That would be kinda disappointing since 980 is barely 15% faster than 780Ti. That would give a 50% increase over GK110 tops, much less than what was achieved with GM107 and GM204 (between 80%-100% increase). I know that they did not have a lot of die space to play with, while those chips had, but still density increased a bit with Maxwell. Well, still an Engineering Sample and we do not know the power numbers either...
Actually, scrap that. There are power numbers and they show the cut down GM200 having a much worse efficiency than GM204... Hmm...
Last edited:
If true, then I might get excited for AMD again, possibly...At least if they get their driver development back under control again.
Same here, if those benchmarks are true then Bermuda looks very tasty indeed (or more likely a "non-X" version of it).
But more or less on paar with the difference between the GM204 and the GK104 no ? ( 50-60% depending the review ).
If thoses numbers are not fake, thoses are early ES version with early drivers... surely not the best who could be achieved. ( so let say 60% more minimum who is effectively in paar with Kepler > Maxwell performance difference )
980 have 2048SP, 680 have 1536SP, i can imagine this can be tied to how much they could put as SP in more over the GK110.
Last edited:
Not really, GM204 vs GK104 was around 70%, in some cases 80% increase really. GK110 is around 50% faster than GK104 as it stands.
Where exactly? TPU shows GTX 980 54 to 63% faster than GTX 770 (900p to 2160p), ComputerBase 45 to 58% (1080p, 1600p, with and without SSAA)
(and those 2 are the only ones i can remember offering nice overall numbers)
edit:
And GK110 vs GK104, TPU says 39-48% and Computerbase 35-47%
In addition, Hardware canuck, 63% @ 1440p ( only resolution tested by them and lately i had allways find their " % faster "way higher of what i have seen elsewhere ). should be lower at 1080p, ( maxwell benefit of better memory efficiency compared to GK104, so the difference increase with bigger resolution )..
Again, i dont trust the numbers leaked by Chiphell, but at 60-65%, it seems at least on paar with what we have seen so far.
Again, i dont trust the numbers leaked by Chiphell, but at 60-65%, it seems at least on paar with what we have seen so far.
Last edited:
As lanek said it was 63% on Hardware Canucks. Give it 65% then for hairsplitting. In some new games like Total War: Rome 2 GTX980 is a whooping 84% faster and in Thief it can be more than 100% faster.
Last edited:
From there, I read 50% tops, hence my comment.
I don't see how it's "hairpslitting" when all reported sites are under what you're claiming
IF thoses numbers are true, early ES sample, early ES drivers ... this 50% could well end at 60-65% ( this what i was saying ) over the GM204...
But there's 2 factors: maxwell GM 204 have really high turbo boost clock ( 1221-1261mhz depending your luck ) , and 33% more shaders than GK104. GM204 gains a lot of performance over the GK104 due to the efficiency in memory bound situation. Im not sure this will all work so well together for GM200 over GK110.
- GK110 was less memory bound that GK104 with hist 384bit bus / 3GB memory setup.
- Its possible that GM200 could not get at the same time 33% more shader and 20-25% more clock speed.
- even 50-55% look credible to me if on the same node ( and again i dont trust this becnhmark leak on chiphell ).
Last edited:
Ok, I admit I was a little bit carried out with the huge energy efficiency gains on the same node. But still, I would not be surprised to see the gap wide a bit in next generation games, as could already be seen from Thief and Total War: Rome 2.
Would any of those 'some new games' be Nvidia Gameworks-"optimized", by any chance...?
Would any of those 'some new games' be Nvidia Gameworks-"optimized", by any chance...?
That argument does not make much sense since:
1) We are comparing Kepler and Maxwell, architectures from nVIDIA, so in theory they would both benefit from Gameworks?
2) Thief is actually an AMD Gaming Evolved title?
But good try at trolling.
Someone please explain to me how and why someone would have:
- AMD's next flagship
- Nvidia's next flagship
- AMD's next 2nd in line card
- Nvidia's next 2nd in line card
- Drivers for all the above
And why that person would be the only one leaking anything. Honestly the chances sound astronomically low.
- AMD's next flagship
- Nvidia's next flagship
- AMD's next 2nd in line card
- Nvidia's next 2nd in line card
- Drivers for all the above
And why that person would be the only one leaking anything. Honestly the chances sound astronomically low.
This someone would have to have a very generous source (or be this source) within one of the main add-in board partners. Unlikely, but possible.
You know, when you put it that way it does sound incredibly unlikely.
I suppose people just really want this (most likely BS) rumor to be true, since we haven't had any flagship cards introduced for over a year now. Wishful thinking can be a really powerful force. Also, the performance figures reported aren't as outlandish as they sometimes are.
Oh gods, don't be so sensitive. It was just a joke, alright?
GM107 and GM204 are far larger than GK107 and GK104 respectively (25% and 35%). In comparison, GM200 is pretty much the same size as GK110 (maybe 2% bigger) so of course the gains will be limited.
That seems to be quite fishy..I dont see why GM200 should be any less power efficient than GM204..unless the clocks/voltages are much higher or its a bad sample (all assuming this "leak" is real of course...)
Given that the architecture is pretty much the same..I dont see how drivers would make much of a difference.
Why not? GM200 will employ similar architectural changes as GM204 did..so why will it not be as efficient with memory bandwidth?
Again..dont understand your logic here. It should be the opposite. GK110 has almost double the units of GK104 but only 50% more bandwidth. For GM200, it should have 50% more units and at the same time 50% more bandwidth.
I was just about the ask the same thing. If it was a leak comparing AMD to AMD..or NV to NV..sure..would have sounded more credible. Presumably these chips would still be in internal testing only so how could someone outside of either NV or AMD get all of the above mentioned GPUs
Blazkowicz
Legend
Does GM200 have fast DP or slow DP?
