NVIDIA Pascal Thread

[C@mM!]

Sorry but this is just completely wrong. When 7970 Launched it had a price of $550 and it most certainly did not have good drivers at launch, the significant driver improvements AMD saw, didn't come until quite a bit later.

Furthermore Nvidia pricing their GK104 die (GTX 680) at $500 also came before AMDs driver improvements.

So AMDs drivers has nothing to do with the increased prices we saw for GCN and Kepler.

Lastly AMD didn't match Nvidias prices, Nvidia matched AMDs prices, seeing as 7970 launched first and 680 second. And at launch the 680 was the faster product.

Want to go back and see what you wrote again, as you just confirmed what I said mostly.

1\ 7970 pricing was one of the last of the 'high end cards priced reasonably'. 680 small die was seen as competitive at the price point, so Nvidia priced GK104 accordingly, thus starting small die at big price strategy, and creating a new halo product in big die GK110 released only 8 months later (and rumour has was specifically held back due to maximise profits from 104) in the form of the Titan (and later Ti cards). Nvidia from 6xx to 9xx continues to abuse its market position and performance lead by increasing card prices, usually as a percentage of how much faster each card is than AMD's competing solution at each price point.

2\ AMD was forced to cut the price on its cards aggressively until its driver rewrite gave its performance back.

3\ With AMD now competitive again, it started slotting cards against Nvidias now higher priced cards.

 

[tviceman]

If they artificially assign the 1080Ti name to GP104, might as well raise the price to $649. The mid-range transition from $249 560Ti to $499 680 to $549 980 to $649 1080Ti will be finally fully completed. NV will have successfully established mid-range at new flagship prices, pushing their gross margins beyond 55%.

And now the true GP100 (102?) flagship probably won't see the light of day until March-June 2017.

It will all depend on performance. If GP104 is 25-30% faster than Titan X (like GTX680 was over GTX580) then I think it'll be $699 for the TI, $549 for the 1080, and $399 for the 1070. If GP104 is only %15 faster than it won't be able to come in that high.

GP102 as the Titan. I didn't believe the GP102 rumors until we all saw how devoted to DP GP100 is, even over GK210 and GF114 for their time at release.

 

[Head1985]

It will all depend on performance. If GP104 is 25-30% faster than Titan X (like GTX680 was over GTX580) then I think it'll be $699 for the TI, $549 for the 1080, and $399 for the 1070. GP102 as the Titan. I didn't believe the GP102 rumors until we all saw how devoted to DP GP100 is, even over GK210 and GF114 for their time at release.

i hope you are not working in Nv marketing.

 

[tviceman]

i hope you are not working in Nv marketing.

It's the logical progression given the state of the industry and past precedence, but GP104 has to be at least over Titan X as GTX680 was over GTX580 at release. If not, then GP104 can't occupy the TI spot with only a ~20% improvement over the 980 TI.

It sucks, but people need to stop looking at die sizes to determine prices. We have long ago kissed goodbye the 1.75-2.0x performance increase within the same price bracket.

 

[Head1985]

By your logic we will pay in 10years 2000USD for GP 104like SKUs,1000USD for GP106SKu/GTX960/1060....


You know its called progress.Old cards getting cheaper or getting replaced by new cards at same price point with better performance.
GTX970/980 are almost 2year old.
980TI 1 year.

 

[Timmah!]

I think nVidia would be shooting themselves in the foot if they made GP104 a top end Ti part. because typically that is reserved for a big die, big price card.

AMD has pretty much said the purpose of the Polaris chips is to get VR to the masses with a well priced chip. meaning it will not be $650 dollars. That will be reserved for Vega come late 2016 or early 2017 (the road map doesn't point down an exact date, just a range).

So I am thinking nVidia cannot make this a Ti part, that has to be reserved for GP102.

Maybe the Ti part from this latest rumor IS GP102. Its only everyone´s assumption all rumored cards are GP104 based.

 

[Snarf Snarf]

This years releases really are looking to be less than exciting :'(. The bright side is that a lot of us who bought 980ti's last year probably won't be upgrading until 2017 with Vega/GP102(if it exists). More money for other hobbies this year at least

 

[extide]

If they artificially assign the 1080Ti name to GP104, might as well raise the price to $649. The mid-range transition from $249 560Ti to $499 680 to $549 980 to $649 1080Ti will be finally fully completed. NV will have successfully established mid-range at new flagship prices, pushing their gross margins beyond 55%.

And now the true GP100 (102?) flagship probably won't see the light of day until March-June 2017. This would be a 3rd and complete generation of confirming bifurcation of a generation into 2 fully distinct parts: 1st half using mid-range as flagship; 2nd part the real flagship comes out.

