** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[tbh]

If small Pascal is first up it won't be much faster than the GM200 cards.

If it offers even a small performance increase over GM200 whilst having a similar price and TDP to GM204 then I'll be happy; I suspect though that it will fall foul on the price front, seeing as it's on an expensive new process and all.

How far can mid- and high-end prices go though before people cry "enough!"? It seems that the sky's the limit for enthusiast parts, but I'm not convinced that will apply to all market segments.

 

[Orangey]

I can't believe people believed that 10x thing.

Did you think your games were going to go from 30 to 300fps?

 

[FoxEye]

If small Pascal is first up it won't be much faster than the GM200 cards.

Which small Pascal? 04 or 06? The latter won't be faster than GM200 at all. Probably a darn sight slower.

Would be nice if the 1070 is as fast as the 980ti, but I'm personally expecting it to be a tad slower. Between the 980 and 980ti would be my guess.

I just don't think nV want to give customers such a big leap in performance, unless they think AMD will force them to.

 

[Boomstick777]

I can't believe people believed that 10x thing.

Did you think your games were going to go from 30 to 300fps?

That 10x thing refers to the pricing.

 

[Seanspeed]

That 10x thing refers to the pricing.

I suspect you're just joking, but it's hard to tell around here, where people seem to believe any wild story you tell them. But anyways:

No, it referred to a very weird combination of tests, most based on compute performance.

 

[bru]

Whatever NVidia launch with first it will be faster than the 980ti/TitanX.

I believe that people are seriously underestimating the performance leap we are soon to witness.

If it is from high end to high end then we could be looking at a similar jump from a NV285 to original Titan (best 55nm part to first full fat 28nm part), that's over 3* performance.

If it is from high end to mid range then we could be looking at a jump from a Nv285 to NV680 ( best 55nm part to midrange 28nm part) even that is over twice the performance.

Now I am not suggesting that we will see 3* the performance, but to think that NVidia will hold back and only just sneak in with a midrange GP104 part that is only just faster than the currant 980ti is sheer madness (in my opinion).

Unless of course you think AMD will be doing the same, as I can pretty much guarantee you that both sides will have vaguely similar performance, just like they always have.

Dissclaimer ( this post is my own opinion and as such is probably complete nonsense)

 

[FoxEye]

Unless of course you think AMD will be doing the same, as I can pretty much guarantee you that both sides will have vaguely similar performance, just like they always have.

Dissclaimer ( this post is my own opinion and as such is probably complete nonsense)

Well AMD have only claimed 2x perf-per-watt for their new generation. It may be that AMD are not capable of delivering quite such a drastic leap in performance with their next gen, compared to what Pascal might be capable of delivering.

And all nV will want to do is roughly match AMD. They won't want to give away tons of performance for free. Although I suppose if they were massively faster they could just charge twice as much

Either way, I think it pays not to get your hopes up for low-end Pascal cards to beat a 980ti. I don't think they will. Consumers these days would settle for much less, and still be happy to open their wallets.

e: nV will still have plenty of 900 series stock in the channel, you would assume, upon the launch of Pascal. What benefit would it be for them, if GP106 (etc) beat a 980ti? They'd be left with inventory that they could not shift unless they practically gave it away.

And when has it ever been the cast that a low end card beat the previous gen's flagship lite? I can just about see the 1070 being close to a 980ti*, but a 1060? I don't think there's a bat's chance in hell of that.

*Personally I think the 1070 will be somewhere between the 980 and 980ti. Not even faster than the ti at all

 

[Boomstick777]

Whatever NVidia launch with first it will be faster than the 980ti/TitanX.

I believe that people are seriously underestimating the performance leap we are soon to witness.

If it is from high end to high end then we could be looking at a similar jump from a NV285 to original Titan (best 55nm part to first full fat 28nm part), that's over 3* performance.

If it is from high end to mid range then we could be looking at a jump from a Nv285 to NV680 ( best 55nm part to midrange 28nm part) even that is over twice the performance.

Now I am not suggesting that we will see 3* the performance, but to think that NVidia will hold back and only just sneak in with a midrange GP104 part that is only just faster than the currant 980ti is sheer madness (in my opinion).

Unless of course you think AMD will be doing the same, as I can pretty much guarantee you that both sides will have vaguely similar performance, just like they always have.

Dissclaimer ( this post is my own opinion and as such is probably complete nonsense)

Ya am fully expecting the new midrange to beat the current high end. Problem will lie in whether AMD / Nvidia sell us the new 'midrange' as the high end, and do a round of milking before the actual high end cards get launched.

If this is the case I will hold off an upgrade for as long as possible. I want a full fat card, whatever the maximum chip capacity of next gen cards is. I.e Titan from the Nvidia side or AMD's best.

Fancy a move to a full AMD setup next year, Zen CPU / Radeon GPU. Depends on what Nvidia's Pascal looks like, have a feeling it will be a real monster. Just plz no pretend high end cards, full fat first..

 

[RanxZy]

There has been a pattern of NVIDIA holding back the full fat high end chips for several months and initially selling smaller, efficient chips on the 680/980 at high prices. This has been very profitable for them and I would be shocked if they released a true high end card with the launch of Pascal that is 70% faster than a Titan X. I think 50% absolute max with an OC while being very efficient

 

[DarrenM343]

There has been a pattern of NVIDIA holding back the full fat high end chips for several months and initially selling smaller, efficient chips on the 680/980 at high prices.

