NVIDIA GK104 Coming Soon?

Status
Not open for further replies.
22a.jpg


PCB prob engineering sample

1 thing I would like to say about this. I think Charlie was right about the card back in January.

"For the doubters, both of the current cards we saw have two DVI plugs on the bottom, one HDMI, and one DP with a half slot cooling grate on one of the two slot end plates. The chip is quite small, and has 8 GDDR5 chips meaning a 256-bit bus/2GB standard, and several other features we can’t talk about yet due to differences between the cards seen. These items won’t change the outcome though, Nvidia wins, handily.S|A"

http://semiaccurate.com/2012/01/19/nvidia-kepler-vs-amd-gcn-has-a-clear-winner/
 
So you honestly think, if this card is lets say 20% faster then a 7970, Nvidia would charge less then AMD?

Since when did Nvidia charge less then AMD on any video card?

P.S. Also remember with 28nm issues. Not making money on a video card isnt smart business sense.

Put the performance piece aside for a second and look at the card itself (assuming the leak is real). It has a GTX 560-class die size, it has a 256-bit memory bus, and it has 2GB or VRAM. Nothing in those specs says ~$500 card - they all say $350 card. So even if Nvidia decides to up-sell it, and it ends up at $450 or so, and they'd still make a killing because the manufacturing cost is going to be low (compared to the GTX 480/580).

I agree with SonDa5 that this is Nvidia's chance to hammer AMD, and I think they'll take it.

In any case, the leaks should start flowing soon if the March date is real.

I have not tried to justify the price of the 7970, I have tried to show that ALL topend cards tend to be very overpriced. My apologies if my post did not convey this intention properly.

I think we can all agree with that. Didn't mean to jump on you.
 
This is bullshit with the extra physx chip. I'm pretty sure Charlie has no idea what he is writing there. But maybe physx will be so fast because of the raw computing power of GK104. Should be in the range of 3 TF compared to only 1.6 for the GTX580.

See my post below about Charlie. He was right about what the card looked like back in January.

So it is possible he is right on the Physx Story....only time will tell (3 weeks)
 
Let's just say I have a source that tells me otherwise. Charlie being right on one thing doesn't make his ramblings on other things true. His reports on Kepler have been too inconsistent for my taste.
 
Let's just say I have a source that tells me otherwise. Charlie being right on one thing doesn't make his ramblings on other things true. His reports on Kepler have been too inconsistent for my taste.

Well considering the only thing we know about kepler is pics from what we seen, and rumors of specs.

And a month and a half ago Charlie posted that he saw the card and what it looked like.....and now the pics prove he was right?

What has charlie been inconsistent about with kepler? IF anything he has been right so far...

P.S. and I hope Charlie is right. I hope Nvidia wins this round easily...
 
Last edited:
First he says it wins in almost every metric and then he backpedals, saying performance varies between sub-Pitcairn level and Tahiti (XT?) territory. After being questioned repeatedly about the PhysX story, there was no answer.

It just doesn't make any sense. Seems like he is trying to find explanations why Kepler is so good and making this stuff up in the process. Oh btw:

GK107 (the real one this time):


http://we.pcinlife.com/forum.php?mod=redirect&goto=findpost&ptid=1844549&pid=33452412

For reference (GT550M):
http://www.cooltoyzph.com/viewitem.php?iid=632

GT650M vs. GT550M (same low bandwidth due to GDDR3) in 3DMark06:

SM2.0: 5279 vs. 3991
SM3.0: 5466 vs. 3687

The GT550M ran at 1280x768. Unfortunately there is no info about the resolution the GT650 machine ran at. But it looks like Kepler is very efficient with bandwidth. Up to 50% more performance at the same bandwidth.
 
First he says it wins in almost every metric and then he backpedals, saying performance varies between sub-Pitcairn level and Tahiti (XT?) territory. After being questioned repeatedly about the PhysX story, there was no answer.

It just doesn't make any sense. Seems like he is trying to find explanations why Kepler is so good and making this stuff up in the process. Oh btw:

GK107 (the real one this time):


http://we.pcinlife.com/forum.php?mod=redirect&goto=findpost&ptid=1844549&pid=33452412

For reference (GT550M):
http://www.cooltoyzph.com/viewitem.php?iid=632

GT650M vs. GT550M (same low bandwidth due to GDDR3) in 3DMark06:

SM2.0: 5279 vs. 3991
SM3.0: 5466 vs. 3687

The GT550M ran at 1280x768. Unfortunately there is no info about the resolution the GT650 machine ran at. But it looks like Kepler is very efficient with bandwidth. Up to 50% more performance at the same bandwidth.

