Nvidia will have a dual GK104 card, too. This time it's likely going to win due to lower TDP.
It's an interesting situation. Who will do the next step? And if you know that the other one is doing it, will it be too late for you?
Printable View
Nvidia will have a dual GK104 card, too. This time it's likely going to win due to lower TDP.
It's an interesting situation. Who will do the next step? And if you know that the other one is doing it, will it be too late for you?
I would like to see dual GPU on 7870 base, which has extremely impressive perf per watt ratio and its way smaller gpu than GTX680. Seing that its 256bit mem controller is really enough to keep up, i dont see any reason why not go for it.
Maybe they can even push clocks up more than Ghz per core.
I find the notion of holding back cards or AMD having the ability to somehow react to a considerably faster card with anything but price adjustments silly. These cards take close to a year to get from tape out to being available in stores. Adjustments you can make after tape out are fairly small unless you're willing to spend a quarter or two and take it for a new spin. If we entertain the thought that GK100 was to be at some point and was cancelled afterwards, and a replacement GK110 was taped out in January like Charlie claimed, we'll be lucky if it launches this year. Nvidia named GK104 GTX 680 which should be indication enough that no faster single GPU is coming any time soon.
You're assuming chips weren't being prepped along side one another... No one ever said the design was cancelled. It just vanished from talks after AMD showed off their 7970.
Also, they don't take a year from tape out unless the design ends up needing a lot of work. A solid example is that the NVidia G80 taped out in July of 2006 and launched in November of the same year. Taking a full year means a LOT of mistakes were made, requiring a few respins to fix the problems.
Finally, realistically AMD's answer will be the 7990 (likely a 7970x2). This won't require a new chip.
Either way, this wouldn't be the first time either company has released a card and trumped it shortly after... The x1800xt had what, 2 months to shine before AMD dropped the x1900?
Personally, I hope they get on with the show sooner than later. Last thing I want to see is the obvious price gouging we have presently going on right now. Highest price single gpu we've seen out of AMD since...well...the x1900xt.
I wouldn't call any other rumors about GK110 any more believable.
It's not like releasing a gk110 wouldn't be nice for Nvidia. It just that considering the circumstances, releasing gk110 later doesn't hurt them that much and releasing it later can be good for them or perhaps both companies even for the long term. With AMD raising prices of graphics cards across the board, and Nvidia can do one of two things, price gk104 aggressively and call it gtx 660 and price it accordingly. This in turn would set off a price war which would be less profitable for both companies. Or do what is rumored and follow AMD pricing/performance, call it the gtx 680 and price it as such. This would also allows the company to raise the price of the rest of their lineup and ease people into paying 100 dollars more for gk110 when it does come out in the future.
Releasing a gk110 right now, if it was available, would make a $550 gk104 not make any sense. Considering gk110 is mass produced and this is the consumer market, i think the max price of such a high end is 650 to 700 dollars. If the die of gk110 is 60% bigger, i would expect gk 110 to be 30-50% faster. With such a performance delta, gk104 would have a hard time justifying its 550 dollar price. Also with GK110 taking the gtx 680 name, gk104 would be stuck with the gtx 660 name and there is no way this card could be sold at $550.
If Gk104 was dramatically slower than the 7970, they would need gk110 to come out much sooner, similar to how AMD needed a successor to the 2900 xt as soon as possible. The high price of the 7xxx and/or relatively small performance jump over last generations high end has giving Nvidia so much flexibility this generation on their release schedule and pricing. Look at the price/performance of their old last generation cards compared to AMD's new cards and you will see this. Beside the gtx 580, both companies last generation have hardly needed much for price cuts. Old stock doesn't need dramatic price drops to sell and Nvidia can thank AMD for this. A GK110 doesn't need to come out right now because AMD has shown Nvidia, people will pay 550 for the performance level of the 7970.
Actually I said there's always other parts in the works, and don't doubt there are more powerful Kepler chips in process.
What I said is:
A. Given the state of the market, it could make good sense to launch what might have been the 2nd or 3rd tier chip first to take advantage of better margins as it would certainly be cheaper to produce and still reportedly beats the competition.
