GM200 was a one time deal. A gaming focused chip the size of that isn't happening on this node, and don't expect one any time soon. Maybe GP102 will show up at 450 mm2ish but expect that to be $899+.
Half the size of GM200, but on the new node.
I think 35% more performance is pretty much expected from a 316 sq mm Pascal chip with clocks close to 1.6 Ghz( 35% faster than GTX 980 Ti) and architectural improvements which should increase perf/cuda core. I think the big die Geforce won't turn up any time before Q2 2017 since Tesla P100 itself is going to widely available from all server vendors only in Q1 2017.
GM200 was a one time deal. A gaming focused chip the size of that isn't happening on this node, and don't expect one any time soon. Maybe GP102 will show up at 450 mm2ish but expect that to be $899+.
Is this card expected to ship with 1.6 ghz clocks ? That would be crazy, then we'll see 1.8-2ghz overclocks ? I haven't read the whole thread.
If that is the case then it changes things. Right now I was getting the impression it would be like last time where the new card would match the old overclocked card and you'd have to overclock it again to get another 20% performance.
1.6 ghz is nuts. I can only just get that clock rate stable on one of my cards and it takes a lot of voltage.
I doubt this. Just will take enough time passing. They'll be stuck on this node for who knows how long. In a year they'll be only too happy to drop a $1000+ Titan card again. We'll definitely see an at least 550mm2 gaming Pascal chip at some point. That is the beast I want for single GPU 1440p.
Given that the P100 was announced as having a 1328 MHz core clock and a boost of ~1.5 GHz
I don't understand this thread or the hate against pasccal at all.
Lets say the GTX 1080 is a 20% increase over the GTX 980 in performance, but greatly increased perf/watt.
Isn't that what NBvidia users wanted? Isn't that why GTX 970 owners picked GTX 970's over R9 290s? Perf/Watt was important right?
Now all of a sudden, the most massive jump in perf/watt we've seen in awhile, and people aren't happy....
Weird....
Perf/watt was more important in allowing NVIDIA win the gaming notebook market than it was in the desktop GPU market.
GM200 was a one time deal. A gaming focused chip the size of that isn't happening on this node, and don't expect one any time soon. Maybe GP102 will show up at 450 mm2ish but expect that to be $899+.
I don't think we'll be seeing Pascal go much further above Maxwell. NV is just going really aggressive with the frequencies on the P100, and it shows in the TDP compared to its predecessors.
With a specified TDP of 300W. Pushing those clock higher would most likely make it actually sip that amount of power. We're talking Uber 290X levels.
Do you see Nvidia ditching their power efficiency mantra? Because I don't.
I don't think we'll see it very soon, but I'd be pretty surprised if there isn't a big die Pascal gaming card eventually.
lol yeah nice logic we will sell fewer but will cost more...did you just seriously said that? what company that wants to be profitable did that
also lol yeah 99% of people didnt bought anything at all the few 1% was the one that bought 970/980/390/390x......
#logic 2016
I think 35% more performance is pretty much expected from a 316 sq mm Pascal chip with clocks close to 1.6 Ghz( 35% faster than GTX 980 Ti) and architectural improvements which should increase perf/cuda core. I think the big die Geforce won't turn up any time before Q2 2017 since Tesla P100 itself is going to widely available from all server vendors only in Q1 2017.
I don't know, it doesn't sound likely. Nvidia's just going to throw away all their GP100 chips that can't pass Tesla binning? I expect at the least we'll see a heavily cut down GP100 gaming card like Nvidia did with the 780 and GK110. I don't think we'll see it very soon, but I'd be pretty surprised if there isn't a big die Pascal gaming card eventually.
If Vega is 4096sp like rumored, we're gonna see a bigger nVidia chip ASAP. Especially if it improves perf/W like was claimed on the road map slide.
If Vega is 4096sp like rumored, we're gonna see a bigger nVidia chip ASAP. Especially if it improves perf/W like was claimed on the road map slide.
tell me a company that actually had profits by selling fewer parts but with more cost per unit...name ONE(and survived to tell the story)Yes I did just say that. Are you seriously arguing that no company has ever gone for high margins over high volume and been profitable? Because if so, then this line of discussion has officially become too dumb for me.
Yep it was indeed the 1% who bought 970/980/390/390x. You don't understand that video cards, has always and will always be niche products, and the vast majority of the worlds population (the 99%) will never buy a GPU, regardless of price.
Indeed
I think Polaris 10 is currently rumoured at 2560SP, so Vega would be a 60% increase over this.
Speculation on the full Polaris 10 chip seems to put it around 390X/Fury level performance, whereas speculation on GP104 seems to be roughly 980 Ti +20-30%. If this holds true then GP104 would be roughly 50% faster than Polaris 10.
So if Vega 10 is 60% faster than Polaris 10, then it would only be about 5-10% faster than GP104, so Nvidia might be able to get by for at least a little while with just GP104 (maybe by releasing some overclocked variants).
You do know that you can't simply count SP's and figure out performance? Especially since they are showing Vega being a major improvement in efficiency over Polaris. Especially since we are comparing rumors for a not released product to a further improved not released product.
That would be so badass if AMD put the pressure on and Nvidia couldn't get away with their $550.00 mid range nonsense. They would be forced to release an actual high end card for a regular high end price of $5-600. That would be amazing.
their own roadmap places polaris 10 at the level of the fury line and 11 for the restLol you're about to be disappointed. I feel like Polaris is entry level/mid-high end. Nvidia is going to win the enthusiast market again IMO. Vega will then spoil that party and come after the 1080Ti does.
You're right usually the performance increase is smaller than the increase in functional units, since you never get perfect scaling, so I guess it might be more like Vega 10 = GP104 in that case.
Where have you seen AMD say that Vega is a major improvement in efficiency over Polaris by the way, I must have missed that?
This road map. Perf/watt over time.
raja said 2.5x and he is the head of rtg and also an engieneer...his words have more valueThe 2.5x already seems wrong since the CEO wont commit to it with investors. And there is no numbering on the left. So you dont even know if the scaling is supposed to be linear either.