Radeon R7 200:  Radeon R7 240 |  R7 250 |  R7 250X |  R7 260 |  R7 265 |  Radeon R9 200:  Radeon R9 270 |  R9 270X |  R9 280 |  R9 280X |  R9 290 |  R9 290X |  R9 295X |  R9 295X2
GeForce 700:  GeForce GTX 760  |  GTX 770  |  GTX 780 |  GTX 780 TI |  GeForce TITAN:  GeForce GTX TITAN |  GTX TITAN Black |  GTX TITAN Z |  GeForce 900:  GTX 970  |  GTX 980

Mar 16th, 2012

GeForce GTX 680 Benchmark Leaks Out, Leaves HD 7970 Behind

GeForce GTX 680 Benchmark Leaks Out, Features MultiDisplayOut
First real tests of GeForce GTX 680 were leaked. Card is much faster than Radeon HD 7970, leaving it behind in almost every test.

Long story short, most of the benchmarks prove that GTX 680 is indeed a good construction. According to previous leaks it was meant to be much faster than Radeon HD 7970. Tests performed by HKEPC confirm this rumor. GeForce GTX 680 was tested with LN2 cooling, with clocks up to 1006 MHz for core and 6008 MHz for memory. Default clock is capped with its own profiles, 1006 MHz is base clock, but card can change the level of gpu clock to 1058 MHz due to computing demand.

HKEPC have put GTX 680 through several tests, which include 3Dmark11 (Extreme and Performance Preset), Heaven Benchmark, Lost Planet 2, Call of Duty MW3, Batman Arkham City, Power Consumption and Temperature.

Benchmarks GTX 680 is 20% Faster than Radeon HD 7970 and 50% Faster than GTX 580

  GeForce GTX 680 Radeon HD 7970 GTX680 vs HD7970 GeForce GTX 580 GTX680 vs GTX580
3DMark 11 Extreme Overall 3165 2669 118% 2077 152%
3DMark11 Extreme GPU Score 2869 2424 118% 1905 150%
3DMark11 Combined Score 3421 2848 120% 2520 120%
3DMark 11 Performance Overall 9303 7713 120% 6578 141%
3DMark11 Performance GPU Score 9298 7406 125% 6027 154%
3DMark11 Performance Score 7655 7010 109% 6392 120%
Batman Arkham City [1920x1080 8xAA] 57 49 116% 42 136%
Call of Duty MW3 [1920x1080 8xAA] 74 74 100% 67 110%
Heaven Benchmark 46.5 41.5 112% 34.9 133%
Lost Planet 2 73.4 58.5 125% 54.6 134%
Temperature Full Load 83 86 96% 86 96%
Temperature Idle 33 40 82% 37 89%

Here are the official, and final specifications of GeForce GTX 680

  GeForce GTX 680 GeForce GTX 580 Radeon HD 7970
Core GK-104 GF-110 Tahiti XT
Architecture Kepler Fermi 2.0 GCN
Process 28nm 40nm 28nm
Transistors 3.54 billion 3.0 billion 4.31 billion
Die Size 294mm2 520mm2 365mm2
Engine Clock 1,006MHz 772MHz 925MHz
Shader Units 1,536 512 2,048
Texture Units 128 64 128
ROPs 32 32 32
Fillrate (GTexel/s) 128.8 52.4 115.6
Memory Type 256bit GDDR5 384bit GDDR5 384bit GDDR5
Memory Size 2GB 1.5GB 3GB
Memory Date Rate 6.0Gbps 4.0Gbps 5.5Gbps
Memory Bandwidth 192.26GB/s 192.384GB/s 264GB/s
PCI-Express 3 2 3
Maximum Wattage 195W 244W 250W
Power Connectors 2x 6-pin 6+8-pin 6+8-pin
DirectX 11 11 11.1
OpenGL 4.1 4.1 4.1
GPGPU CUDA, OpenCL CUDA, OpenCL OpenCL 1.2
Maximum Monitors Output 4 2 4(6 w/MST)

HKEPC test rig

Test Platform
CPU Intel Core i7 3960X (With Turbo Boost)
Motherboard ASUS Rampage 4 Extreme X79 MB
Memory GEIL Evo 2 DDR3-2200 2GB x 4 CL9
SSD / HDD Corsair Force GT 120GB
VGA

MSI GeForce GTX 580/ GeForce GTX 680 /AMD Radeon HD 7970

Power Supply Corsair AX1200 PowerSupply
OS Windows 7 Ultimate SP1 64Bit

GeForce GTX 680 GK104 GPU

A 28nm GK104 GPU is built with 1536 CUDA cores. Everything is packed onto 294mm2 die with almost 3.54 billion transistors. CUDA cores were trippled since GTX 580. GK104 is built with GigaThread Engine, which is placed between GPU and PCI-Express 3.0. GPU is divided into four graphics processing clusters (GPCs) which are connected to the same L2 cache and have all four have one Raster Engine. Every cluster has its two SMX, which is basically a streaming multiprocessor. SMX is equipped with PolyMorph Engine 2.0, which is accompanied by 192 CUDA cores. GK104 is also built with 8 Geometry Units, 32 Raster Operating Units and 128 Texture Units. Every Cluster has its own two SMXs, since every SMX has 192 cores, it makes 1536 CUDA cores in total.

GeForce GTX 680 – The Card

First thing is the clock, which reaches the GHz level on base settings. Card will automatically change the clock according to demand. GTX 680 has included profiles, where the Boost Clock is set to 1058 MHz (which is reference maximum clock). Card is built with 3.56 billion of transistors, which is half the billion more than its predecessor GTX 580. Fillrate was measured at 128.8 GTexels per second which is 146% more than GTX 680 and 11% more than Radeon HD 7970. Card has a similar bandwidth to GTX 580 of 192 GB/s. PCI-Express was updated to version 3.0.

NVIDIA 3D Vision Surround

GeForce GTX 680 has a support for operations on four monitors simultaneously. Since card has four outputs for displaying devices (HDMI, DisplayPort and two DVI-DL) it is possible to connect devices to all of them, and operate on multiple displays at once. This setting has a support of 3D Vision Surround technology. Every gamer that has heard of 3D Vision from NVIDIA will understand how interesting this option can be. Connecting three of four displays and play on them with stereoscopic 3D support may be quite a fun.

New Anti-Aliasing Algorithm TXAA

As we reported earlier, NVIDIA was prepaing a new anti-aliasing algoritm, which was supposed to increse image quality without decreasing framerate. It was comparable to 8xMSAA but much better. TXAA stands for super-efficient temporal anti-aliasing algorithm. It has two main levels TXAA1 and TXAA2. TXAA1 can be comparable to offer quality similar to 16 x MSAA, but performance penalty of 2xMSAA. TXAA2 is actually even better solution, there is no algorithm which would offer this quality of anti-aliasing. Performance penalty is measured to be at the level of 4xMSAA. There are few games which natively support this algorithm. To enable it we need to use NVIDIA’s Control Panel.

Adaptive VSync and GPU Boost

NVIDIA has presented a new VSync technology, which is mainly offered for gamers. Adaptive VSync is a good tool, when framerate is too slow or too fast. What new Adaptive VSync is a technology which will automatically remove the cap when framerate reaches critical 30 or less frames per second. Since low framerates offer noticeable lag, and high framerates offer page-tearing , Adaptive VSync is said to soften those transitions by making them more organic.