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

[Simon Belmont]

Those benchmarks wont be a good sign of how well the 1080 can do: "3DMark11 Performance preset is rendered at 1280×720 resolution".

Pascal will draw a bigger lead in more modern games at higher resolutions in more demanding situations. Then there is the usual driver maturity differences.

Its looking like the 1080 is decently faster than a 980ti, as expected.

By certain peoples reckoning AMD should have launched Polaris back in November if Nvidia is 6 months behind AMD!

I have a 970. If true it should be a great upgrade for me

 

[andybird123]

If that is correct 20-30% faster then a 980Ti is not great. Good upgrade for anyone 970 or lower. This could be false?

30% is about what I'm expecting

 

[Simon Belmont]

30% is about what I'm expecting

I guess even Titan X users will see a upgrade.

 

[AlamoX]

30% is more than what i was expecting

 

[Cooper]

30% is precisely what Nvidia want it to be, not what the card is actually capable of if it were let of the leash of GPU market maneuverings.

 

[CAT-THE-FIFTH]

There seems to be a 7.5GB card with quite a high 3DMark score which has also appeared:

http://www.3dmark.com/3dm11/11061015

Edit!!

Scores look weird though!!

 

[Jono8]

Negative: http://forums.overclockers.co.uk/showpost.php?p=27387417&postcount=1232

http://forums.overclockers.co.uk/showthread.php?t=18428457&highlight=Official

GPU scores don't look that impressive to me, with the caveat that we don't know what the stock clocks are on the 1080 and how far it can be pushed.

If it is stock ( presumably it is) - 10,102 is not bad for the Firestrike X bench. That is already up there with the top overclocked and watercooled Titan X cards.

 

[Jono8]

There seems to be a 7.5GB card with quite a high 3DMark score which has also appeared:

http://www.3dmark.com/3dm11/11061015

Edit!!

Scores look weird though!!

7.5gb? The 1070!?

 

[JediFragger]

Hoping it is 25-30% faster than the Ti!!

Looks like clock for clock it may be on a par though....

 

[Gregster]

Async is the new "big thing" but not sure how often/much game devs will use it in truth and especially if NVidia don't have it working on Pascal. I suppose it is the same for ROVs and CR for AMD and if they can't deal with it in Polaris, we won't see much of that being used either. Fingers crossed that both Pascal and Polaris have good working ACE, ROVs and Conservative Rasterization for some nice IQ and performance boosts in the future.

 

[CAT-THE-FIFTH]

Another one from AT forums:

https://www.chiphell.com/thread-1579388-1-1.html

The Maxwell various public version, Master Card, frequency + 500mhz, Pascal is the current public version of the GM204 in 1228 ~ 1291mhz, corresponding to 1734 ~ 1797mhz is the master in the actual frequency between 1316 ~ 1455mhz corresponding Pascal is 1823 ~ 1962mhz frequency is close to the upper limit now, consider the redundancy must add up

...Pascal pure infrastructure upgrade at most 10 to 15%, with a frequency ratio came out, certainly not so much Maxwell

Reference cards in China will cost 3999 CNY and custom cards 4500-5000 CNY at least.

So,that works out around £425 for the reference cards.

 

[Seanspeed]

If that is correct 20-30% faster then a 980Ti is not great.

How is that 'not great'?

It's not mind blowing, but it's also a 320mm^ card. It means we could easily see upwards of 60-70% faster for GP100, probably edging up to 80-100% at high resolutions using HBM2 seeing how well Fiji used HBM1 at 4k.

I'd say the results are pretty much exactly where I was expecting it to be. With driver maturity, I expect the 1080 could eek out a general advantage of 20-30% in most titles over a 980Ti.

Meaning that a 1070 should be able to match/beat a 980Ti, as many said couldn't ever happen.

I'm pretty happy with that. I know people want 'big Pascal' yesterday, but for everybody who doesn't own a 980Ti, this level of power at a lower market position is great.

 

[LoadsaMoney]

Nice, looks like its a 1080 for me then

 

[wunkley]

Another one from AT forums:

https://www.chiphell.com/thread-1579388-1-1.html



Reference cards in China will cost 3999 CNY and custom cards 4500-5000 CNY at least.

So,that works out around £425 for the reference cards.

.........Add on the usual UK general rip off tax and bumpage, and it'll be around 980Ti prices.

 

[Simon Belmont]

How is that 'not great'?

It's not mind blowing, but it's also a 320mm^ card. It means we could easily see upwards of 60-70% faster for GP100, probably edging up to 80-100% at high resolutions using HBM2 seeing how well Fiji used HBM1 at 4k.

I'd say the results are pretty much exactly where I was expecting it to be. With driver maturity, I expect the 1080 could eek out a general advantage of 20-30% in most titles over a 980Ti.

Meaning that a 1070 should be able to match/beat a 980Ti, as many said couldn't ever happen.

I'm pretty happy with that. I know people want 'big Pascal' yesterday, but for everybody who doesn't own a 980Ti, this level of power at a lower market position is great.

I guess I was expecting closer to 50%. Yes it's still a good showing, but these benches could be wrong?

 

[Seanspeed]

I guess I was expecting closer to 50%. Yes it's still a good showing, but these benches could be wrong?

Sure, there's a chance they are wrong and just trying to take advantage of the pre-announcement hype with some FUD.

It seems awfully plausible, though. Nothing they're claiming is outrageous or necessarily clickbait-worthy sounding, ya know? Well, the clock speed stuff is pretty crazy, but Maxwell's clock rates were also crazy when we first started discovering what they could do.

 

[flopper]

Even funnier fact: Async compute is not part of DX12, only compute and graphics queues are, which all Nvidia cards back to the Fermi support.

Best performance form Nvidia's compute queues comes from large work batches. GCN can handle smaller batch sizes better, but at the cost of increased overhead.

Don't buy into the AMD Pr machine. async compute is a an AMD marketing spin to solve a problem that Nvidia doesn't suffer from to the same extent. GCN has a terrible time exploiting all the CUs, Nvidia's architecture is much better at reaching closer to theoretical performance.

thats why amd crusch nvidia with Dx12 when a 390 owns the 980ti.
whats good with theory when in practice its not working?

 

[Jono8]

Uh oh, the crazies are coming.

 

[bfar]

.........Add on the usual UK general rip off tax and bumpage, and it'll be around 980Ti prices.

That would probably put the 1070 at around £3-350, ballpark 980ti performance level. Will be really interesting to see where Polaris fits in. Rumours about performance are pretty vague atm, but it sounds like it might put some pressure on the 1070 if the price is right.

 

[Gregster]

Uh oh, the crazies are coming.

Sadly this is more and more commonplace. Several people in this thread want to see fail from NVidia, where I would hope that any sane person would want to see both AMD and NVidia having beast cards. I don't understand the mentality of people like Flopper and Muzigaz.

Sitting at my computer today, if the 1080 is 20% faster than a 980Ti and comes with 8GB of GDDR5 (hell, even if it is 256 bit) and costs £420, I will be seriously thinking of grabbing one. The same for Polaris and will be considering that also. I don't really want to jump early and would rather hold out for the big cards but hell, I cleared well over £400 in OT last month, so what the hell