GeForce GTX 680 Overview
Here we are, NVIDIA has officialy launched the GeForce GTX 680 based on the Kepler GK104 GPU. This card packs a bunch of new features (and some of them can be tweaked with some line of code!) such as GPU Boost (dynamic clocking depending on GPU load), adaptive VSYNC, bindless textures, new power management functions (Power Target), TXAA (temporal anti-aliasing), NVENC (encoding engine) and optimized power consumption. The architecture has been changed (no double speed anymore for shader processors, this is the new SMX unit) and priced at USD $499, this card targets AMD Radeon HD 7900 series.
Here are the main features of the GeForce GTX 680:
- GPU: Kepler GK104 @ 1006MHz (base clock) or 1058MHz (boost clock), 28nm TSMC
- Shader processors (CUDA cores): 1536
- Streaming Multiprocessor-X (SMX): 8 (192SP per SMX)
- Memory: 2048MB GDDR5 @ 1058MHz real speed or 4232MHz effective, 256-bit
- Color ROP units: 32
- Texture units: 128
- TDP: 195W
- Power connectors: two 6-pin
- 3D APIs: OpenGL 4.2, Direct3D 11.1
- GPU computing: OpenCL 1.2, CUDA, PhysX and DirectCompute
- Bus interface: PCI-Express 3.0
- Price: USD $499
Kepler GK104 GPU architecture:
Adaptive VSYNC (vertical synchronization):
Bindless Textures:
TXAA: Temporal Anti-Aliasing
NVENC: dedicated encoding engine
GeForce GTX 680 Performances
when I played Bioshock with VSync enabled I had 30-40 fps, with VSync disabled it was more than 80 fps (on Radeon 4850), so I can’t get what the hell NV says about VSync
As I suspected… new arch is great for gaming but is also step back in GPGPU. Well, maybe not whole big step, but a little one. Nevertheless it’s step in the right direction. Lower temps and energy consumptions, great performance and visual quality. That’s what made NV famous.
I don’t think that it is step back in computing. I would like to see some CUDA benchmarks to verify that. NVIDIA invested to much in CUDA and it’s ecosystem to make a step back now. I think that low performance in OpenCL benchmarks is becaus earyl drivers.
well I guess for compute you must not consider it an improvement over GF100,GF110 it clearly is a successor for GF104 ,GF114
as long as its compute ability still enough for gaming (physx, directcompute/cuda effects), then it’s no big deal. just need more games to exploit this compute feature.
Reviewers don’t seem too happy with the opencl perfs, but if nvidia can keep a lower price than whatever amd reply is gonna be, this can still be a good choice for gaming.
Great card, ridiculous price.
I ain’t paying 500 for the successor of the GTX 560 Ti. Jesus.