Topic: 7990 delayed (Read 7972 times)

Latest reports has it debuting in early June at the Computex show.  I'm going to hang onto my 5970s til the end of May I guess.

1250 MHz core that is under a full 7970 waterblock I meant to say; at stock voltage for 24/7 between 38 to 45 C (above ~1260 MHz, and you need more volts). Under a waterblock, 1300 to 1375 MHz is probably the upper limit (brief test: ~880 MH/s), sounds like a terrible idea to run at such settings for 24/7.

Electromigration would probably be more of a concern on an air cooled 7970.

Firstly, I don't think 1250 MHz for a 7970 is safe or maintainable for long periods without electromigration taking a toll on the card.

Secondly, I think they will launch 7990 in May.

Its actually quite interesting to get all this insight for such a delay. Since the 680s are already out, it is possible to predict how a "690" would perform ideally, which is why I raise the question on AMD holding out longer with out a dual gpu product. If they cannot make a product that can compete with 690, then what you said would make sense for a delay: the possibility that AMD is looking for higher TDP envelope (and cooling problem & leakage @+80C) or more aggressive binning. Oh well, was looking forward in having a 7990 with a longer career in my rig... , but dual kepler and big kepler Tesla are coming up soon.

I have seen many tech and youtube reviews describing the 680 as slightly better than the 7970, at times the 7970 surpasses in certain applications under certain conditions. Still an odd delay in my opinion (both can be described as equal in a sense). Tech review sites will be on top of the real issue behind the delay in a little while.

AMD could do what NVIDIA did with GTX590, ala bin the really good GPUs and overclock the heck out of them for 7990. The 7970 is only 5% slower clock for clock in gaming compared to GTX680. A good binning process could easily make up that difference.

The only "fishy" thing that is going on is that they might have problems releasing a 7990 that can compete with gtx 690 performance, slapping two 7970 cores on the same PCB is no more challenging than doing it with 58xx and 69xx as long as they used the same approach with lower clocks/vcore. The problem is that it most likely wont be enough to gain the performance crown this time around over the GTX 690, the situation was the other way around with gtx 590 vs 6990 where the 590 was held back by thermals.

My point is had it not been for GTX 680 "unexpectedly" beating 7970 we would have seen the 7990 on schedule with them following the same approach as they have done before with earlier dual cards to keep the TDP in check. At this point they are most likely either looking at a higher TDP envelope or simply doing more aggressive binning to get chips that can handle higher speeds with lower power usage, first alternative takes redesign and second one just takes times since you are either waiting for the chips to pile up or the yields to improve.

The dual cards have never been a high volume product and as long as they weren't making more cards than they could sell (GTX 680 launch) there was no incentive to do volume binning for 7990 (If you can sell one card for a premium today why save it and hope that you can still charge a premium tomorrow).

I might as well be wrong tho since i'm just speculating, if the 7990 is released with <900 core you are allowed to slap me :p

There is one thing that does give credit to the whole TDP/Cooling problem however and it's the leakage that has been reported at 80C+, power usage can skyrocket 30-40W per 7970 when they start approaching 90 degrees and this might potentially be a huge cooling problem for a dual card. So if there is a problem this is most likely the one they are trying to solve.  

Edit: Nvidia isn't focusing on the "high end" in a traditional sense, like I said gtx 680 is not a high end GPU looking at the chip size. Rather their whole mid segment lineup is going to be largely consistent of harvested GK104s that didn't make the cut, iirc the gtx 660/670 are both gk104, might even be TI variants of these. They will be released when GTX 580/570/560 stocks reaches a level where nvidia wont be competing with themselves. AMD was in a whole other situation since the 6950/6970 stocks have been dwindling for some time and at least here in Europe the prices has gone up for these cards since end of last year, while on the other hand the GTX5xx prices has plummeted hinting at a "spring cleaning" before they get the lower kepler SKUs out the door. :p  

Yes. The 7970 launched in early January, the only 28nm product on the market, new technology, AMD took advantage of this position, and set a desired price. nVidia countered with their 680 in March, in order to attract sales; there should be reason why a customer would pick nVidia's product over AMD's (market has brand fans and non-brand fans): releasing a product superior to a 7970 is no doubt the focus, despite being a little over two months late after the 7970's launch. nVidia launches their 680, sales decline for AMD's 7970 (though the entire single gpu lineup is already being sold in the market from AMD, nVidia only has its most expensive single gpu card out), so AMD re-prices its products to be more competitive against a now present competitor in the 28nm market.

