NVIDIA Pascal Thread

[Kris194]

RussianSensation, did you born yesterday? It's normal that at some point you're selling better product for lower price. What you have to understand is that lower price is not equal to less money to manufacturer. GTX 970 is the best example.

http://www.techpowerup.com/mobile/2...-ti-gtx-780-gtx-770-cuts-gtx-760-pricing.html

 

[RussianSensation]

RussianSensation, did you born yesterday? It's normal that at some point you're selling better product for lower price. What you have to understand is that lower price is not equal to less money to manufacturer. GTX 970 is the best example.

http://www.techpowerup.com/mobile/2...-ti-gtx-780-gtx-770-cuts-gtx-760-pricing.html

When 970 launched, AMD had 290/290X. Now apply 2X perf/watt to 970 today to the most recent TPU charts, and you end up with a card that wipes the floor with a 980Ti. 970 never wiped the floor with a 290X/780Ti. So that means one of these:

1) 2X perf/watt is a lie just like it was for Maxwell;
2) NV will lowball clock speeds on 1070 to not wipe the floor with a 980Ti
3) NV will raise the price above $330
4) There will be a 3rd tier 1060Ti GP104 chip
5) AMD is toast if 1070 is a $330 970 replacement with 2X the performance of a 970. Alternatively, AMD has a GP104 competitor but what is it? Dude, if 1070 ~ 980Ti for $329 in a 145W TDP, AMD would need to drop the Fury X to $249 to sell any.

The market dynamics of when 970/980 showed up and where Nano/Fury/Fury X sit, and NV's market share are completely different. If I am NV today, why the hell would I sell a card faster than a 980Ti for $330 when my customers are paying $450+ for a 980 and AMD has nothing for the next 6 months? Give me some good reasons why. When 970 launched, after-market 290s were selling for $350-375. That means it was in NV's best interest to slightly undercut the 290 to make sure it's not relevant. What does NV need to undercut today to sell the 1070? Nothing since AMD has nothing worth buying above the $300 R9 390. So unless AMD pre-emptively drops the Nano to $349, why wouldn't NV sell 1070 for $399-429? You are telling me if 1070 ~ 980Ti and costs $399-429, it wouldn't sell like hot cakes? The market is changing -- both companies want to raise prices to maximize profit margins. Nearly every market segment had risen in price since 2012. I mean with the 960 replacing 760, NV literally didn't even try, delivering the worst performance increase for a x60 series card in 5 consecutive generations. And yet, 960 is one of the best selling cards this gen!! If I am NV, I am sitting and laughing. $$$. No need to lowball anything since the market has spoken.

If anything, I'd raise prices even more. Move GP106 to $199-249, GP104 1060Ti to $299-349, GP104 1070 to $399-499, GP104 1080 to $599-649.

 

[VirtualLarry]

I would be very surprised if GP104 used standard GDDR5. Assuming a 256-bit bus width (which has been the case on every "4"-series Nvidia GPU up until now), it would be completely hamstrung by memory bandwidth limitations and not running anywhere near its full potential. This chip isn't going to be viable without GDDR5X.

Why not a "Ti" release later on, with GDDR5X? (Are they pin-compatible, just different clock-speeds? Or is there a pinout difference? Still, they could do a re-layout of the board, I suppose.)

 

[Silverforce11]

The way I am reading the rumors is that Polaris 10 and GP104 are not competitors. 1070/1080 are 970/980 replacements --> $330-550 market segment. By definition Polaris bringing 970/290/290X minimum spec to the masses is a completely different market for next gen than where 1070/1080 will be. That's my take on it. That's why I insist that AMD has no plan for most of 2016 to counter 1070/1080 other than lower prices on Nano/Fury/Fury X.

We already called it ages ago.

One only has to look at the die sizes to know instantly where Polaris 11 and 10 are going to fall in the market. These are not mid-range next-gen products. It's only entry and mainstream and Raja & others at AMD have been quite clear they are aiming to bring great perf per dollar to the market.

AMD don't need to counter GP104 because it's at another market level. If it launches in June, then AMD will just forfeit that market segment until Vega is ready. Fiji 4GB can't even compete with the 980Ti, it has no chance competing against GP104 with 8GB vram.

Their best bet is to bargain discount it prior and hope to sell all the last production runs of Fiji SKUs, while the rest goes into Fury VR/Duo for professional market and the new Firepro for HPC. Consumers won't touch 4GB Fury or Nano when there's a 8GB next-gen mid-range for ~$400.

That's assuming GP104 launches in June.

Vega could also arrive in Q4 2016. According to the Roadmap, that's as solid as Q1 2017.

 

[ShintaiDK]

We already called it ages ago.

One only has to look at the die sizes to know instantly where Polaris 11 and 10 are going to fall in the market. These are not mid-range next-gen products. It's only entry and mainstream and Raja & others at AMD have been quite clear they are aiming to bring great perf per dollar to the market.

Exactly. :thumbsup:

One also has to remember what AMD and Nvidia have to pay for these chips.

