I'm a little puzzled by this bit -Kepler taped out in September
Seems strange to me that Nvidia would do a die shrink of Fermi, and if they do, what is the time frame for this? It needs to be fairly soon, otherwise it is going to run into Kepler and what would be the point? Unless Nvidia plans to shrink some of the lower end stuff and rebrand it.Before Kepler arrives, the GPU maker is expected to release a series of die shrinks of Fermi built using TSMC's 28nm fabrication process.
Could be 2 months, could be 6 months. No one really knows right now. There might be some more concrete dates coming out just before the 7970 is available next year...
And judging by your most recent post, closer to 3 months...
Sorry, I thought you were talking about another AMD card.
I'm not stating worse case scenario, its just logical given how much problems TSMC has been having with their 28nm node. The worse case scenario for NV is gk100 is DOA and just cannot be made by TSMC. Require a total redesign and wouldn't see the light of day until 2013.
Making a 520mm2 GPU is hard enough on a mature 40nm process (remember the gtx560ti-448, selling low yield chips for $$), making an even bigger GPU on a new process?
Your quote, the last sentence. Read it. Q2 2012. It means they need a respin for gk104. The smaller easier to make GPU.
So if everyone is talking about another card coming out in a couple of months, why would you get the 7970 now?
Wasn't this the thinking when the 5870 launched? And then early adopters not only got in at lower prices but they got to enjoy their DX11 cards for some 4-5 months (or was it longer?) before competition showed up?
A bird in the hand is worth two in the bush. History has taught us to not count our Nvidia eggs until they hatch. It'll get here when it gets here, until then we've got 7970 coming on Jan 9th and it looks mighty impressive if you are still on an original 5870 or GTX460 or older.
There's no merit to any of your dire projections for Kepler, i'm not saying you might not be right(time will tell) but i have not seen anything of the sort that your talking about.
Latest rumors have GK104 and GK100 released in Q1 and Q2.
Edit: thread is about the 7970 so ill move on.
There was a similar sentiment going around with Fermi, how would Nvidia possibly compete price wise with AMD due to the gargantuan die size. Turns out, Nvidia did just fine and in fact has shows good margins. AMD on the other hand has barely made money with their GPU division in total. It is not really apples to apples though because AMD has their CPU division in the mix and are saddled by debt, but the point is, Nvidia has shown they have no trouble eeking out profits despite their philosophy of "big bad huge monstrous" GPUs.What everyone also needs to factor in is the size of the die. At 365mm^2, AMD can probably sell those at the 300-400$ price range no problem. Something that Kepler will probably not be able to do. So when everyone says Kepler can easily beat 7970, you might also want to think about: Hey! What if AMD dropped the prices? Will Kepler still be competitive on a perf/$ ratio? AMD has huge maneuverability margin here.
There was a similar sentiment going around with Fermi, how would Nvidia possibly compete price wise with AMD due to the gargantuan die size. Turns out, Nvidia did just fine and in fact has shows good margins. AMD on the other hand has barely made money with their GPU division in total. It is not really apples to apples though because AMD has their CPU division in the mix and are saddled by debt, but the point is, Nvidia has shown they have no trouble eeking out profits despite their philosophy of "big bad huge monstrous" GPUs.
Whereas there's definitely solid indication of a very late mid-range kepler and even later high-end.
Agreed. It was not that long ago that the Radeon brand was on life support, so it's easy to understand why AMD pushed out high value parts to gain back marketshare. But that is not a sustainable model, unless your majority profit generator is from other products.Which is all a very good reason why AMD has gone to nVidia's pricing structure.
I am huge disappoint. Obviously I would love to have a 7970, but HOLY crap at the cost. It's over 2x more expensive than a 6950 2GB (which I already have). It's hardly 2x faster, particularly if you look at CF 6950 2GB, which is STILL cheaper.
The 6xxx series didn't make the 5xxx obsolete at all, looks like the 7xxx series leaves the 6xxx and 5xxx series still in place as well.
From my understanding of reviews, the biggest thing to take out from this gen of cards is the fact that tessellation is no longer an nvidia dominated zone.
The other thing is that this card is running on what I would consider to be an alpha driver. the performance will definitely improve over six months with new catalyst releases.
Third, these cards respond much better to ocing than the 6xxx series. I have an unlocked 6950 that I run on 6970 clocks and most of the time, I can hardly feel the difference in performance.
These used to be amd's weak areas and looks like they have worked on all of them. That leaves nvidia fans grumbling because they have to wait for the green team to show up to the party.
Pricing is a bit of a sore spot but amd needs to make money so they are justified in charging what they want in a scenario where they can stay ahead of an oced 580 comfortably.
My expectations from die shrinks is more realistic. I don't expect new gen tech to trounce previous gen dual card tech those days are long gone.
As a consumer I can't wait for kepler to launch as this will mean amd will reduce the prices which would be a win win scenario.
Some of the reviewers are not able to hide their bias. Techradar seems to be the worst. They find it difficult to pat amd for doing something right for once lol.
Solid - coming from someone or something other than Nvidia. Like Charlie. Solid.
Coming from NV themselves. I already posted info on that, look it up. NV admited no kepler until late Q2.
People seem to have forgotten the HD 5870 release. It wasn't all that impressive if you compare it to the GTX 285.....
I don't think people have forgotten the HD5870's original launch. A lot of people just don't consider TPU's HD5870 launch benchmarks relevant at all since they included a lot of ancient games.
How about this?
vs.
Thankfully, HD7970 overclocks like hell! :biggrin:
Seems strange to me that Nvidia would do a die shrink of Fermi, and if they do, what is the time frame for this? It needs to be fairly soon, otherwise it is going to run into Kepler and what would be the point? Unless Nvidia plans to shrink some of the lower end stuff and rebrand it.
I was a little disappointed as well since I just sold my 6950's but after reading a few reviews I think the performance difference will be worth it. According to ComputerBase, the 7970 is 61% faster than the 6950 at 1600p. From early indications the 7970 overclocks like stink and performance scales well. The 6950 generally gets a 8-10% overclock on stock volts and performance doesn't scale as linearly. I know my 6950's would hit 900MHz on the core with 1.175V. On stock volts they were closer to 860-870MHz.
With a healthy overclock the 7970 should be just as fast as Xfired 6950's which is perfect IMO. You get the performance without the power draw and multi-card compromises. Bang for your buck still isn't as good as the 6950's since they can be had for ~$200 on the used market or $230 new, but that's always been the case with single vs multi-gpus.
The only reason I would be disappointed buying a 7970 is if AMD decides to release Tahiti 1.1 in a month or two like they did with the X1900XT.