Ok let me rephrase myself. I ment 104GK. not mobile Kepler. I do not care about mobile kepler.

Also. It is possible the reason he back tracked on 104GK performance, could be because of Canned benchmarks (which Kyle himself mentioned that he is seeing 40-50% increase in canned benchmarks compared to 580 GTX). Canned Benchmarks could be 3dmark 2011 that uses Nvidia Physx, which could make Nvidia win hands down since AMD does not have physx.

Of course this is all speculation, but that doesnt take away from the fact that Charlie was right about the Ports and the memory configuration of 104GK.
 
GK 107 is shown to have 384 unified shaders. So they are still counting the shaders the same way as Fermi?
 
3DMark11 doesn't use physx. Also he said "680 vs 580" and the 580 obviously has physx ;)

@Hornet:
Yes, but hotclock is gone, so you need twice as many.
 
3DMark11 doesn't use physx. Also he said "680 vs 580" and the 580 obviously has physx ;)

@Hornet:
Yes, but hotclock is gone, so you need twice as many.

104GK is rumored to have some super efficient Physx built into the card itself. This is just rumors, but that could explain the performance increase.

And I thought 3dmark used physx?....of course I havent used 3dmark is years so im probably wrong..

P.S. Ahh in 3dmark11 they decided against using Nvidia Physx. Interesting.
 
There are other things that make GK104 fast vs GF110 (GTX580):
+100% texture fillrate
+80% compute power
+33% VRAM
more efficient use of bandwidth
 
There are other things that make GK104 fast vs GF110 (GTX580):
+100% texture fillrate
+80% compute power
+33% VRAM
more efficient use of bandwidth

Vram wont do a damn thing other then help in the higher resolutions.

Otherwise....its all speculation, We do not have any form of details on the specs other then rumors (although the 104gk pics show 2gb of memory).
 
Yes, so? Performance at higher resolutions matters, too.

Considering Nvidia (almost) doubled the shadercount the past few generations, the step from 384 (hotclock) to 1536 (non-hotclock) is believable. A great increase of texture fillrate is also a given. Remember how many TMUs the 8800GTX had? 64! That number has not increased for 6 years now. Even GF114 had 64. It is very likely that they will finally put more TMUs in it, especially with that much shaderpower.
 
Yes, so? Performance at higher resolutions matters, too.

Considering Nvidia (almost) doubled the shadercount the past few generations, the step from 384 (hotclock) to 1536 (non-hotclock) is believable. A great increase of texture fillrate is also a given. Remember how many TMUs the 8800GTX had? 64! That number has not increased for 6 years now. Even GF114 had 64. It is very likely that they will finally put more TMUs in it, especially with that much shaderpower.

I totally agree it does help in the high-end. Just like the 580 GTX 3gb helped in triple monitor gaming compared to the 580 GTX 1.5gb.

But what im saying is....specs of the 104GK are all rumors right now...We can guess all we want.

but this comes back to Charlie being right about the ports on the 104GK, and the the 2gb of memory and the Die being small.

everything he mentioned in the middle of January was correct after looking at the pic today.

Performance and specs.....no one (except people in China) know what they are.
 
Last edited:
Well, I know my source, so I know most of the specs. If you don't believe it, it's your call ;)
 
Well, I know my source, so I know most of the specs. If you don't believe it, it's your call ;)

You got proof?....then ill believe ya!

My Source says your source is wrong.

My Source also told me that Aliens are real too!
 
You got proof?....then ill believe ya!

My Source says your source is wrong.

My Source also told me that Aliens are real too!

Errrr.... no. I'll take boxleitnerb's info with a grain of salt, but posts like this I would discard outright for credibility ;). You've been around awhile... you know people can't provide "proof" in these cases.
 
Charlie has no proof either, so he is a troll, too? No one would reveal his/her sources that's just common sense. You are of course allowed to not believe me. However, troll-calling is not okay and you know it.
 
Get this thread back on topic and knock off the name calling, Its not acceptable and there will be consequences
 
True, but this was just to show that there is more potential in GK104 than some mythical physx leprechaun that is running around in the GPU ;)
 
The GT550M ran at 1280x768. Unfortunately there is no info about the resolution the GT650 machine ran at. But it looks like Kepler is very efficient with bandwidth. Up to 50% more performance at the same bandwidth.

