GK110 / GTX 780 pre-launch leaks & info

[TheHunter]

Found a chinese site-pchome.net today that has posted some specs of the high-end GTX 700 series - 790, 780 & 770.. Apparently all are based on GK110..
although I am not fully buying this, I am posting the info nonetheless (as the details were updated today and might hopefully hold some truth). also the price quote says 'upcoming'.. here are the translated links anyway

gtx 790 - 6GB, 384 bit x2
gtx 780 - 3GB, 384 bit
gtx 770 - 2GB, 256 bit

this info might be half true or even a dud.. so dont forget the salt

Yeah 780 & 770 only 32 ROPS?

Seems kinda off, 48 would be more like it, AMD HD8970 apparently has 48 as well.


I hope they'll keep HyperQ and Dynamic parallelism

 

[EarthDog]

Yawn... another rumor...

 

[TheHunter]

^
Well its a fake, everything is (esp. that low 32ROP count), 3dcenter seems the most legit atm

http://www.3dcenter.org/news/ausblick-auf-nvidias-2013er-kepler-refresh-generation


780 GTX
6 Raster-Engines, 2880 (1D) Shader-Einheiten, 160 TMUs, 48 ROPs, 384 Bit Speicherinterface

 

[Frag_Maniac]

If there's gonna be a Tesla K20 by the end of the year, then it's perfectly plausible that nvidia will make a consumer graphics card at the same time.

You seem to be forgetting or glossing over the fact that Tesla product is limited production, super high end, where huge yield lots are not critical as with consumer product, so it's not in fact "perfectly plausible".

The VR Zone article has it right. Everyone, including Intel, are pushing back latest production goals because every manufacturer out there right now is struggling with these ultra small dies, which obviously affects yields, which in turn affects the quantity and consistency of each model.

Those are things that have to be dialed in BEFORE mass production begins. Nvidia is not going to want to repeat the base layer defect problem that affected the launch of the 680s, so I doubt they'll be overly anxious to beat AMD to the punch, esp considering the 700s will be a mere reboot of the existing architecture, vs Maxwell.

I'd be excited about AMD's 8000 series if it weren't for the fact that they never talk about or seem to even acknowledge the need to improve their software. Right now 7970s are going for as low as $330, but the customer feedback isn't so good.

 

[qubit]

You seem to be forgetting or glossing over the fact that Tesla product is limited production, super high end, where huge yield lots are not critical as with consumer product, so it's not in fact "perfectly plausible".

The VR Zone article has it right. Everyone, including Intel, are pushing back latest production goals because every manufacturer out there right now is struggling with these ultra small dies, which obviously affects yields, which in turn affects the quantity and consistency of each model.

I didn't gloss over anything, it's how I felt about it at the time - 2.5 months ago.

Judging from a couple of stories I've seen recently (I forget where now, sorry) NVIDIA might still release a graphics card based on this GPU. Perhaps it'll be cut down a bit? I've seen that suggested and would help with the yield problem. Of course, the softening enthusiast market really isn't helping any of this, is it? In the end, I think that there's quite a bit of uncertainty around this, but I still wouldn't rule out a launch by end of December or early January, perhaps.

 

[Frag_Maniac]

Actually I said forgetting OR glossing over, as it sounded to be one or the other. Again, I feel counting on ANY 700 being released this year or even by Jan seems overly optimistic. If even Intel says they are pushing back production 6 months, that says a whole lot regarding how difficult it is to produce these small die chips in quantities.

 

[qubit]

Again, I feel counting on ANY 700 being released this year or even by Jan seems overly optimistic. If even Intel says they are pushing back production 6 months, that says a whole lot regarding how difficult it is to produce these small die chips in quantities.

Yeah, you could be right. Too much uncertainty, can't say much more than that, really.

 

[NHKS]

just for the sake of gossip/speculation..
GeForce 700 series card - GTX780(GK110) according to source(29Nov).. not to say that i am buying this info without salt!

8+6 pin eh? and whats with that orientation?!!