We as gamers are screwed now because it means an extra year or so wait for a real flagship (but then next gen Volta is 1 or so year away), or put up and shut up and pay flagship prices for mid-range if you want cutting edge tech right away. This is a brilliant strategy reversal of Fermi. Essentially front-load a highly clocked 460 at $500+, then release cut-down GTX470/480 9-12 months later, and finish it off with a flagship 580, but call it Titan series. Then once they launch the full $1000 Titan X Pascal and $699 GP100/102 gaming card, they will conveniently lower the price of the mid-range GP104 to $449-499, much like they did with the 980. The end result is the mid-range next gen series is overpriced for the entire duration of the generation. Well played.

NV is so obvious now, it's not even funny, but it is sad that the graphics card market became this way in just 4 years.

Dude, while it is true they moved the xx04 chips up in the stack, in some cases, the whole concept of nVidia ripping off the consumers is total BS. They, and every other company out there, sell their products at the max prices the markets will bear. That means they need to take into account the other products they are competing against at the time, etc. Ultimately WHO CARES what the card or the chip is called, it's ultimately all about price and performance, That's IT.

 

[Headfoot]

Dude, while it is true they moved the xx04 chips up in the stack, in some cases, the whole concept of nVidia ripping off the consumers is total BS. They, and every other company out there, sell their products at the max prices the markets will bear. That means they need to take into account the other products they are competing against at the time, etc. Ultimately WHO CARES what the card or the chip is called, it's ultimately all about price and performance, That's IT.

Exactly. You pay the price of the flagship card, and you get a midrange card's performance. That's precisely the problem.

 

[msi2]

http://pctuning.tyden.cz/component/...p100-pujde-i-do-hernich-grafik-nejen-do-tesel

It has to be 30% over 980Ti...

 

[Glo.]

What if X80 will have 30% higher performance over GTX980?

What you expect from improved Maxwell architecture with higher boost clocks?

Everything falls perfectly in line. GDDR5X - X80 Ti Model. GDDR5 - cut down full GP104 die, with 1400 MHz boost clock. 30% higher boost clock than GTX 980. On a node that brings 70% reduction in power consumption, or 30% higher performance. With slight tweaks compared to Maxwell.

Pascal IS Maxwell. With slight tweaks that allow to differentiate them. Mahigan was right about it, that After Maxwell was supposed to come Volta, but 20 nm failure made Nvidia go back to 28 nm with this architecture. And Pascal Is what Maxwell was supposed to be from the beginning.

 

[96Firebird]

Exactly. You pay the price of the flagship card, and you get a midrange card's performance. That's precisely the problem.

Pretend there are no larger chips, because there won't be in a gaming card.

There, flagship performance in a flagship card.

amd is faster in every game coming out with dx12 so thats a lack of competition?

I think you missed the part where I said "If that does happen...", referring to the future scenario outlined by Russian. Nothing about today's lineup.

Re-read it again, I think you'll be able to understand it this time!

 

[Pinstripe]

GP104 for 1080Ti?

Would make more sense to have a 1080, 1070 and 1060 Ti.

 

[Krteq]

Maybe the Ti part from this latest rumor IS GP102. Its only everyone´s assumption all rumored cards are GP104 based.

Yeah, but no one ever seen this mythic GP102.

 

[Timmah!]

Yeah, but no one ever seen this mythic GP102.

True that. But nobody really has seen GP104 either, bar that single picture, which may or may not be a fake.

Personally, i believe there will be GP102. But IMO it will be a Titan card, at least at first.

 

[Stuka87]

Yeah, but no one ever seen this mythic GP102.

Nobody has seen ANY pascal based part outside of the tesla GP100 card.

The GP102 has shown up in drivers, and of course in rumors, and it is expected to be the big chip gaming GPU (ie: Titan and later 1080Ti?).

 

[Actaeon]

So far these rumors have been disappointing. I know it is a few years out, but at this point it is looking like Volta will be the next meaningful upgrade for Maxwell owners. These rumors make it seem like Pascal will just be shrunk Maxwell with minor performance and efficiency improvements across the lineup. Maxwell was efficient enough, so I'd rather they have optimized performance instead. For anyone on pre-Maxwell, Pascal is looking like a great upgrade.

Volta will bring Improved Async, brand new architecture, broader HBM2 support, and built on matured 16nm finfet process. Assuming it comes out in 2018 that will be a perfect time to build a brand new rig to replace my trusty OC'd Sandy Bridge.

 

[Headfoot]

Pretend there are no larger chips, because there won't be in a gaming card.

That's very, very far from confirmed... I'm 99% certain there will be a large die (500+mm2) on 16nm at some point in its lifetime sold to consumers usable for gaming (e.g. Titan). The question is when and for how much, not if.