Even Vega 10, AMDs top bin is essentially just a Fiji shrink without a chocking low VRAM limit and some balancing. And I expect Nvidia to do the exact same, if we exclude the GP100.

Rumours also seems to mention a GP102, perhaps a smaller gamer version of GP100 that will park GP100 into the Tesla/Quadro area for good.

GDDR5 will be a limit on any performance increase as well, until GDDR5X hits. Because HBM isn't going to be anything than flagship for a very long time.

I can easily imagine more 14/16nm GPUs released as late as in 2020 as leading edge.

 

[Kris194]

RussianSensation, no one, literally no one said, that GTX 1070 will cost 330$. It's more than likely, that it will cost 349 or even 399$ just like GTX 670. Moreover, 2x perf/watt doesn't mean that new cards will be 2x faster. No matter if it will cost 349 or 399, it will be still selling like hot cakes if it will beat Titan X. I guess that Nvidia may want to lower TDP slightly to put GP104 in notebooks, just like they did it with Maxwell.

 

[jpiniero]

If GP104 actually did use a 256-bit bus with standard GDDR5, then it's not going to be beating GM200 at all; it's going to be lagging behind because the needed memory bandwidth just won't be there. I don't think Nvidia is going to release a product like that.

384-bit is possible but I wouldn't count on it. I suppose Pascal could have some bandwidth saving features to help partially mitigate the issue. I suppose they could only hard launch the mobile parts and the X70 in July (which won't be fast enough to really need GDDR5X) and just paper launch the X80.

 

[SteveGrabowski]

What would you do if you were NV? Honest question -- I can sell 980 for $450+ in 2016 so why the heck would I sell 1070 ~ 980Ti for $329?

If it's decently profitable, I would sell it for about $360 so to make sure no one bought or waited for AMD and to get all the 900 series owners to upgrade in addition to 600 and 700 series owners. That way they can pre-emptively kneecap Vega, it would be a small enough price increase over the 970 that people would still buy it like crazy just like the 970. If Nvidia thinks longterm here maybe in a couple of years they can jack the prices way up and keep them there. I mean Intel isn't going to duplicate that performance on their integrated graphics considering the difficulties they're having with node shrinks and if Vega fails I'd imagine Nvidia would drive them into the entry to lower midrange level, leaving them alone for supplying enthusiast gpus. I have to imagine that was the plan for the 970, and it slaughtered AMD just as they were about to recover from the end of the mining boom. With how unstable AMD is the last thing I want is for them to get a nice revenue stream from Vega if I'm Nvidia.

 

[dangerman1337]

I can easily imagine more 14/16nm GPUs released as late as in 2020 as leading
edge.

I can see Nvidia's Volta & AMD's Navi being on an improved 14/16nm process & HMC with big architecture improvements with DX13 (perhaps?) and then the GPUs in 2020 being on 10nm maturely developed or even 7nm along with extra architectural improvements.

 

[Kuiva maa]

The way I am reading the rumors is that Polaris 10 and GP104 are not competitors. 1070/1080 are 970/980 replacements --> $330-550 market segment. By definition Polaris bringing 970/290/290X minimum spec to the masses is a completely different market for next gen than where 1070/1080 will be. That's my take on it. That's why I insist that AMD has no plan for most of 2016 to counter 1070/1080 other than lower prices on Nano/Fury/Fury X.

This seems like the fastest way for AMD to take its market share at single digits, if true. If Polaris 10 is as slow as you expect it to be, AMD will be offering no new SKU that would be an actual upgrade for hawaii/GM204 users and upwards.Imagine loyal AMD customers on hawaii since late 2013 waiting for something better than Fury X - this sends them straight to nvidia. AMD's only chance to sell anything midrange or higher would be to attract a small portion of the crowd that did not find Fiji worthwhile the first time round and budget buyers that still use pitcairns etc and will be looking for an upgrade in the 150-200 range. AMD will have to price their Fiji respins really low as well, since nvidia will be ruling the roost at 450+ with much higher performance. AMD getting pigeonholed in cheap brackets while competing with their 28nm SKUs smells disaster.

 

[Sweepr]

5 out of 10 Amazon best-sellers are in the $300-350 price range. Targeting current Hawaii/GM204 buyers doesn't sound like a bad strategy.

 

[Kris194]

No way that Polaris 10 will cost 350$. Polaris is way too small GPU to cost that much. I'll be surprised if it will cost more than 250$. Nvidia and AMD are targeting way different market segments with their upcoming cards.

 

[Head1985]

If polaris 10 is fast as GTX980 and have TDP 120w and cost 250-300USD then AMD have winner.
ALL nv cards will be obsolete crap vs polaris 11 and 10(entry to GTX980 level almost entire market) because Die shrink.AMD will gain good market share...

 

[Adored]

The 7870 was $350 on release, seems unlikely that the full Polaris 10 will be cheaper this time around, but the cut-down part could easily be $250.