 

[Slizzo]

just for the sake of gossip/speculation..
GeForce 700 series card - GTX780(GK110) according to source(29Nov).. not to say that i am buying this info without salt!
http://www.arabpcworld.com/images/2012/11/NVIDIA-Kepler-GK110-GTX-780-CARD-APCW-01.png
8+6 pin eh? and whats with that orientation?!!

That's the way they used to orient the power connections on their cards. If it's full blown GK110 this is great news.

 

[qubit]

Again, I feel counting on ANY 700 being released this year or even by Jan seems overly optimistic. If even Intel says they are pushing back production 6 months, that says a whole lot regarding how difficult it is to produce these small die chips in quantities.

Hey, December 2012 has been and gone and not a word from nvidia about a GTX 780 type card, lol. Let's hope they launch the bloody thing sometime in the not too distant future. I haven't even seen rumours about it lately, which doesn't bode well.

 

[radrok]

Yeah I do agree, I feel kinda annoyed by the lack of leaks about any of the new GPUs.

Could it be that this time the two of Nvidia and ATI will launch their new gen close to each other?

 

[LAN_deRf_HA]

If the 680 was really meant to be the 660 then the GK110 will be nvidia's most polished product ever. We've known the specs for months because it's been ready to go for months so I'm expecting a lot.

 

[qubit]

If the 680 was really meant to be the 660 then the GK110 will be nvidia's most polished product ever. We've known the specs for months because it's been ready to go for months so I'm expecting a lot.

Thng is the card has already been released as HPC accelerator a couple of months ago, so it's not that much of a stretch to expect a graphics card version of it, which is why I'm surprised at the lack of news.

And yeah, I bet it will destroy every other card out there if they ever make it. Perhaps it's the lack of games requiring that level of horsepower that's putting them off making it, as sales would likely be very low.

 

[radrok]

Thng is the card has already been released as HPC accelerator a couple of months ago, so it's not that much of a stretch to expect a graphics card version of it, which is why I'm surprised at the lack of news.

And yeah, I bet it will destroy every other card out there if they ever make it. Perhaps it's the lack of games requiring that level of horsepower that's putting them off making it, as sales would likely be very low.

I look forward to see that giant GPU overclocked, I imagine that if the HPC version is 225-250W TDP (as far as I remember) then the desktop version could very well draw 400W overclocked with a waterblock

Or maybe not cause Nvidia doesn't want us to play with voltages :shadedshu

 

[T4C Fantasy]

I look forward to see that giant GPU overclocked, I imagine that if the HPC version is 225-250W TDP (as far as I remember) then the desktop version could very well draw 400W overclocked with a waterblock

Or maybe not cause Nvidia doesn't want us to play with voltages :shadedshu

I love new gpu generations, all the benchmarks, the trolling, the architecture white papers!

 

[ViperXTR]

GPU pre release excitement , sometimes wanting is better than having

And i heard the kepler refresh would be bringing back its muscle in the GPGPU part?

 

[T4C Fantasy]

GPU pre release excitement , sometimes wanting is better than having

And i heard the kepler refresh would be bringing back its muscle in the GPGPU part?

if it uses gk110 then yes

 

[HammerON]

Would be great for a "crunching" GPU as AMD is currently the "king of the hill" (World Community Grid).

 

[the54thvoid]

Everyone needs to remember that the full force GK110 (if fully gpgpu enabled) will require a lot of power*. GK104 is so frugal because it's gpu capability is far less than it's Fermi ancestor (compare 580 compute benchmarks/scores with 680).

*I'm also sure we'll see the same people bitching about how much power the 7970 consumes suddenly say it (power consumption) doesn't matter if/when the proper daddy kepler arrives.

I do think that had Nvidia got their shit together last year and managed to manufacture a working GK100 then it would have been the same old story of recent years, faster than AMD but more power hungry. I see GK110 being that very card, but working.