B. Could be the bigger chip isn't able to be brought to market profitably now due to yield issues.
C. It doesn't make much sense to endlessly debate unreleased rumored products, because we can't buy them.
I think that is pretty "good analysis".
If there is a GK110 then in the unlikely event it's even ready now yields would be far below GK104. It would be rather stupid to sell a large low yielding chip into the consumer segment when you can get the same price for a much smaller higher yielding die.
AMD did nVidia a huge favor with Southern Islands pricing. Both companies win.
What's interesting though is how AMD will react to GTX 680. Let's say it's 10-20% faster than 7970 and costs 549$ like the leaks indicate. AMD has the bigger chip and more vRAM on its card, so competing with price will be more expensive for them. Yet if Nvidia can duplicate the success of GK104 in the lower end too, AMD is in a lot of trouble. The only hope of AMD is that Nvidia won't respond to price cuts and will simply milk GK104 for all its worth.
My estimation was for 3DMark and 1920x1080 4xAA performance when there's no need for the extra bandwidth grunt 7970 has. You can already see which cases those would be based on 7870 performance. No doubt 7970 willl be faster in those games and 30" and eyefinity resolutions, unless Nvidia really has some sort of trick up their sleeves for Surround. Compute goes for AMD too, but does a gamer really care?
If AMD drops prices, Nvidia can always do the same, and it will most likely be much easier for them, at least after they get production quantities in order. My guess is AMD will have to drop their prices 50$ or so and Nvidia simply won't have to since demand will exceed any supply they might have, given their warnings of low 28 nm production quantities. AMD can probably match the GTX 680 with OC 7970 cards like the MSI Lightning, but we'll just have to see how Nvidia cards will OC in turn. :up:
looks like nVidia's mid-range line-up will be occupied by GTX 570 & 580
Amazon now repriced it's ASUS GTX570 DirectCU II to $299.99
http://www.amazon.com/gp/product/B00...00_i00_details
I find it hilarious that people claim AMD can't compete with Nvidia because now, for the first time since RV770/GT200 Nvidia has the better perf/mm˛ and perf/W if leaks are believable. Nvidia managed to compete with AMD "only" by throwing a whole lot of die area and transistors on the chips. AMD has been doing this with CPUs for a long time, they will probably be able to do it on the graphics market as well.
Here's something from OCUK
Unforturnately the cpu and res is a disappointed pairing.
EDIT:
Here's what a user at overclock.net posted using his 7970
http://i60.photobucket.com/albums/h2...3/b2d283f1.jpg
He's running 32bit, not sure if that matters.
I just ran heaven at his settings with 2500k @ 4.4
FPS 59.8
Score 1507
MIN 12.7
MAX 157.1
7970 @ 1000/1425
GTX 680 looks to be at least as fast as 7970
what resolution you was running ?
I ran same as him 1600x900 all same settings in heaven
@Rattle, thanks .
Anyway, the guy on OCUK report a weird thing, 706mhz as default clock ( i let you read their threads lol, too much pages ). nvidia control panel, GPU-z say the same. Dont know if theres different default clock or not. (outside turbo mode ) ...
I hope it's an error on reading the default clock.
http://img13.imageshack.us/img13/4623/sanstitre1oxa.png
GPU-Z 0.60
Quote:
Added optional GPU-Z installer which launches when GPU-Z is started for the first time, or can be started manually via the GPU-Z system menu
Added support for AMD Radeon HD 7870, HD 7850, HD 7670M, FireStream 9370, FirePro V3900
Added support for NVIDIA GeForce GTX 680, GT 640M, GTX 560 SE, GT 550M, GeForce 510, NVS 2100M
Fixed clocks not showing correctly on some systems with Catalyst 12.2
Fixed board id reading on pre-Fermi cards
Fixed OpenCL detection for ATI Fusion (Sumo & Ontario)
Fixed GPU-Z window getting cut off at certain DPI settings
yep good find, and it have fix the report of dual gpu ,..
