 & 1.5 x the power of a GTX580. Wouldn't this give the GTX680 around twice the performance of the 7970?

March 9, 2012 | 10:42 AM - Posted by Anonymous (not verified)

There was no real indication which setup was faster...

The Tri-SLI 580 GTX could have generate way higher FPS than one GTX680... it's just that GTX 680 alone is sufficient to run it at a comfortable frame rate

March 9, 2012 | 10:54 AM - Posted by Josh Walrath

It may be that the GTX 680 is just more efficient at that particular workload. NVIDIA looks to have done a lot of work on geometry and lighting with the new chip, but in terms of pure pixel pushing power... it is not that massive step up (eg. 3x the perf of GTX 580s).

March 9, 2012 | 03:38 PM - Posted by staryoshi (not verified)

The GTX680 is the midrange/enthusiast Kepler part. It was intended to replace the GTX560 Ti. It will compete with the HD7970. nVidia needs to get moving on their high-end parts. It would have been an amazing win for nVidia if their lineup had been complete at this time, but it looks like we'll be waiting for their flagship card for quite some time.

March 10, 2012 | 04:54 PM - Posted by AndrewST (not verified)

Ick, this is slightly depressing news. I was really hoping for something to justify the expected price tag, but that doesn't seem to be the case. Hopefully it will outperform AMD enough for AMD to drop their price, then maybe I will ditch Nvidia and go AMD for the first time in years.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.