 

[Kaynar]

*I'm also sure we'll see the same people bitching about how much power the 7970 consumes suddenly say it (power consumption) doesn't matter if/when the proper daddy kepler arrives.

I do think that had Nvidia got their shit together last year and managed to manufacture a working GK100 then it would have been the same old story of recent years, faster than AMD but more power hungry. I see GK110 being that very card, but working.

You know that the 7970 actually consumes about the same as the 680 at standard clocks during gaming... (source: http://www.guru3d.com/articles_pages/geforce_gtx_680_review,9.html) I saw that a few days ago while looking for some other stuff on several different websites. For 20Watts difference in practice is not much of a difference.

And i'm sure that nVidia was just waiting for ATI to release their card and see if they would have to "deploy the big guns" or just keep them for the next round, which is what they certainly did.

 

[Recus]

You know that the 7970 actually consumes about the same as the 680 at standard clocks during gaming... (source: http://www.guru3d.com/articles_pages/geforce_gtx_680_review,9.html) I saw that a few days ago while looking for some other stuff on several different websites. For 20Watts difference in practice is not much of a difference.

And i'm sure that nVidia was just waiting for ATI to release their card and see if they would have to "deploy the big guns" or just keep them for the next round, which is what they certainly did.

And GTX 680 is faster. Since most recommended is GHz edition: http://www.guru3d.com/articles_pages/radeon_hd_7970_ghz_edition_review,8.html

 

[Vlada011]

I think same it's to early for new series too.
GTX680 4GB is perfect card, special solder Classified version with backed EVboot connectors.
Cheat NVIDIA.
Intel/NVIDIA that's the way.
And this connectors on back is OK and I love cards when need much power.
With 8+8 pin and more...

 

[the54thvoid]

And GTX 680 is faster. Since most recommended is GHz edition: http://www.guru3d.com/articles_pages/radeon_hd_7970_ghz_edition_review,8.html

If you post on TPU, it's polite to use TPU's reviews.

http://www.techpowerup.com/reviews/ASUS/HD_7970_Matrix/26.html

The 7970 GHz (not the Matrix) beats the 680 on every resolution except 1280x800. And a GHz still clocks lower than your average 680. I'd also like to point out W1zz's test suite is one of the largest out there.

Point is, had the GK100 seen the light of day, IMO, it would have comfortably beaten the 7970 (10-20%?). But it would have consumed far more power and definitely cost more. 680's are still over inflated price wise.

 

[Recus]

If you post on TPU, it's polite to use TPU's reviews.

http://www.techpowerup.com/reviews/ASUS/HD_7970_Matrix/26.html

The 7970 GHz (not the Matrix) beats the 680 on every resolution except 1280x800. And a GHz still clocks lower than your average 680. I'd also like to point out W1zz's test suite is one of the largest out there.

Point is, had the GK100 seen the light of day, IMO, it would have comfortably beaten the 7970 (10-20%?). But it would have consumed far more power and definitely cost more. 680's are still over inflated price wise.

And if TPU review is wrong? Unlike you I read several different reviews of the same product.

I said GTX 680 is faster than HD 7970, not GHz.

Because of transistor count I can guess that GK100 is much faster.

 

[Crap Daddy]

Point is, had the GK100 seen the light of day, IMO, it would have comfortably beaten the 7970 (10-20%?). But it would have consumed far more power and definitely cost more. 680's are still over inflated price wise.

Well of course. Nvidia has taken AMD and us for a ride. The incompetent masterminds at the helm of AMD, the same that are driving the company to disaster, launched the 7970 clocked conservatively and at an astonishing price - remember? $550. Nvidia on the other hand wasn't ready with the big chip, you know the one that was supposed to be the flagship and compete with the 7970, the infamous GK110. In turn the very manufacturable, smaller and very powerful midrange chip was almost ready. Some tweaks, the ingenious boost and bam!
You have another flagship which at launch was better by a fair margin than the 7970 and claimed the most important title of the fastest GPU.

Now imagine what would have happened if AMD would have launched the 7970 clocked as a GHZ edition?