 

[swilli89]

Dude, while it is true they moved the xx04 chips up in the stack, in some cases, the whole concept of nVidia ripping off the consumers is total BS. They, and every other company out there, sell their products at the max prices the markets will bear. That means they need to take into account the other products they are competing against at the time, etc. Ultimately WHO CARES what the card or the chip is called, it's ultimately all about price and performance, That's IT.

Ah the old capitalistic fallacy. With global markets and such constricted source of supplies, this concept is sorely outdated but still gets reiterated by corporate defenders. The price to start a company with relevant technology and skills and compete (in economists circles known as The Barriers to Entry in Porter's Five Forces) is so astronomically high in this game that it doesn't mean that a competitor will sprout up to take the place of a consumer gouging company.

There most certainly ARE fair margins and returns you can expect a company to net if you afford yourself to be a humanist. Greed is a natural, normal human trait and once you add all moving parts of a corporation, the company itself becomes almost maliciously greedy in pretty much every industry and application you can think of. It simply has to be curtailed at a unified consumer level or the federal level every now and then.

 

[antihelten]

Want to go back and see what you wrote again, as you just confirmed what I said mostly.

1\ 7970 pricing was one of the last of the 'high end cards priced reasonably'. 680 small die was seen as competitive at the price point, so Nvidia priced GK104 accordingly, thus starting small die at big price strategy, and creating a new halo product in big die GK110 released only 8 months later (and rumour has was specifically held back due to maximise profits from 104) in the form of the Titan (and later Ti cards). Nvidia from 6xx to 9xx continues to abuse its market position and performance lead by increasing card prices, usually as a percentage of how much faster each card is than AMD's competing solution at each price point.

2\ AMD was forced to cut the price on its cards aggressively until its driver rewrite gave its performance back.

3\ With AMD now competitive again, it started slotting cards against Nvidias now higher priced cards.

I'm perfectly aware of what I wrote, and no it does not confirm what you wrote at all (that the price spikes were related to better competitiveness from AMD due to better drivers).

1\ The 7970 was most certainly not one of the last "high end cards priced reasonably", in fact the 7970 was the the card that originally broke that trend. AMD's previous flagship cards, the 5870 and 6970, cost $380 and $370 respectively, which is a far cry from the $550 of the 7970. Furthermore this was hardly due to AMD now selling a big die GPU (breaking with their small-GPU strategy), since the 7970 was only 352mm2, compared to 334mm2 and 389mm2 for the 5870 and 6970. So basically AMD had a 45-50% price hike, hardly what you would call "reasonably priced", not to mention that there was plenty of complaining on these forums about the price during the launch. And before you start mentioning that $550 is a reasonable price because that's what Nvidia sells their flagship GPUs for, note that those GPUs are 500+ mm2, and not at all comparable.

2\ So? that has nothing to do with the discussion. The discussion is about why Nvidia (and AMD) raised their launch prices relative to previous generations. the fact that prices goes down over time is nothing new, that has always happened when one or the other isn't competitive enough (just look back at what happened to the GTX 280)

3\ Again AMD launched their cards (the 7000 series) first, so no they did not slot their cards against Nvidia, Nvidia slotted theirs against AMD. Since then there hasn't been any comparable midrange (300-350mm2) GPUs launched. The closest we get would be the 400-450mm2 GPUs (Hawaii and GM204), but once again AMD launched theirs first.

 

[tviceman]

By your logic we will pay in 10years 2000USD for GP 104like SKUs,1000USD for GP106SKu/GTX960/1060....


You know its called progress.Old cards getting cheaper or getting replaced by new cards at same price point with better performance.
GTX970/980 are almost 2year old.
980TI 1 year.

Take a breath of fresh air and chill out. That wasn't my logic. Prices went up at 28nm on both camps. From what everyone is saying, fin-fet is no more (and probably less) cost effective at this point in it's life cycle than 28nm. Coupled with newer, high cost memory (GDDR5X and HBM2), and we end up with a perfect storm for higher prices yet again.

The $699 price I listed for a GP104 @ Titan X + 30% speeds would, in fact, would require a stock performing 980 TI to be ~$385 brand new to be equal in price/$. So again, your exaggeration is duly noted but that's all it is - an exaggeration. I don't want to see prices go up anymore than the next consumer but we all live in a concept called reality and unfortunately reality can suck sometimes.

So again, if GP104 performs above GM200 like GK104 did over GF114, I expect
Titan X + 30% - $699
980 TI + 15-20% - $549
980 TI - $379-399

 

[Arachnotronic]

I find it frustrating that many of the people in here who are claiming that NVIDIA is "ripping off we the gamers" seem to be arguing from false premises.

First of all, using die size to compare whether one chip is a worthy successor to another is ridiculous. In this day and age, in order to move to new nodes, the capital intensity goes substantially up due to the need for double patterning.