I think there might be an element of the higher end chips just not actually being ready or available initially. I personally doubt that nVidia suddenly has all these chips from mid to high end at the very launch of a new range, and simply holds back the high end chips, but I could be wrong. Even if they can choose which one to manufacture first,the better chips to manufacture will be the mid-range as more will buy them, even if it has a slight premium initiallly as the higher end chips are not available.

if those with high end current gen want high end next gen they should sit it out for a while - the card(s) they already have will be plenty fast enough

 

[James J]

Theoretical max performance and actual max performance are two completely different things. I'd expect the top SKU pascal chip to be 40% faster than the top SKU maxwell one.

 

[Orangey]

It will be interesting to see if GP100 can actually use all the CUDA cores. It is supposedly more than twice the transistors.

 

[tbh]

*Personally I think the 1070 will be somewhere between the 980 and 980ti. Not even faster than the ti at all

I'd be disappointed if this turns out to be the case, given that this will be the first new node for over 4 years by time of release.

 

[TNA]

I'd be disappointed if this turns out to be the case, given that this will be the first new node for over 4 years by time of release.

tbh me too

 

[D1craig]

you are all delusional if you think they are going to release only cards that are faster than what is out atm. i dont think they have done that previously.

i think they will release something that is close to 980ti being top end card for a few months, if AMD beat them then the gap between cards will be shorter.

they know they have a similar fan base to apple - will buy this stuff even though its **** and ****es all over them whilst getting a direct link to their bank accounts.

we have many members on these forums that buy the top end whenever it is released. whether something is released 1-3 months later that is better than it.

EDIT - i bought a 970 at launch and it was only £260. if the pascal equivalent is more than £100 more expensive i wont be buying it.

 

[Orangey]

The process is tuned for mobile. That means if they try and make a speed demon it's going to suck inordinately more power. Add to that they are still using GDDR5 on mid-range. They will also be keeping the die sizes in check on the earliest cards to keep costs down.

I don't see 1070 being faster than 980Ti.

 

[Seanspeed]

you are all delusional if you think they are going to release only cards that are faster than what is out atm. i dont think they have done that previously.

i think they will release something that is close to 980ti being top end card for a few months, if AMD beat them then the gap between cards will be shorter.

they know they have a similar fan base to apple - will buy this stuff even though its **** and ****es all over them whilst getting a direct link to their bank accounts.

we have many members on these forums that buy the top end whenever it is released. whether something is released 1-3 months later that is better than it.

EDIT - i bought a 970 at launch and it was only £260. if the pascal equivalent is more than £100 more expensive i wont be buying it.

They've been able to do this because they've been able to put out products competitive with what AMD has at the time.

The 970 and 980 would not have gone nearly as well had they had to go toe-to-toe with the 390 and Fury line being released at the same time.

They will not have that luxury this time out. AMD will be itching to put out their next generation cards at the same time as Nvidia. If Nvidia pull out the weak guns and AMD pull out the big guns, Nvidia would be left looking silly. I would certainly go AMD in that situation as it'll be time for me to upgrade, especially with all the demanding VR content out by then. I wouldn't be the only one, either. Nvidia have some loyal buyers, but they'd generally been able to justify their good sales with highly competitive products. I think people's loyalty will only go so far if the 'high end' Nvidia product is being smoked by the genuine high end AMD product that releases at the same time.

 

[drunkenmaster]

Whatever NVidia launch with first it will be faster than the 980ti/TitanX.

I believe that people are seriously underestimating the performance leap we are soon to witness.

If it is from high end to high end then we could be looking at a similar jump from a NV285 to original Titan (best 55nm part to first full fat 28nm part), that's over 3* performance.

If it is from high end to mid range then we could be looking at a jump from a Nv285 to NV680 ( best 55nm part to midrange 28nm part) even that is over twice the performance.

Now I am not suggesting that we will see 3* the performance, but to think that NVidia will hold back and only just sneak in with a midrange GP104 part that is only just faster than the currant 980ti is sheer madness (in my opinion).

Unless of course you think AMD will be doing the same, as I can pretty much guarantee you that both sides will have vaguely similar performance, just like they always have.

Dissclaimer ( this post is my own opinion and as such is probably complete nonsense)

In absolutely no way is the drop from 55nm to 28nm akin to the drop from 28nm to '16' nm, none.

20 + 16nm together is producing the kind of optical shrink and power characteristics that previous single node drops have achieved. A 500mm^2 16nm chip should only be looking at twice the transistor count and similar power usage to a 28nm 500mm^2 chip at similar voltage/clock speeds.

The 28nm to 16nm drop in node is far far closer to the 40nm to 28nm drop than 55/65 > 28nm.

The 20nm node drop effectively didn't come anywhere near achieving the power drop required in a normal node which is why no one liked it. Ultimately most new nodes bring around a 1.85-1.95x transistor count into the same area with a near halving of power usage per transistor at similar performance level. 20nm achieved fairly close on the transistor density but achieved more like 15% power drop rather than 50%. 16nm which is built on top of 20nm metal layers, doesn't shrink much at all, 5-15% smaller, no where near a 50% shrink of a normal node but brings around another 30% power drop.

Add the 20 and 16nm nodes together and you pretty much achieve the normal double transistor density at half the power.

 

[Seanspeed]

The process is tuned for mobile. That means if they try and make a speed demon it's going to suck inordinately more power. Add to that they are still using GDDR5 on mid-range. They will also be keeping the die sizes in check on the earliest cards to keep costs down.

I don't see 1070 being faster than 980Ti.

I would bet money it will be. x70 cards are upper midrange and if they do use GDDR5, I expect it will be GDDR5X, which seems promising according to what we've heard.

As for sucking power, both AMD and Nvidia are claiming 2x performance per watt, so I highly doubt they're going to struggle that much here. Maybe we wont get 1300Mhz GPU's out-of-the-box or whatever, but there's still a lot of power efficiency gains to be utilized.

 

[DmitryD]

Those chips won't overclock well.