Graphics card do improve in performance from one generation to another, I would be tempted to say at least by 30% (not set in stone), just to throw a number out there for reference. Early launches do pose a risk, the possibility that the competition will tweak/reconsider product specs and later down the road release a superior product (better than 30% improvement, into a competitor saturated market). I am guessing this what you are referring to? I am not an electrical engineer (though I do know a few), but I know clock settings on a chip are specified by a number of factors in its design, my initial reaction in the 7970 being able to operate from a stock 925 to a 1250 MHz range told me they wanted to have headroom that can compete performance-wise with a later released nVidia product.  

nVidia intends to release their 690 sooner than anticipated according to their road map (originally Q3?), an interesting marketing strategy.  Indicating they want to release all high-end/enthusiast cards first (then mid/lower end later), in order to capture back more of the high end-market (expen$ive) that they been shut out from since January (7970). Had the 7990 been released in March (just before or after 680 release), then nVidia would most likely be under more pressure in attempting to establish itself in sales/market share of this generation of GPUs (again brand fans are the loyal buyers, expanding by luring non-brand fans is important in the market). Long story short, we have a delay for the 7990, something fishy is happening... (it is difficult to slap two Tahiti cores onto a pcb)


A new Tesla Kepler is being released on May 14, double the transistor count than gtx 680, nVidia's first move for this generation of high performance computing?

In the high end market holding the performance crown is everything and means you can charge a premium. Just look at what happened with the 7970 when GTX 680 launched. Nvidia pretty much pushed the 680 as far as they safely could to make sure it would be faster than then 7970 by a decent margin and can now charge more for it than they should by logic. GTX 680 is essentially a mid range card looking at die size and the card itself and should sit somewhere around/just below the 7950 in terms of pricing (6950 vs 560 TI is a similar situation looking just at the physical cards)

Had AMD known about the performance level of 680 they could possibly have been able to launch the 7970 at 1000-1100 core since the headroom is huge atm with the stock at 925. Had they done this they could have stuck to their price premium for a much longer period since 680 vs 7970 performance hadn't been as clear as it is now, and there wont anything else coming from Nvidia except GTX 690 in the near future afaik.

That's why it would make sense to wait with launching the card, another thing is that the higher they plan to push the card to beat the "expected" 690 performance, the more aggressive they have to be with binning the 79xx cores and it takes longer to build up enough supply for a hard launch.

In this market you either have the performance crown and you get paid for having the "best" card or you sell the second best card in the world cause it has good performance/price, guess where they would rather sell their cards.

Well, trying to beat a gtx 690 is a rather bad reason to justify the delay, business is the priority, selling the product brings in revenue; holding out for later is counterproductive (lose potential customers in market): AMD wants to be first in a new generation of VGA cards for a reason, no? Remember Fermi from nVidia? It was delayed, sales were not the greatest, so nVidia later re-launched Fermi a second time; obviously a reason exists for this occurrence. 7990 launching Q3 2012, and not in Q1: AMD has come across some obstacle on this high end product to keep having it delayed. (Why was 7990 put as Q1, then the more recent MSI slide say "7990" TBD?)

Trying to beat gtx 690 most likely, else they would just have made a "5970" with 2x7970 at 7950 speeds and voltage if heat was the only thing holding them back. They have the "Fastest graphics card in the world" throne still technically with the 6990 (when crossfire works and it's not a game that favors Nvidia, at least I guess that's how AMD looks at it) and I guess they want to keep it.

I figured it would be yet another delay. A release at practically Q3, there is a reason for this, surely they are having problems with something; faulty yields?

I almost spit out my coffee when I read that one!

http://www.techpowerup.com/164746/AMD-Chooses-Computex-2012-as-Radeon-HD-7990-Launchpad.html

I prefer ssateneth's. Time to update urban dictionary.

Quoted for comedy gold. It means Fixed Your Post For You.

No idea what FYPFY means. "Fuck you pussy, fuck you"?

7970 has a TDP of 210W according to AMD. I'm not sure where that 660W number came from.

FYPFY.

People who use TDP to measure power draw are just naive.  TDP is a SAFETY rating.  It indicates the maximum amount of power (and thus heat) the device can safely dissipate without damaging itself, nearby components, or causing melting, scorching, or fires.

I'm pretty sure they're talking about max TDP, which is usually a higher number than what you'd typically see while mining.

That's BS. I'm mining on a 3930K with a pair of stock 7970s and my TOTAL system consumption doesnt break 600W