Some people are trying to dismiss and/or are worried GK104 will be extremely bandwidth constrained, but they're failing to look at the past to see how Nvidia's newest mid-range offers competed against their previous flagships in regards to bandwidth.

gtx285 - 159 gb/s mem bandwidth
gtx460 - 115.2 gb/s mem bandwidth
gtx560ti - 128 gb/s mem bandwidth

The gtx460 arguable outperformed the gtx285 (if we all agree that it was underclocked and use an average of factory overclocks then it definitely outperformed the gtx285) despite it's noticeably lower bandwidth. And the gtx560ti, with still noticeably lower bandwidth, absolutely smokes gtx285, running 40-50% faster across the board.

AMD has, without a doubt, a better understanding of memory controllers and have been able to maximize vram speed as a result. But, to me, it seems that Nvidia has been able to optimize their architectures and improve cache efficiency in such a way that they are able to squeeze a little more performance at similar or lower bandwidths than AMD.

EDIT: Throw the 8800GT, and 8800GTX into the mix and it's a similar story. Sure, the 8800GT ended up being about 10% slower, but it had 40% less memory bandwidth and it wasn't a next generation product replacement of the 8800GTX.

8800GTX - 86.4 gb/s memory bandwidth
8800GT - 56 gb/s memory bandwidth
 
So if the GK104 part is 45% faster than a GTX 580, you are going to be fine with Nvidia charging $700 for it? Because that's the same logic that says the 7970 is fairly priced. You're cool with the 7990 being $900?

The GTX 580 replaced the GTX 480 with 10% higher performance at the same price. AMD replaced the HD 6970 with the HD 7950, with ~15% higher performance, for $100 more (~30%).

I don't understand why anyone is defending AMD's pricing. Do you like paying more?
most people here do not give a damn whether it is AMD or Nvidia. its people like you that act all psycho and try to make this an AMD vs Nvidia issue when all most people want out of the next gen are much faster cards at the same price points.

7970 costs 50% more than the 6970 while only being about 40% faster. so much for getting a faster card at the same price point. and the gtx580 was considered a poor value by just about everybody. funny how AMD's next gen card comes out and only beats it by 20-25% and while also costing more and all of sudden that is supposed to be a great deal. these prices are going to trickle down and we end up paying high end prices for what is really just next gen upper mid range performance.

sorry but if it was not for the great overclocking, the 7970/7950 would be a joke for a next gen card IMO. and if the opinions of these "nvidiots" bother you so much then stay out of the Nvidia sub forum.
Because I understand simple economics, which a lot of people here seem not to. The supply of 28nm wafers is incredibly low and AMD still brought a competitive part to market that outclasses anything in the high end. Period. There is nothing worth getting in the high end at the moment, and for the last two months, unless it's a 7950 or 7970. If you want lower prices, tell both TSMC and NVIDIA to get their acts together and produce some competition.

A few people consistently pan AMD for releasing a 28nm-based enthusiast GPU that delivers, first, but "oh, it wasn't at a price I like, I'm going to go cry about it." I would think on a hardware forum people would be excited for new cards, but instead there's this whiny narcissistic drivel in every thread. If GK104 is released and lowers the prices of the 7950 and 7970 in the process, great, everybody wins. But that's a long shot with the way 28 nm production currently is.
 
You realize that
1. Tahiti is a small die for a 550$ card
2. the GPU itself is only about 1/4 or 1/5 of the complete BOM (bills of material) - and then there is still room for profit.

28nm does account for the high prices a little, but +50% MSRP over the 6970 is just insane. No way their chip costs 250-300$. The GF110 cost 120$ and it is 50% larger than Tahiti.
 
You realize that
1. Tahiti is a small die for a 550$ card
2. the GPU itself is only about 1/4 or 1/5 of the complete BOM (bills of material) - and then there is still room for profit.

28nm does account for the high prices a little, but +50% MSRP over the 6970 is just insane. No way their chip costs 250-300$. The GF110 cost 120$ and it is 50% larger than Tahiti.
This is a simple argument of supply and demand. AMD is probably making a healthy profit on each chip, but by market forces there's no reason they shouldn't. Because of the stunted production of 28nm chips, there's a very limited supply. AMD could price Tahiti-based cards lower and still make a profit, but then they'd sell out even more quickly than they are now. That's just throwing away profits, and no competently run business does that.
 
