GTX680, images from THG review leaked

[Grooveriding]

Hopefully this clears up some confusion about Nvidia releasing a high end part later this year. Apologies if already posted.

Fail.

Wonder what nvidia will release for mid-range then. The 680 is a smaller increase over the 580 than the 7970 was over the 6970. Leaves them less wiggle room to stuff other cards in there.

Look at AMD, the 7870 nips on the heels of the 7950.

Really surprised nvidia did not deliver a flagship with the performance increase we are used to seeing from them. Maybe these guys are at the point where they can't easily crank out these big dies anymore on the new process. The troubles they had on 40nm could logically be magnified now on 28nm.

 

[imported_Smurf]

I think these new high end cards, despite being "high end" for what is available right now, need to be around $400 for me to consider purchasing. But I feel like by the time hd7970 and gtx680 get down to $400, refreshes will literally be 1-2 months away. So then, to answer your question, yes I think that is what it will take to get me to buy one of these cards. Either that or lots of overtime at my work.

If it makes you feel any better, Big Kepler will certainly be more than $400.

 

[BallaTheFeared]

GF104 is a mid ranged die with no compute power. It is the same as the GTX 460, the same as the GTX 560ti and anything else based on the GF104/GF110 design.

This card is purely for graphics it is not a work station card, it is not high end. GK100/110 are the high end cream of the crop, high performance parts which won't produce as good of perf/watt as this card has, but it will offer way more brute force. With more ROPs, bigger bus, more vRAM, bigger and more everything.


Wait, did you guys think this was Nvidia's Quadro/TESLA node process??

Ha, ha, ha, ha.

 

[Concillian]

I've heard that games currently can't use more than 4GB total RAM including VRAM. Can anyone clarify if this is true? If it's true, I can't see 3GB cards being all that useful.

32 bit games cannot address more than 4GB total RAM, yes.

It is a hard limitation of 32 bit. (2^32 = 4 GB).

However
I am not sure if this has any bearing on extra memory used at higher resolution...
- For example, 4xSSAA will render a frame with 4x the pixels and then "downsize" that to the current resolution. I am not sure if the game address requirements are the downsized image or the full 4x image. My suspicion is that these kinds of sub-pixel memory requirements are strictly handled by the card and not even seen by the game engine, and thus not a part of any addressable limit.

Typically you see addressable space issues with texture maps that the game loads into video memory, and not memory size issues with MSAA, TrAA, SSAA etc.... I suspect a video card can utilize more memory than just what a game can address.

 

[omeds]

The point is simple for those who don't think this is a mid-range GPU being sold at enthusiast prices: Enthusiast GPUs have always focused on giving the highest performance at the highest resolution and AA. 1080p is not enthusiast resolution, sorry for those who think it is.

1080p @ 120hz/fps and 1080p 3D @ 60hz/fps per eye each take more GPU power than 1600p @ 60hz/fps.

 

[Joseph F]

Hopefully this clears up some confusion about Nvidia releasing a high end part later this year. Apologies if already posted.

Is he an nV employee?

 

[bunnyfubbles]

The point is simple for those who don't think this is a mid-range GPU being sold at enthusiast prices: Enthusiast GPUs have always focused on giving the highest performance at the highest resolution and AA. 1080p is not enthusiast resolution, sorry for those who think it is.

The fact that the performance delta drop significantly at 1600p (and by extension, worse at 1080p x 3) should really tell you its a mid-range GPU, aimed at 1080p performance, castrated in dp and compute performance. It's traits are reminiscent of the gtx460, 560, 560 ti, we've seen it before, we can smell it. It smells like a mid-range product being sold at high prices. The same for the 7970, which is why I've always said its prices were ridiculous.

The architecture is great, amazing perf/w. But the entire line-up from both companies are ~$100 too high.

Really, even 7870 with its tiny GPU is being sold near $400, that's just silly season.

1080p is an enthusiast resolution, its the only one where I can get 120Hz from.

and while I otherwise agree that this really isn't shaping up to be a typical nVidia flagship part, it is hard to argue against its performance when its clearly faster at 1080p (something I actually care about very much for my 120Hz 1080p monitor) and appears to be at worst just as fast at 1600p. What we're arguing over is that nVidia's midrange part is just as good if not better than AMD's flagship and thus is well above nVidia's average for such a part.

So while fundamentally its design would be more appropriately labeled as a 660, in practice its performance really would justify a 670 naming if not better when compared to the competition for a majority of customers. Really I think the only people who would be truly disappionted would be the multi monitor users, but I'd think they'd have already upgraded their rigs with 7970s by now and thus would largely be indifferent to this release.

Is $500 too expensive? Not really, not when compared to its competition. But then we really could argue everything this generation is too expensive, and if it is the market will likely reflect that and prices will naturally come down.

