For me it was quite interesting. Does anybody know what the next gen mem that was stated for Navi chip is about? Is it succeeding HBM?
Not to mention Nvidia released massive GPUs on brand new node before, so its not like such thing never happened.
Nvidia's last attempt to release a massive GPU on a brand-new node (Fermi GF100) resulted in terrible delays - nearly a year - and serious yield issues that prevented them from producing a full SKU - every chip had to be cut down. Thermals were also through the roof. It took a bunch of respins to get it right (GF110). That's why they went with the smaller Kepler GK104 as their debut chip on 28nm in March 2012. Note that although they had Tesla chips for Oak Ridge a few months later, they didn't release GK110 Tesla to the public until November, and the first consumer GK110 Titan card wasn't released until early 2013. (And even these were cut-down with a few SMXs disabled.)
Well, if that's the case, Nvidia better not release a 14nm card better than the GTX 980 TI during the same time frame as Polaris or AMD is, yet again, behind.
Will upgrade irregardless for low wattage crossfire while I wait for Vega....
As I have said before, it's possible that AMD and Nvidia are going for completely different things in 2016. AMD, based on their public statements and demonstrations, seems to be going for laptop contracts ("console-grade performance in a thin-and-light laptop" for the small Polaris chip) and "affordable VR" (i.e. performance in what up until now has been the high end and will now be the mid-range, at lower prices). In contrast, what little we know about Nvidia's Pascal so far indicates that they are likely going to be prioritizing GP100 for a Tesla refresh, perhaps before any consumer-focused products hit. I doubt there will be any GP100 consumer cards in 2016, since it would make no sense to sell even a $999 Titan as long as someone is willing to buy the (low-yield) output for multiple thousands of dollars as Tesla cards. Only after that market is saturated will we see a Pascal Titan, probably in early 2017 - about the same time as AMD has Vega scheduled on today's roadmap. We probably will see GP104 later this year as the flagship consumer card, but I still think that won't be until September to October. I do think it will probably have a raw performance edge over AMD's 2016 products, though.
We probably will see GP104 later this year as the flagship consumer card, but I still think that won't be until September to October. I do think it will probably have a raw performance edge over AMD's 2016 products, though.
An up to date road map.
As I have said before, it's possible that AMD and Nvidia are going for completely different things in 2016. AMD, based on their public statements and demonstrations, seems to be going for laptop contracts ("console-grade performance in a thin-and-light laptop" for the small Polaris chip) and "affordable VR" (i.e. performance in what up until now has been the high end and will now be the mid-range, at lower prices). In contrast, what little we know about Nvidia's Pascal so far indicates that they are likely going to be prioritizing GP100 for a Tesla refresh, perhaps before any consumer-focused products hit. I doubt there will be any GP100 consumer cards in 2016, since it would make no sense to sell even a $999 Titan as long as someone is willing to buy the (low-yield) output for multiple thousands of dollars as Tesla cards. Only after that market is saturated will we see a Pascal Titan, probably in early 2017 - about the same time as AMD has Vega scheduled on today's roadmap. We probably will see GP104 later this year as the flagship consumer card, but I still think that won't be until September to October. I do think it will probably have a raw performance edge over AMD's 2016 products, though.
I think that we will see GP104 launch this May and available around June/July. I also think that it will be using GDDR5 (not 5x or HBM2).
You were not the audience for this. Gamers were not the audience. The audience is game developers @ GDC. AMD is simply trying to show the real developers what they've been doing over the past 6 months and what they plan to continue investing in.
True.
then dont fuxxing send it on internet for ffs.
Keep it in house.
AMD PR are engineer based which isnt doing them any favors.
Their Fiji launch with the drunk on drugs guy who is working for AMD another engineer screamed AN OVERCLOCKERS DREAM
with people like that in a company doing things that dont work to tell the customer that buys their stuff if people stop buying your stuff you dont have a job soon anymore.
and who wanna hire people who sank a company?
AMD are doing everything wrong except hardware which beats Nvidia ten times over.
polaris 11 120mm2 7790successor GTX960 like performanceOfc, we suspected as such for a long time now and it's logical. Only tiny and small chips need apply for the node transition.
That's what we have, Polaris 11 is a tiny chip, ~120mm2, Polaris 10 is a small chip, ~232mm2. These aren't even real "mid-range," that's normally around 300 to 380 mm2.
So Polaris 10 is more a low-midrange chip. If it's capable of beating the 390X, that's pretty impressive already.
Basically to get their 2.5x perf/w claims over 28nm GCN, they would have to match Fiji at 100W. So that's the ballpark.
So 2017 is the year to get a new videocard and monitor....
the damn box is far more in 2016 then it is 2017, why is everyone calling it 2017?
Polaris is coming mid year, that makes the last day of 2015 is the middle of 2016 (20 in 15, 16 in 16) that makes vega 2/3 in 16 1/3 in 17, or more likely a Q4 launch.
I've heard such rumors but I do not believe them. Nvidia likes to give its '4'-series chips a 256-bit memory bus, and GM204 is pretty much the limit of what you can achieve with that narrow of a bus on standard GDDR5. They will need GDDR5X to get adequate bandwidth to feed GP104 on a 256-bit bus. That's even if GP104 only has the same 3072-shader count as GM200, and I expect it to have 4096 shaders (a '4'-series chip needs to beat the previous '0'-series chip, especially after a node shrink).
polaris 11 120mm2 7790successor GTX960 like performance
polaris 10 232mm2 7870successor 390x/GTX980 like performance.It will fight against 1060.
Vega-early 2017 Hawaii successor.GP104 like performance(30% above TITANX).Around 6 months later than GP104.
It looks like AMD will have lead in low-end to mainstream.Untill NV launch their low-end and mainstream.NV after they release GP104 they will have no competetion with GP104 in 400-550USD market.
BIG question also is what AMD put againts BIG pascall.Vega x2?
Factor in that Pascal is 2x perf/watt of Maxwell while re-incorporating FP64 units, more complex FP32 units (capable of FP16) and your calculations are way off.
Maybe not for GP104, gaming focused and won't need those HPC features to weight it down.
GP100 for sure, going back to a compute root.
GK104 at launch was much faster than the big Fermi though. Could you imagine if the 680 is slower than the 580?
Exactly.
The Oak Ridge GPU's were late and cut down as well. Fermi was a mess. The faithful defended it though and bought it in droves. Go figure. :\
If GP104 uses GDDR5x then it will be competitive, if it doesn't then we're looking at the same sort of thing AMD are delivering.
In that case GP100 and Vega are where current GTX 980 Ti and FuryX users will be looking towards.