I do think the pricing is incorrect at the moment. Prices have changed very little in the last two years, this is the reason I haven't upgraded since a 460, everything seems overpriced (oh and the gtx460 will run almost everything).

I'm hoping Nvidia's next gen will spur some competition which has been absent recently.
 
This is a simple argument of supply and demand. AMD is probably making a healthy profit on each chip, but by market forces there's no reason they shouldn't. Because of the stunted production of 28nm chips, there's a very limited supply. AMD could price Tahiti-based cards lower and still make a profit, but then they'd sell out even more quickly than they are now. That's just throwing away profits, and no competently run business does that.

I understand that, of course. It is still bad value for a next-gen card, that is my opinion as a consumer. In fact, every 7000 card is too expensive right now. I would not buy GK104 for 400-450$ either. I'd rather wait until this silliness is over.
 
I understand that, of course. It is still bad value for a next-gen card, that is my opinion as a consumer. In fact, every 7000 card is too expensive right now. I would not buy GK104 for 400-450$ either. I'd rather wait until this silliness is over.

I agree but not by much. IMHO the HD 7970 and all other top end single GPU cards should be around $50 (£30) cheaper on release. In the UK the new gen top end single GPU usually retails for £400-£430 inc VAT.

It might sound ironic agreeing with this considering I purchased a HD 7970 at MSRP (I avoided the price gouging retailers). AMD are obviously following the usual trend as far release as pricing goes. Thankfully by the end of this month Nvidia should have GK104 released. We need competition to force prices lower IMHO.

HD 7970 is an excellent card and a considerable improvement over my 6970, but price/perf is poor going by release prices. Eg. On release my HD 6970 cost £290, the HD 7970 is ~40% faster so logically it should be £406. So price gouging retailers asside, the HD 7970 is roughly similar in prece/perf at its MSRP of £420. This sounds bad but it needs to be remembered that HD 6970 was not the top end card when it was released. Lets consider the following points.

1) Performance is meant to increase with price staying static for each new gen released, because if it doesn't then prices just keep going up each new gen. Imagine if the HD 8790 is released in a year and it is 50% faster than HD 7970 and priced at £630. Should we defend that price based on the fact that price/perf is identical? Of course we shouldn't, but this will not happen.

2) The GTX 580 was priced around £400 on release here in the UK, before that the GTX 480 was released at £430. AMD released the HD 7970 at a similar price of £420 (price gouging retailers asside). It was a faster card than the previous gen at a similar price, so as is the usual trend, price/perf had improved a the high end.

So IMHO all AMD did with the HD 7970 was take the absolute high end performance crown and it was priced at the usual high end price. AMD could not do the same with HD 6970 because that crown belonged to GTX 580. When Nvidia or AMD release a high end card that beats HD 7970 in the future, it will NOT be priced higher, it will be priced similarly. This is the normal way of things.

We need to remember that when we are comparing HD 7970 release pricing we shouldn't be comparing price/perf to HD 6970, but to GTX 580. GTX 580 was overpriced compared to GTX 570 and HD 6970. GTX 480 was overpriced compared to HD 5870. HD 5870 IMHO was the first time an absolute top end GPU was released at a realisitic price. Unfortunately it was because AMD wanted to gain marketshare, not because they were starting a new trend.
 
Last edited:
Because I understand simple economics, which a lot of people here seem not to. The supply of 28nm wafers is incredibly low and AMD still brought a competitive part to market that outclasses anything in the high end. Period. There is nothing worth getting in the high end at the moment, and for the last two months, unless it's a 7950 or 7970. If you want lower prices, tell both TSMC and NVIDIA to get their acts together and produce some competition.

A few people consistently pan AMD for releasing a 28nm-based enthusiast GPU that delivers, first, but "oh, it wasn't at a price I like, I'm going to go cry about it." I would think on a hardware forum people would be excited for new cards, but instead there's this whiny narcissistic drivel in every thread. If GK104 is released and lowers the prices of the 7950 and 7970 in the process, great, everybody wins. But that's a long shot with the way 28 nm production currently is.

Great post Mr k6. Sums it up really well.
 
Great post Mr k6. Sums it up really well.

