NVIDIA Maxwell Speculation Thread

 
I've only looked at the TechReport review, but it's almost embarrassing how this tiny thing that sips BW gets roughly the performance of R9 285.

From a customer point of view, I don't find this class of products very exciting, but if this is the segment with the highest revenues, Nvidia will be taking in monster margins on this one.
 
silent_guy said:
I've only looked at the TechReport review, but it's almost embarrassing how this tiny thing that sips BW gets roughly the performance of R9 285.

From a customer point of view, I don't find this class of products very exciting, but if this is the segment with the highest revenues, Nvidia will be taking in monster margins on this one.
Click to expand...

The overclocked one ... only a few review have the stock 960.. the Asus Strix they are using is refered as 1316mhz boost ( surely way more in reality, HC report 14216 mhz on their card )..... basically the card seems to sit between the 280 and 285. depending the clock speed.

This put them at roughly 8% over the GTX760.. ( hence why Nvidia was compare them with the GTX660 ( non TI )..

If someone wanted a confirmation than a 960TI based on GM204 was coming soon, i think we have it now.
 
Last edited:
What I find the most impressive with Maxwell is that each iteration comes with new features:
GM107 as gen1 / initial release with the great new arch efficiency
GM204 came with VXGI / DSR / MFAA accelerations
finally GM206 and tegra X1 bring full hardware HEVC decoder.
wondering what big daddy GM200 will bring on the table...
 
I'm probably the only one that doesn't see the same power efficiency in the GM206 compared to GM107 or GM204. Despite the disgrace of an only 2GB framebuffer for a mainstream solution, it might be a quite good offering overall but it most certainly lacks the "ooomph" effect its other Maxwell brothers have.
 
xpea said:
What I find the most impressive with Maxwell is that each iteration comes with new features:
GM107 as gen1 / initial release with the great new arch efficiency
GM204 came with VXGI / DSR / MFAA accelerations
finally GM206 and tegra X1 bring full hardware HEVC decoder.
wondering what big daddy GM200 will bring on the table...
Click to expand...

X1 also had the important addition of 2-wide SIMD fp16 FMA. I wonder if GM206 shares this? (Reviewers, what does CUDA deviceQuery say about the capability level of GM206? GM107 is sm_50, GM204 is sm_52, X1 is sm_53.)
 
Ailuros said:
Τοuche, but I said GM204 and the 970 isn't exactly the end station. Now riddle me this: considering the 970 has a power target of 145W, if I'd take hypothetically a 10 SMM GM204 at comparable clocks to the 970, what would you think the power target would look like on that one?
Click to expand...
Contrary to the GTX970 and 980, the typical power consumption of the GTX960 seems to be quite a bit lower than the TDP.
 
Ailuros said:
I'm probably the only one that doesn't see the same power efficiency in the GM206 compared to GM107 or GM204. Despite the disgrace of an only 2GB framebuffer for a mainstream solution, it might be a quite good offering overall but it most certainly lacks the "ooomph" effect its other Maxwell brothers have.
Click to expand...

I fully agree with you. Judging by TechPowerUp numbers this thing has double the TDP from full GM107 and is barely more than 50% faster than that part at 1080p! o_O

Compared to Kepler it's still impressive BUT it really needs to be clocked up sky high to really show competitive performance. GM107 and GM204 already reached high clocks but they did not need to be clocked as high as GM206 to be considered impressive.

Unfortunately this gives me this feeling that GM200 will follow the same route, except it will not clock as sky high and therefore it will not be considered impressive at all...

GM107 and GM204 might grow to be considered as the sweet spots for the Maxwell architecture...
 
Last edited:
silent_guy said:
Contrary to the GTX970 and 980, the typical power consumption of the GTX960 seems to be quite a bit lower than the TDP.
Click to expand...

According to techpowerup that's not true. They have the average power consumption at 108W for reference and 111W for EVGA OC part. Not very far away from official TDP of 120W...
 
This only has 27% more bandwidth than the GTX 750 Ti; in that sense it is still quite impressive! What I find more interesting is the die size. Effectively everything is halved from GM204 which makes it very easy to figure out what the die size of GM200 could be if it was exactly +50%:

GM204: 398mm2
GM206: 227mm2 -> +171mm2
GM200: 398 + 171 = 569mm2 (for 384-bit GDDR5 / 6xGPC / 3072 ALUs).

*IF* it is true that GM200 is really +-620mm2 then it might be possible for NVIDIA to have 5 SMMs per GPC ala GM107 for a total of 240 TMUs & 3840 ALUs. They already had 5 SMM/GPC on GM107 so it's definitely possible and could be quite attractive at high resolutions.
 
Picao84 said:
According to techpowerup that's not true. They have the average power consumption at 108W for reference and 111W for EVGA OC part. Not very far away from official TDP of 120W...
Click to expand...
On the other hand, if you compare to the official 165W of the GTX980... ;)

Not saying that the GTX960 sometimes doesn't come close to 120W, but it doesn't seem to go over it in egregious ways.

E.g. in the TPU numbers, the peak consumption is almost exactly 120W, where it's 181W and 184W for GTX970 and GTX980 resp. (stock clocks.)
 
Picao84 said:
According to techpowerup that's not true. They have the average power consumption at 108W for reference and 111W for EVGA OC part. Not very far away from official TDP of 120W...
Click to expand...

Is 120W really the "TDP"? Not that it really matters. Have a look at the "torture" graph: http://www.tomshardware.de/nvidia-geforce-gtx-960-maxwell-test,testberichte-241727-17.html


Granted these are done with a power virus, but I fail to see any "sweet spot" in terms of power consumption. At default clocks it stays within the 120W range. But once you let "overclocking dream" come into play it reaches nearly GTX980 level in that graph.
 
Arun said:
This only has 27% more bandwidth than the GTX 750 Ti; in that sense it is still quite impressive! What I find more interesting is the die size. Effectively everything is halved from GM204 which makes it very easy to figure out what the die size of GM200 could be if it was exactly +50%:

GM204: 398mm2
GM206: 227mm2 -> +171mm2
GM200: 398 + 171 = 569mm2 (for 384-bit GDDR5 / 6xGPC / 3072 ALUs).

*IF* it is true that GM200 is really +-620mm2 then it might be possible for NVIDIA to have 5 SMMs per GPC ala GM107 for a total of 240 TMUs & 3840 ALUs. They already had 5 SMM/GPC on GM107 so it's definitely possible and could be quite attractive at high resolutions.
Click to expand...

You are forgetting the die space *REQUIRED* for the additional FP64 units.
 
You must log in or register to reply here.

Similar threads

T
GPU Stability testing for Nvidia cards - PowerSweep
Replies
0
Views
448
T2098
T
D
Speculation and Rumors: Nvidia Blackwell ...
45 46 47
Replies
928
Views
103K
Kaotik
Kaotik
Rys
  • Sticky
Hello again
3 4 5
Replies
91
Views
12K
homerdog
homerdog
M
AMD made a mistake letting NVIDIA drive Ray Tracing APIs
5 6 7
Replies
122
Views
20K
MfA
M
xpea
Nvidia Hopper Speculation, Rumours and Discussion
7 8 9
Replies
177
Views
38K
Deleted member 2197
D
Back
Top