ASUS GeForce GTX 670 DirectCU Mini review -
The Technology and Specs
Reference technology and specs
We'll first look at the reference (original design) based specs and architecture. The GeForce GTX 670 is based on the a Kepler GPU architecture, the very same 28nm GK104 GPU which is used on the GeForce GTX 680. The GeForce GTX 670 boasts 1344 CUDA (shader) cores whereas the GeForce GTX 680 has 1536 CUDA (shader) cores. That's 192 Shader cores less, and that's precisely one CUDA core clusters (SM) less out of the eight available. The product is obviously PCI-Express 3.0 ready and has a TDP of around 170 Watt (with a typical draw of 150~160W). But let me first show you GK104 die:
NVIDIA GK104 Kepler architecture GPU, you can see the eight SM (CUDA/shader core) clusters, one of these has been deactivated for the GTX 670.
An immediate difference to the GPU core versus the shader processor domain is that both will be clocked at 1:1, meaning both the core and shader domain clock in at 915 MHz. The boost clock for the reference GTX 670 cards is set at 980 MHz though that can vary a bit per card and available power envelope (topping 1 GHz is common). As far as the memory specs of the GK104 Kepler GPU are concerned, the boards will feature a 256-bit memory bus connected to 2 GB of GDDR5 video buffer memory. On the memory controller side of things you'll see very significant improvements as the reference memory clock is set at 6 GHz / Gbps. This boils down to to a memory bandwidth of 192 GB/s on that 256-bit memory bus. These graphics adapters are of course DirectX 11.1 ready. With Windows 8, 7 and Vista also being DX11 ready all we need are more new games to take advantage of DirectCompute, multi-threading, hardware tessellation and the latest shader 5.0 extensions.
For your reference here's a quick overview of some past generation high-end GeForce cards opposed to the new Kepler based GeForce GTX 680.
670 DCU Mini
|Stream (Shader) Processors||480||512||1344||1344||1536||3072|
|Core Clock (MHz)||700||772||915||928||1006||915|
|Shader Clock (MHz)||1400||1544||-||-||-||-|
|Boost clock (Mhz)||-||-||980||1033||1058||1019|
|Memory Clock (effective MHz)||3700||4000||6008||6008||6008||6008|
For Kepler, NVIDIA kept their memory controllers GDDR5 compatible. Memory wise NVIDIA has nice large memory volumes due to their architecture, we pass 2 GB as standard these days for most of NVIDIA's series 600 graphics cards in the high range spectrum. The hardware engineers of NVIDIA reworked the memory subs system quite a bit, enabling much higher memory clock frequency speeds compared to previous generation GeForce GPUs. The result is this memory speeds up-to 6 Gbps. Each memory partition utilizes one memory controller on the respective GPU, which will get 256/512 MB of memory tied to it.
- The GTX 580 has six memory controllers (6x256MB) = 1536 MB of GDDR5 memory
- The GTX 670 has four memory controllers (4x512MB) = 2048 MB of GDDR5 memory
- The GTX 680 has four memory controllers (4x512MB) = 2048 MB of GDDR5 memory
As mentioned in the introduction, a 4 GB version would be very possible as well. It all depends on the board partners.
The graphics architecture that is Kepler
As you can understand, the massive memory partitions, bus-width and combination of GDDR5 memory (quad data rate) allow the GPU to work with a very high framebuffer bandwidth (effective). Let's again put most of the data in a chart to get an idea and better overview of changes:
|Graphics card||GeForce GTX 580||GeForce GTX
670 DCU Moni
|GeForce GTX 680||GeForce GTX 690|
|Streaming Multiprocessors (SM)||16||7||7||8||16|
|Graphics Clock (Core)||772 MHz||915 / 980MHz||928/ 1033 MHz||1006/1058MHz||915/1019MHz|
|Shader Processor Clock||1544 MHz||915 / 980MHz||/ 1033 MHz||1006/1058MHz||915/1019MHz|
|Memory Clock / Data rate MHz||1000 / 4000||1502 / 6008 MHz||1502 / 6008 MHz||1502 / 6008 MHz||1502 / 6008 MHz|
|Graphics memory||1536 MB||2048 MB||2048 MB||2048 MB||4096 MB|
|Memory bandwidth||192 GB/s||192 GB/s||192 GB/s||192 GB/s||192 GB/s|
|Power connectors||1x6-pin PEG, 1x8-pin PEG||2x6-pin PEG||2x6-pin PEG||2x6-pin PEG||2x8-pin PEG|
|Max board power (TDP)||244 Watts||170 Watts||180 Watts||170 Watts||300 Watts|
|Recommended Power supply||600 Watts||550 Watts||550 Watts||550 Watts||750 Watts|
|GPU Thermal Threshold||97 degrees C||98 degrees C||98 degrees C||98 degrees C||98 degrees C|
So we talked about the core clocks, specifications and memory partitions. Obviously there's a lot more to talk through the GPU architecture for example. To understand a graphics processor you simply need to break it down into pieces to better understand it. Let's first look at the raw data that most of you can understand and grasp. This bit will be about the Kepler architecture, if you're not interested in g33k talk by all means please browse to the next page.