Now if someone can send it to this guy on OCUK lol ..
min fps mean nothing in Unigine Heaven if the tester doesn't cycle through all the 26 scenes at least once by pressing enter. Only then are the results realisting - streaming problems otherwise.
Yup it was recently brought up, he's running 300.65 not 300.99 as seen in another screenshot, he also need the newer GPUZ I think :)
Just apparead on Tweekdown, apparently the GTX 680 is listed at $566.75
http://www.tweaktown.com/news/23058/..._75/index.html
Looking forward to some noise results.. If it idles quieter than my 480's I'm sold!!
Title of this thread should be 680?
Yeah I posted what the original link was days ago in this thread. Likely a jacked preorder price but it's listed.
http://www.xtremesystems.org/forums/...=1#post5070420
From overclock.net again.
For reason the Heaven is showing the card as an intel IGP, although he says it might be because he ran it in Windowed mode.
http://i60.photobucket.com/albums/h2...3/7ef2e0b4.jpg
http://i60.photobucket.com/albums/h2...3/4e932bd0.jpg
^ link please? So we get the larger original picture.
Copy/paste the url's and it gives the original size:
http://i60.photobucket.com/albums/h2...3/7ef2e0b4.jpg
http://i60.photobucket.com/albums/h2...3/4e932bd0.jpg
:)
EDIT: Notice it says release date March 22nd, which is Thursday... :D!
Thanks. 78 fps...that is 35% faster than a 7970 (57 fps):
http://www.forum-3dcenter.org/vbulle...postcount=6815
Impressive ^^
Someone with a 7970 care to bench? :)Quote:
Batman AC: 4x MSAA, 16xAF, Global Settings on very high, PhysX off (to compare to AMD)
112 FPS at 1300/6500 on 1920x1080.
(I assume he used DX11).
Someone asked earlier in that thread for a DX11 batman, so it probably is... :eek:.
Beginning to question if I really even *need* 2x GTX 680 instead of just 1x, for my 2560x1600 gaming... for now... certainly *want* two but... haha.Quote:
Thanks, lolitsallwrong, it's way better than original OP numbers! You can run free bench like Alien vs Predator etc.. i wonder how it runs with Batman Arham City + DX11+PHyX HIGH + 1080p smile.gif
A GTX570 gets 40fps there (after the DX11 patch), with 8xMSAA though. Still most impressive.
http://www.legitreviews.com/article/1798/3/
Batman AC bench was on very high not extreme? That would make it slower than 7970. So I assume he made a mistake
and it was actually on extreme.
As for the 1600x900, he wanted to compare it to the bogus benchmark from the OP. Heaven with extreme tess and AA is not CPU bound.
Yes, very high apparently. Maybe he repeats it later. Everyone wants him to bench something different, lol.
Anyone who uses the AC benchmark for a review will be getting very inaccurate results as both NVIDIA and AMD have built optimizations into their driver stacks JUST for benchmark. In-game results in exterior scenes (most of the game) differ wildly from the benchmark.
How did you get such high score on a stock 7970?
EDIT: nm, you only ran 1440x900
Probably cheated with the tessellation slider in the CCC. 92fps is BS. 55-57 is more realistic, maybe 62 at 1440x900 as opposed to 1600x900.
The same way as the guys with 680 passed from 54fps to ~78.
Ok believe in what you want. I can get wtv Fps that I wat from the ~55 with default tess to ~93 with costum tess, and the same can be done to nvidia cards, just takes a bit more work then slide a bar. Just my 2 cents. Btw as far as it showed the guy run a 7970 too and got 54fps so they were "on par".
So 55fps then, not 92. As I said - you cheated.
He ran the 3.0 Benchmark again:
http://www.overclock.net/t/1231113/g...#post_16752543
If I cheated all the top results in Heaven are as you say... Cheats. As for this matter I was only trying to prove a point. It's easy to mess with tess with both brands drivers, if I really wanted to "cheat" I would have made it more realistic ;)
Then tell me, how can you influence the tessellation level with Nvidia drivers?
Why would I ? But believe me it can be done ;)
I would just say that is in the same way you can use bitmap cheat/tweak in AMD drivers ( it's stright forward in nvidia and dificult in AMD ).