 

[imported_Smurf]

NVIDIA won't release a faster single GPU this year, the next faster product from both AMD and NVIDIA shall be dual GPU based.

source

Hopefully this clears up some confusion about Nvidia releasing a high end part later this year. Apologies if already posted.

Who is Gibbo and why is he a reliable source? As far as I can tell, he's a forum moderator who runs computer store with a vested interest in both selling video cards today and keeping people off the fence. I have a hard time trusting information from a user with a troll face / anti-NVIDIA avatar.

 

[Concillian]

Hopefully this clears up some confusion about Nvidia releasing a high end part later this year. Apologies if already posted.

Haven't you learned yet that nothing ever gets cleared up in this forum?

If the data is clear, the source is not... or is biased... or is suspect... or there is a conspiracy.... or the "magic 8 ball" disagrees... or...someone's aunt's best friend has said something to the contrary... or....

 

[AnandThenMan]

Really surprised nvidia did not deliver a flagship with the performance increase we are used to seeing from them. Maybe these guys are at the point where they can't easily crank out these big dies anymore on the new process. The troubles they had on 40nm could logically be magnified now on 28nm.

I believe this is the case. Taping out a new design on a new process is getting more and more difficult and more expensive.

The fact that the performance delta drop significantly at 1600p (and by extension, worse at 1080p x 3) should really tell you its a mid-range GPU, aimed at 1080p performance, castrated in dp and compute performance. It's traits are reminiscent of the gtx460, 560, 560 ti, we've seen it before, we can smell it. It smells like a mid-range product being sold at high prices. The same for the 7970, which is why I've always said its prices were ridiculous.

Not that I blame them, but I think people are being a bit too impatient. Give it 2-3 months and I bet we see $100+ price drop on these cards.

Is he an nV employee?

My understanding is yes, or maybe I'm wrong? To be honest I don't know for sure.

 

[BallaTheFeared]

Who is Gibbo and why is he a reliable source? As far as I can tell, he's a forum moderator who runs computer store with a vested interest in both selling video cards today and keeping people off the fence. I have a hard time trusting information from a user with a troll face / anti-NVIDIA avatar.

There is no question GF100/110 will be out this year, Nvidia needs their workstation cards. You could argue it won't ever come to the desktop, but I'd find that a bit far fetched. It's not like Nvidia can just sit on their hands, AMD may have gotten smacked around in every facet this time, but they're a good company and will always be there ready to strike.

 

[AnandThenMan]

I have a hard time trusting information from a user with a troll face / anti-NVIDIA avatar.

I posted a link, that is not my information. As for my avatar, it is not anti-Nvidia or not meant to be any such thing.

There is no question GF100/110 will be out this year...

There are nothing BUT questions about when/if GF100 or whatever it is will be out this year.

 

[BallaTheFeared]

There are nothing BUT questions about when/if GF100 or whatever it is will be out this year.

So you're of the opinion that Nvidia is done, they're not going to make the GF100/110 work station card, and are going to let GCN sit on top of their ancient Fermi workstation cards?

GK104 can not replace Fermi in that area.

 

[ocre]

Did you see the benchmarks? 1600p performance is clearly inferior to the GTX 590. You don't need to take my word for it. Look at the data:

Source: http://videocardz.com/31116/nvidia-geforce-gtx-680-performance-comparison

And there is little doubt the GK110 is on its way.

At 1600p the 680 is on 10% faster on average with these results yet people keep saying the 7990 is a better card at that resolution??? WTH!!

the 3 cases it looses, its barely anything to ride home about. its 1fps or less 2 of the times. Then the big one metro,, where the 7970 wins by a mere 3fps. thats it. Out of 13 cases in 1600p the 690 wins 10 times and by a lot more than 3 fps. when it loses its by 1fps or 3. on average its 10% faster than the 7970 at 1600p, so please everyone quit saying that its only a card for 1080. It also is a better performer at 1600p.

Another thing to bring up is that the 680 just seems week in metro period. It has little to do with 1600p because in every resolution tested the 680 loses to the 7970 in metro. So its kinda ignorant to focus only on the 1600p results!!!

just saying, people are ignoring the actual facts in these numbers leaked. The numbers show clearly the 680 is a fast than the 7970 even in 2500x1600

 

[AnandThenMan]

So you're of the opinion that Nvidia is done, they're not going to make the GF100/110 work station card, and are going to let GCN sit on top of their ancient Fermi workstation cards?

GK104 can not replace Fermi in that area.

AMD is abysmally behind Nvidia in the workstation/professional market. They have no software infrastructure to speak of and are no threat to Nvidia. That may change at some point, but I would not hold my breath AMD has been historically weak on the software support side. I never said Nvidia is not going to make a compute based Kepler, I'm saying it all likely hood such a part is a far ways away, think 8 months at best. I base this on the rumored tape out dates among other things.