So above we see the GK104 block diagram that entails the Kepler architecture. Let's break it down into bits and pieces. A fully operating GK104 will have:
- 1536 CUDA processors (Shader cores)
- 192 CUDA core clusters (SM).
- 8 geometry units
- 4 raster Units
- 128 Texture Units
- 32 ROP engines
- 256-bit GDDR5 memory bus
- DirectX 11.1
Above thus a fully operating GK104 as used on the GTX 680. The GTX 670 uses the same chip, but has one SM (CUDA / Shader core cluster) disabled. So the more important thing to focus on are the SM (block of shader processors) clusters (or SMX as NVIDIA likes to call it for the GTX 680, which has 192 Shader processors. That's radically different from Fermi, the GeForce GTX 580 for example had 32 shader processors per SM cluster. 1536 : 192 = 8 Shader clusters (SMs). Let's blow up one such cluster:
Above the block diagram for a single Shader processor cluster, aka SM or SMX as NVIDIA now calls it. The new SMX has quite a bit more bite in terms of shader, texture and geometry processing. 192 CUDA cores, that's six times the number of cores per SM opposed to Fermi. Now, at the end of the pipeline we run into the ROP (Raster Operation) engine and the GTX 680 again has 32 engines for features like pixel blending and AA. There's a total of 128 texture filtering units available for the GeForce GTX 680. The math is simple here, each SM has 16 texture units tied to it.
- GeForce GTX 580 has 16 SMs X 4 Texture units = 64
- GeForce GTX 670 has 7 SMs X 16 Texture units = 112
- GeForce GTX 680 has 8 SMs X 16 Texture units = 128
Above the GK104 host interface - The Gigathread engine, four GPCs, four memory controllers, the ROP partitions, a 768 KB L2 cache. Each GPC has eight polymorph engines - ROP partitions are nearby to the L2 cache, Each shader cluster then is tied to L1 and a shared L2 cache. Shading performance is going be increased quite bit, geometry performance will get a nice boost as well. NVIDIA is using 64KB Shared Memory/L1 per SMX please note that they have a 16/48 48/16 ratio here for graphics/compute, as before with Fermi. For L2, 128KB per 64-bit memory controller. So that adds up to 512KB L2
In regards to architectural changes, on top of the pipeline NVIDIA has now added new Polymorph 2.0 (world space processing) engines and raster (screen space processing) engines, they act like a mini CPU really.
In this article we review ASUS GeForce GTX 760 DirectCU MINI edition, this particular model comes with a dual-slot DirectCU Mini cooling solution that will fit small form factor based motherboards like Mini ITX up-to 17cm. ASUS has re-engineered the DirectCU cooler to fit small form factor cases. While shorter, it introduces a copper vapor chamber placed directly on top of the GPU for faster heat spreading and dispersal with lower temperatures than reference GTX 760.
ASUS GeForce GTX 760 DirectCU II OC review
In this article we review ASUS GeForce GTX 760 DirectCU II OC edition, this particular model comes with a dual-slot DirectCU II cooing solution. That boils down to a silent product versus and more than excellent cooling performance. Inno3D offers the card factory slightly overclocked at some pretty impressive clock frequencies.
ASUS GeForce GTX 780 DirectCU II OC review
We test and review the ASUS GeForce GTX 780 DirectCU II review edition. The graphics card comes with a factory overclock and an updated DirectCU II cooler that has CoolTech fans. That would be two silent 90mm fans.
ASUS GeForce GTX 670 DirectCU Mini review
In this article we review the ASUS GeForce GTX 670 DirectCU Mini edition, a compact performance graphics card designed primarily for small form factor PCs with mini ITX motherboards. The dual-slot card measures just 17cm and features the NVIDIA GTX 670 GPU. ASUS has re-engineered the DirectCU cooler to fit small form factor cases. While shorter, it introduces a copper vapor chamber placed directly on top of the GPU for faster heat spreading and dispersal with 20% lower temperatures than reference GTX 670.