 

[Qwertilot]

Not that I totally disagree, but to pedant - 980's are already only 165w, so would be far from obselete crap vs a 120W / identical performance die shrunk competitor. They'd have to price cut a lot of course but fine for stalling for a few months.

That's presumably the plan if they're doing GP104 first.

Trickier down at lower tdp's perhaps.

 

[Kris194]

The 7870 was $350 on release, seems unlikely that the full Polaris 10 will be cheaper this time around, but the cut-down part could easily be $250.

Look at the current price of 390. I really doubt that Polaris 10 will beat 390.

 

[Adored]

Look at the current price of 390. I really doubt that Polaris 10 will beat 390.

If Polaris 10 (the bigger one) can't beat the 390 then AMD has really messed up somewhere. Even if we just assume a basic 50% shrink then Polaris 10 should be 464mm2 on 28nm, which is larger than Hawaii.

The 7870 beat the GTX 580. Bandwidth might be an issue but Polaris will have compression so I don't think so.

 

[Kris194]

The question is whether Polaris 10 will really have 2560 shading units. If yes then it may beat Fury easily with all announced improvements.

 

[Kuiva maa]

5 out of 10 Amazon best-sellers are in the $300-350 price range. Targeting current Hawaii/GM204 buyers doesn't sound like a bad strategy.

Targeting is not enough. This crowd have had this performance offered to them at that price for almost two years now and they will be looking for upgrades, not for the same performance. Either AMD is fully relinquishing desktop to Nvidia or Polaris 10 is much more powerful than we think.

 

[Sweepr]

They don't need a Fiji killer at mainstream prices. Give users better than Hawaii performance, new features and lower power consumption at the same price point and the masses will buy it.

Back to topic, Sweclockers mentioned two Pascal GPUs launching after Computex (GM206 successor included) and mobile chips ready for back to school season, so a 'GP106' might not be far from desktop launch.

 

[jpiniero]

Back to topic, Sweclockers mentioned two Pascal GPUs launching after Computex (GM206 successor included) and mobile chips ready for back to school season, so a 'GP106' might not be far from desktop launch.

I would be surprised if they released GP106 desktop parts at Computex. Mobile parts yes. I'd have to say October for the better one and January for the worse one.

 

[Qwertilot]

Some sort of mobile pascal pre Xmas presumably more or less a given with the profits riding on it.

Targeting is not enough. This crowd have had this performance offered to them at that price for almost two years now and they will be looking for upgrades, not for the same performance. Either AMD is fully relinquishing desktop to Nvidia or Polaris 10 is much more powerful than we think.

Umm, neither? With Vega we know that they're going to be rolling out a full 14nm desktop range out over a period of 6-12 months. Surely a perfectly sane time scale for a full stack refresh?

It looks a bit odd, but this sort of thing will likely be visible with Pascal to some extent too. Simply down to the die shrink being such a big jump that using 28nm parts to fill the ranges in will be distinctly non trivial.

 

[digitaldurandal]

If Nvidia launches a 980 replacement and AMD is ~6 months behind I will be hard pressed not to switch to back to Green Team. Once a DP 1.3 120hz 1440p Ultrawide without ghosting hits with either Freesync or Gsync the current market will determine my team, probably for five years due to the monitor purchase.

As 1440p Ultrawide monitors drops in price I think that a growing portion of enthusiast gamers will end up locked into a team as well.

 

[Silverforce11]

AMD did their 28nm product stack pretty quick.

6 months is a long gap between Polaris and Vega, and the reason I think, is HBM2, the stacking process itself adds months to the production line.

If GP104 is on GDDR5, performance isn't going to be as great as some folks hope. Even the 970/980 now is already bandwidth bottlenecked, it needs VRAM OC to scale. What's their angle?

1. 384 bit bus on GP104? Would be unprecedented but definitely possible. Basically 980Ti class bandwidth. With improvements in prefetch, cache and compression, they can hit a slightly higher performance profile than the 980Ti with GDDR5.

2. 256 bit bus, performance would be very bandwidth bottlenecked, in games where the compression is efficient, there will be good performance, otherwise it's < 980Ti.

Both of these are not ideal. The ideal is 256 bit bus + GDDR5X. But this means delays to release, paper launches etc. However, when NV is so far ahead, and their current GPUs sell, they can afford delays go for the ideal scenario. GP104, 256 bit bus + GDDR5X will be a great performer while being cheap to manufacture. Win-win scenario.

 

[Kris194]

If Polaris 10 (the bigger one) can't beat the 390 then AMD has really messed up somewhere. Even if we just assume a basic 50% shrink then Polaris 10 should be 464mm2 on 28nm, which is larger than Hawaii.

The 7870 beat the GTX 580. Bandwidth might be an issue but Polaris will have compression so I don't think so.

You shouldn't look like that at this. You have to remember that AMD have to lower TDP comparing to their 28nm GPUs. Everything depends whether full Polaris 10 has 2560 Shading Units. If yes then with all improvements made it will easily beat 390x.While its shader compute looks good, memory bandwith seems to be really shitty.