This means that even though the foundries can achieve improved cost/transistor, cost/mm^2 is actually on the rise. Intel estimates that in going from 22nm (single patterning) -> 14nm (double patterning) led to a 30% increase in wafer cost. That's NOT small.

If we assume something similar for TSMC in going from 28nm to 16nm, then this means that the cost of 300mm^2 of silicon in 16FF+ is about that of ~390mm^2 of TSMC 28nm silicon, assuming normalized yields. TSMC 16nm yields are said to be crazy good right now, so let's assume similar yields.

In addition, the performance/transistor has gone way up, so while some people will sit there and gnash their teeth about how they're not getting the "right amount of silicon die area" it's ultimately delivered performance/$ that matters.

If 300mm^2 of 16FF+ silicon gives you way better performance than 600mm^2 of 28nm silicon, and it is being sold to you for the same price as the 600mm^2 part, then why complain? Why whine that your "rights" as "we the gamers" are being trampled upon by evil, greedy NVIDIA?

If you don't like it, buy an AMD alternative. And, I guess if AMD does the same thing (and they appear to be doing just that), the best I can suggest is to use the "free" Intel/AMD iGPU on your CPU of choice.

 

[swilli89]

I find it frustrating that many of the people in here who are claiming that NVIDIA is "ripping off we the gamers" seem to be arguing from false premises.

First of all, using die size to compare whether one chip is a worthy successor to another is ridiculous. In this day and age, in order to move to new nodes, the capital intensity goes substantially up due to the need for double patterning.

This means that even though the foundries can achieve improved cost/transistor, cost/mm^2 is actually on the rise. Intel estimates that in going from 22nm (single patterning) -> 14nm (double patterning) led to a 30% increase in wafer cost. That's NOT small.

If we assume something similar for TSMC in going from 28nm to 16nm, then this means that the cost of 300mm^2 of silicon in 16FF+ is about that of ~390mm^2 of TSMC 28nm silicon, assuming normalized yields. TSMC 16nm yields are said to be crazy good right now, so let's assume similar yields.

In addition, the performance/transistor has gone way up, so while some people will sit there and gnash their teeth about how they're not getting the "right amount of silicon die area" it's ultimately delivered performance/$ that matters.

If 300mm^2 of 16FF+ silicon gives you way better performance than 600mm^2 of 28nm silicon, and it is being sold to you for the same price as the 600mm^2 part, then why complain? Why whine that your "rights" as "we the gamers" are being trampled upon by evil, greedy NVIDIA?

If you don't like it, buy an AMD alternative. And, I guess if AMD does the same thing (and they appear to be doing just that), the best I can suggest is to use the "free" Intel/AMD iGPU on your CPU of choice.

I actually agree with all that stuff you mention. The only exception is that nVidia is not the foundry. They are not absorbing the cost of switching nodes and they are not getting hit with 30% increase in the cost of doing business. TSMC wisely would absorb much of this cost as it wants to remain competitive against other foundries.

At the end of the day nVidia is probably getting less yields on the new node to begin with but as their financials have shown, their margins have grown and grown the past several years. It boils down to a strict couple of options:

A)As a consumer you shouldn't be ok with this.

B)As a stockholder or other vested stakeholder you should be ok with this.

Sigh.. C) The internet devil's advocate arguer. Arguing "rights" and such. Maybe you're ok with this because you just love capitalism so much you don't mind forking over more money as a sign of your uber patriotism.

 

[DooKey]

That's very, very far from confirmed... I'm 99% certain there will be a large die (500+mm2) on 16nm at some point in its lifetime sold to consumers usable for gaming (e.g. Titan). The question is when and for how much, not if.

Exactly right. Unfortunately the process isn't mature enough for consumer parts so we are stuck with the small die/pipe cleaner for almost a year.

 

[DooKey]

I actually agree with all that stuff you mention. The only exception is that nVidia is not the foundry. They are not absorbing the cost of switching nodes and they are not getting hit with 30% increase in the cost of doing business. TSMC wisely would absorb much of this cost as it wants to remain competitive against other foundries.

At the end of the day nVidia is probably getting less yields on the new node to begin with but as their financials have shown, their margins have grown and grown the past several years. It boils down to a strict couple of options:

A)As a consumer you shouldn't be ok with this.

B)As a stockholder or other vested stakeholder you should be ok with this.

Sigh.. C) The internet devil's advocate arguer. Arguing "rights" and such. Maybe you're ok with this because you just love capitalism so much you don't mind forking over more money as a sign of your uber patriotism.

No one but this echo chamber can hear your whine. Do what every consumer of every product does and vote with your wallet. It really doesn't get any simpler than that.