 

[Creig]

just saying, people are ignoring the actual facts in these numbers leaked. The numbers show clearly the 680 is a fast than the 7970 even in 2500x1600

And the other actual facts people seem to be ignoring is that these are only six game benchmarks from one website. Why don't we wait until all the results are in before we proclaim an unreleased card as the new champ, okay?

 

[ocre]

they could be separating the markets. We do know that the 28nm ramp up is so considerably slow that there just isnt enough wafers to go around. Perhaps nvidia needed to maximize the chips per wafer and this is why they 680 is so small. the graphics market makes money in bulk.

So the tesla brands make a lot more money per chip and perhaps they decided to not waste the resorces and the only way was to run two different designs. The large quantity gaming market is done in small dies to get more parts per wafer. The big die will only serve in the telsa brands where the quantities are very tiny.

It would make perfect sense.

 

[ocre]

And the other actual facts people seem to be ignoring is that these are only six game benchmarks from one website. Why don't we wait until all the results are in before we proclaim an unreleased card as the new champ, okay?

I have stated this already, in post earlier. So you are telling me nothing.

But you can clearly see that people are going on and on about the 680 being relevent only for 1080 and the 7970 is the card for 1600. But according to the data in these leaks, that is absurd!!!!

anyway, having more data........ i am all for it. i cannot wait to see it. I dont even care who is the winner.

but lets be factual.
if you read my post you will see i clearly stated this was based on the data from the leak. The same data that have some ppl claiming the 7970 is better for 1600p.

 

[BallaTheFeared]

Too much is going on, though I thought GK104 was rumored to have tapped out just 3 months before GK100?

There is a Nvidia driver roadmap floating around which shows workstation cards in Q3, I'd assume desktop would come before workstation but who knows... Either way I haven't seen anything that has made me think bigK won't be coming out later this year.

Also I don't believe AMD's lack of presence is going to affect Nvidia wanting to get a new revenue stream out there, you don't make money as a business sitting on your old products. Intel is an example of a company that keeps pushing the limits year after year even though there is no competition and that market is grossly swayed towards Intel with something like 90+% market share.

But at this point it doesn't much matter, it's a phantom at this point and the only way we'll know for sure is when it actually launches.

 

[96Firebird]

GF100 and GF110 have been out for a while now...

 

[Joseph F]

GF100 and GF110 have been out for a while now...

Lame.

 

[BallaTheFeared]

GF100 and GF110 have been out for a while now...

GF GK GR GD is what I got.

Ha

 

[Grooveriding]

Intel is an example of a company that keeps pushing the limits year after year even though there is no competition and that market is grossly swayed towards Intel with something like 90+% market share.

Intel, unlike nvidia, doesn't just design their chips, they also own and design their fabs as well, as you know. Not to mention they are a monster compared to nvidia in resources and footprint.

Nvidia and AMD are at the mercy of TSMC and what TSMC can and cannot do for them, all while attending to more customers than just them. AMD/Nvidia can't do much if TSMC says 'no, can't accommodate you right now'

I think we could safely assume nvidia plans to at least pull 6 months at a minimum out of the 680 as a flagship $500+ part. That takes us through to September, I won't be surprised if we don't get a GK110 until 2013 at this rate. They're not exactly under the gun to get something out like they were with GF100 and all its issues.

Unless they're swayed by customers raising a stink about nvidia not being consistent and delivering in the 680 what is a mediocre flagship perf. increase compared to prior ones. Yeah, not going to happen. I have no idea what sort of expectations their professional customers have and/or how important it is to nv to give them new and faster products.

 

[-Slacker-]

To BallaTheFeared

Is there a reason why nvidia can't make a GK104-based Quadro line?

 

[BallaTheFeared]

Nvidia has a big market in GPGPU, letting that sit out of their quarterly statement makes no sense to me, but perhaps my sense making isn't what it used to be. Intel is after their market there, Knights Corner is an example of how Intel is trying to cut into Nvidia HPC shares... AMD isn't a threat atm, but Intel as you mentioned is a monster and Nvidia wouldn't be very wise to sit on their 2 year old 40nm power hungry HPC cards.

To BallaTheFeared

Is there a reason why nvidia can't make a GK104-based Quadro line?

Because GK104/GF104 lack compute power, that's why they have much better perf/watt than their big brother counter-parts. They're basically graphics cards with CUDA cores. They can take advantage of it in some things, but at their core they're not workstation cards. GF100/110 have lower perf/watt than GF104/114, but they offer considerably more GPGPU performance. Nvidia actually cuts their GPGPU performance so they can sell the same cards for $4000 in the HPC market.


Here is a GTX 470, which sold for $350 retail and I bought at the end of 2010 for $150 each.

http://www.newegg.com/Product/Produc...82E16814133347

That's a $4000 GTX 470 with 6GB of ram. Nvidia wants to replace that card, GK104 will not replace it.