So you basically accuse the poster over at OCN of cheating but don't say how that is actually possible. Nice, very believable.
He has posted a new screenshot of fullscreen mode to show perf wasn't particularly affected:
http://img805.imageshack.us/img805/8...ullscreenf.jpg
I don't really care about what you think. I'm just watching two different results, and if you don't find the diferences suspectfull there's nothing I can do.
That would just mean that these 680 would be equivalent to a 7970 1250//1650.
http://dl.dropbox.com/u/20116032/Heaven2.png
Yes, in this particular synthetic benchmark. And we all know how strong Nvidia is with tessellation. But we'll see in 2-3 days.
Wah wah, wait for the real review/ers
Uhm true, I could swear they were run with 1600x900. I'll re do them for your mental sake.
What's the point of running Unigine Heaven if you don't atleast run it at 1920x1080? :P
Not really, if Nvidia could release a stronger GPU now they would, but they can't.
The only thing that could have happened is, that NV went with more conservative clock speeds to lower the TDP, when they saw that HD 7970 was "only" GTX 580 +30%.
there is no reason for them to back off if the competition is weak. if they spent a hundred million making kepler, they loose millions with every day they dont launch it. right now a card twice as fast as a 580 would be worth 800$+, in a few months it might be worth only 500$
if they had both cards ready, they would have released both cards
my take it that they learned their lesson with woodscrews and large chips thanks to the fermi situation and decided to launch their 460/560 type part first, since thats where most of the market share can come from. however due to their timelines, or thanks to competition, they have the ability to call it an upper level part. the fact its called a 680 means their large gpu is still a long ways out and/or that it competes with the 79xxs. it dosnt even have to be faster than a 7970 either, since they might get a whole bunch of sales thanks to people who have believed thanks to their track record that its faster than the competitions high level because they usually are.
i think nvidia has played it very smart this time around. the only thing we wish they did was show up, or atleast give us real press, when amd launched their cards. too much silence when your running late usually means you have not much good to say. imagine if they gave us one teaser add a few months ago saying, "hold on, we have something guaranteed faster and/or cheaper coming" and by that late in the game they probably could have said it too since they would know where they stand, and they set their prices. but instead they let competition soak up some market share, which i dont think was very smart.
Wishful thinking.
Nvidia certainly had no qualms releasing the 8800 GTX and then the 8800 Ultra at $800+ prices when AMD had nothing, they certainly had no problem releasing the GTX 280 when no one knew what the 4800's had in store, etc.
They'll release what they can, when they can, if money and market share make sense (and market share is more important than ever, now that the PC GPU industry is shrinking)
Bingo. If millions are spent on R&D, you would try to recoup the costs, rather than let a product sit around doing nothing.
Playing fair has never been part of the lexicon of GPU makers, so why play 'fair'?
^^ that's exactly why I keep harping on about "business decisions" when people (specially fans) talk about lack of need to release products.
It seems my comments go unread since people keep posting how "110" was not needed to be released... lol I have never seen this kind of behaviour ever...
Inter saying to shareholders: "Let's not release any more CPU's for the next 4 years, AMD needs to catch up and our CPUs are fast enough!"
If anyone find the above comment extremely stupid that's exactly what's being said right now on this thread about Nv's bigger chip.
does the mighty 104 seem like a good product? all indications say yes, do people want to go on having a mental diarrhoea and say the first thing that pops in their head? most likely.
Compare
http://img853.imageshack.us/img853/4920/680m.jpg
with my 7970 @ 1005/1500
http://img20.imageshack.us/img20/295...2008491694.jpg
with my 7970 @ 1005/1500 and AMD OPTIMIZED
http://img706.imageshack.us/img706/1...2009023656.jpg
A 800 dollar gtx 680 would not sell very well at all. That's limited edition card pricing and the ultra fell from 800($829 to $699) really quickly and that was in a bit better economy. There's a point where a high enough price will curb off demand and I think at that point, it is reached and passed. Dual cards can sell for that much because ultra enthusiast will pay more for dual cards. And even then, 800 is alot of money to spend and we haven't reached that point yet unless we count the ultra limited editions.