Nah. New generations released well over a year later are supposed to offer a lower price for the same performance, not a large hike (25% price) for a bit more (20%) performance. That's why everyone keeps bringing it up as a poor value: it's 14 months later, on a new node, and yet it isn't even a good price/perf for what it does deliver, which in and of itself is underwhelming. It isn't compelling at all for the high-end buyer.

The alternative is for them to be the same price for a lot more performance, which also happens often, but the Tahiti cards failed completely on that too as I mentioned above. Hopefully Kepler gets the market back in line to normal... AMD is feasting on the large amount they probably are creaming off each sale of the 79xx's. The cards they've delivered simply are overpriced.

Looking forward to seeing Kepler... hopefully we see it as indicated for GDC 2012 (March 7th to the 9th)!
 
It is just hilarious to defend NV pricing. Anyway, my sources indicate 7950 performance or slightly slower, you guys will see in a few weeks. The big bomb (110/112) is the one you want which unfortunately won't be released until late this year. That one is plausible for 50% faster than 580, the GK104 will not be.
 
It is just hilarious to defend NV pricing. Anyway, my sources indicate 7950 performance or slightly slower, you guys will see in a few weeks. The big bomb (110/112) is the one you want which unfortunately won't be released until late this year. That one is plausible for 50% faster than 580, the GK104 will not be.

Not hearing that from anywhere else, but we'll see quite sooner than a few weeks I think by now :). As far as defending pricing, the 79xx keeps being defended as based off nVidia pricing, and we're saying that the 79xx is overpriced this much later... so... no one's actually defending either pricing at this point in time :confused: . Anyway, Kepler should bring things back down to Earth a bit at minimum, if not a lot, with any luck. :D
 
It is just hilarious to defend NV pricing. Anyway, my sources indicate 7950 performance or slightly slower, you guys will see in a few weeks. The big bomb (110/112) is the one you want which unfortunately won't be released until late this year. That one is plausible for 50% faster than 580, the GK104 will not be.

My sources indicate a 30000000000% increase over the HD 7970 and it will consume 5 watts on load.
 
Nah. New generations released well over a year later are supposed to offer a lower price for the same performance, not a large hike (25% price) for a bit more (20%) performance. That's why everyone keeps bringing it up as a poor value: it's 14 months later, on a new node, and yet it isn't even a good price/perf for what it does deliver, which in and of itself is underwhelming. It isn't compelling at all for the high-end buyer.

The alternative is for them to be the same price for a lot more performance, which also happens often, but the Tahiti cards failed completely on that too as I mentioned above. Hopefully Kepler gets the market back in line to normal... AMD is feasting on the large amount they probably are creaming off each sale of the 79xx's. The cards they've delivered simply are overpriced.

Looking forward to seeing Kepler... hopefully we see it as indicated for GDC 2012 (March 7th to the 9th)!
It probably seems that way to someone who doesn't understand simple economics.
 
No it seems that way to people that actually know anything about video hardware.

7 series is overpriced, offers less IQ than the competition, less support, less 3D capability, small bump in performance, 14 months later. Knock $100 off and it would be an ok buy.
 
No it seems that way to people that actually know anything about video hardware.

7 series is overpriced, offers less IQ than the competition, less support, less 3D capability, small bump in performance, 14 months later. Knock $100 off and it would be an ok buy.

Yet it's sold out everywhere. Somebody's laughing on their way to the bank.
 
Yet it's sold out everywhere. Somebody's laughing on their way to the bank.

At least not with my money.

It probably seems that way to someone who doesn't understand simple economics.
Economics didn't help the consumer in this situation, only the seller received the benefit of it.
Thankfully we have two companies to choose between and loads of older gen handware at good prices.

What the product costs and what the consumer is willing to pay must be in alignment. When it's not, the value is lost no matter what the economics is.
Video cards are a "want" item, so people like myself have to see value before purchasing.
 
Last edited:
Value is completely subjective, and it's dumb to argue otherwise. It's also silly to say stuff like "it's not what people in the high-end market want" considering it's always sold out in a few hours.

You think it's a poor value? Great. Just don't buy it. I don't know why certain posters have to insist in telling everyone that their 7970 wasn't worth it.
 
Yet it's sold out everywhere. Somebody's laughing on their way to the bank.



HD7950 isn't sold out every where.

HD7970 is a good card but it's really not a great leap in performance from the GTX580.
 
Status
Not open for further replies.
Back
Top