Hey folks, looks like the leak period of the Maxwell Geforce 800 Series has finally started because specifications and benchmarks of the GTX 870 have leaked out. Even though the leak comes from a source (Coolaler.com) that has leaked authentic information in the past, I would still recommend just the tiniest pinch of salt just in case.
Geforce GTX 870 Specs and Performance Leaked – 1664 CUDA Cores, 4GB GDDR5, 256Bit BUS and 4GB GDDR5
The entire leak consists of a couple of screen shots and at close inspection everything appears to be legit with not a hair out of place. As expected, the GPU core, which is GM204, is not properly recognized and the name detected shows Nvidia’s internal nomenclature. I ran the name “Nvidia 17D-20″ in a few databases but it didn’t net me anything else. This probably the engineering sample of the 870 and the code allocated to the GPU is 13C2. I am guessing that the 13 Stands for 13 SMMs and C2 stands for Consumer Chip no. 2. I am assuming the actual flagship will be named either 16C2 or 17C2 depending on the amount of Streaming Multiprocessor Maxwells (SMMs). A lot of other specs are not properly recognized and this is to be expected since there is no way that the GPU-Z tool has been updated to support Maxwell GPUs yet. The screen shots not only show the specifications but also reveal the 3D Mark Extreme and Performance scores as well, which is great because it allows us to rank the GPU among Nvidia’s existing cards. Since they are using an i7-4820K we can assume no CPU Bottleneck. Without any further ado, here are the complete specs of the high end Geforce GTX 870:
WCCFTech
|
GeForce GTX 780
|
GeForce GTX 780 Ti
|
Geforce GTX 870
|
Geforce GTX 880
|
Codename
|
GK110
|
GK110
|
GM204
|
GM204
|
Process
|
28nm
|
28nm
|
28nm
|
28nm
|
GPU Core
|
Kepler
|
Kepler
|
Maxwell
|
Maxwell
|
SMX/SMM
|
12
|
14
|
13
|
17 (?)
|
CUDA Cores
|
2304
|
2880
|
1644
|
2176 (?)
|
ROPS
|
48
|
48
|
32
|
32
|
TMUs
|
192
|
240
|
138
|
TBC
|
Core Clock
|
863 MHz
|
876 MHz
|
1051 MHz
|
TBC
|
Boost Clock
|
900 MHz
|
928 MHz
|
1178 MHz
|
TBC
|
Memory
|
3 GB GDDR5
|
3 GB GDDR5
|
4 GB GDDR5
|
4 GB/8GB GDDR5
|
Memory Clock
|
6.0 GB/s
|
7.0 GB/s
|
7.0 GB/s
|
7.0 GB/s
|
Memory Bus
|
384-Bit
|
384-Bit
|
256-Bit
|
256-Bit
|
Memory Bandwidth
|
288.6 GB/s
|
336.0 GB/s
|
224.5 GB/s
|
224.5 GB/s
|
Texture Fill Rate GT/s
|
166
|
210
|
145.0
|
TBC
|
TDP
|
250W
|
250W
|
~190W
|
~190W
|
Now there is some pretty interesting information about this card. The 3d Mark performance benchmark of the GTX 870 is 11,919 and the extreme score is 4625. Now the GTX 780 scores around that point (4625) so any fears of the Maxwell series actually being weaker than its predecessor should be allayed. If the GTX 870 can score around the GTX 780 then the 880 should score more than the 780 Ti. The specs listed show 1664 CUDA Cores, which means that we are looking at a grand total of 13 SMMs (128 Cores per SMM, 1664/128 = 13). Following the previous generation trend, the Maxwell flagship should have 17 SMMs with a CUDA Core count of 2170 respectively. Basically, you should expect an SMM value of 14 – 17. I Will be really surprised if it goes any higher. The Pixel Fillrate appears to be 33.6 GPixels per second and the Texture Fillrate is 145 GPixels per second. In comparison the values of the 770 were 33.5 and 134 respectively. The bandwidth is 224 GB/s which is to be expected since the card has a 256 Bit Interface. Another interesting thing is that the memory used belongs to Samsung and not Hynix. So there you go folks, the first authentic looking leak regarding the Geforce GTX 870 Maxwell.
If you want to contact RDT please feel free to call between the hours of 9am-8pm AEST. deon@robertsondt.com Ph: 0426279566 or visit https://www.facebook.com/rdtaus
No comments:
Post a Comment