If GK110 was 800 dollars not only would the card not sell well, there would be no way even with that high a price that the Gk104 could be priced at $550. The only way GK104 can be priced so high is because it is the fastest thing from Nvidia. No way a midrange gtx 660(naming means alot) could be priced liked that.
Volume is the key. And releasing a half assed gk110(not fully mature) with terrible yields are hardly going to do anything with their profits. And would hurt the price the gk104 could be sold at
Intel does exactly that to some extent. Not for AMD to catch up but to get maximum return on research and development. In addition a weak AMD that is not a threat gets the trade commission off their back for monopoly practises. The less often Intel introduces new products the less they have to spend on R and D. Look how early we get leaks for future Intel products and we can see Intel could be way more aggressive with their release schedule.
Money is not infinite for R and D and is one of the biggest costs for both companies. I remember NV spends somewhere between 300-500 million a quarter on R and D, if they continue to release products as soon as they are finished, they would have a crazy hard time recouping costs. Unless the product is like bulldozer which was a total bomb, holding off to recoup money on R and D does have its pro's.
Releasing something half baked like the original fermi could do more harm then good and it might not be bad to hold off.
I partly agree with your statement, but this was aimed at the comments saying GK110 was not released because they didn't need to, not because it wasn't ready.
edit:**
they could be more aggressive?.... mate I do not think they could be any more aggressive than what they currently are. and yes we got info about Haswell already (because they spent a lot of money on R & D,and look, we are just getting SE-e and IB) these guys live of the verge of technology and they're very aggressive spending their resources on research...
Who said Fermi was half baked? Fermi aside from heat was and still is a fantastic chip. Frankly most people forget that the CPU designs they use are modular and therefor they could and do scale the chip up or down based on the needs of the day. The real RND is not in the actual die itself but the interconnects and the modules themselves. Thats part of what bit fermi in the ass; they couldnt scale it down because each module was configured too large and they couldn't effectively harvest half-dead chips like AMD could thus now we see a super-scalar design from nVidia with much smaller stream processors.
I am quite sure that is without a doubt a GK110 exists; most likely what is going to happen is they are probly going to do another respin and add directX 12 much like what we saw with Cayman. Then they'll put it up against Hectoncheries
Yields alone made it a bad chip. It was very fortunate for Nvidia, that they had a contract in place where they only had to pay for working chips which didn't carry over to 28nm. With the increase in wafers costs and the limited allocations they are getting, the better the yields of gk110, the more profitable it will be.
The gtx 480 original launch was hardly well recieved and was made a joke of from the AMD camp(i.e the commercial of gamers getting busted by the cops for too much power usage).
If Nvidia can release the fully realized gk110(ala gtx 480 vs gtx 580), it will probably be a better seller and allow them to make a better professional products that are not severely underclocked with a bunch of cores not activated like the fermi professional products.
Asus has their subforum up for GTX 680:
http://vip.asus.com/forum/topic.aspx...Language=en-us
...and another vendor lists a (probably high for preorder) card... http://www.pcnation.com/web/details.asp?item=NG8027 $558 shipped for an Asus.
EDIT: $585... lol... http://www.costcentral.com/proddetai...2GD5/11583991/
So... you have nothing to refute?
What are you even trying to say? History alone would show that you're very wrongQuote:
"Playing fair", or milking the masses? You decide.
That's what I was trying to say. All these stupid conspiracies about GK110 not being released because AMD isn't competing is stupid. If GK110 isn't ready, it isn't ready - hence why we have a GTX 680 first at $550. And for all we know, GK110 isn't close to being ready and may not be til the end of the year.
All the talk about Nvidia makes it sound like Nvidia can't adopt a similar strategy to AMD and go with getting a mid-range out first, especially on a new process, especially after learning from Fermi and past history
GK110 is probably not ready yet, give it 3 months. Early speculation was we were gonna see the X60 card early anyways. So saying that is a more powerful card coming soon might not be off at all. Just 3 months or 6 months is the question.