It’s been a while since NVIDIA launched the first GPU and Graphic cards using that GPU belonging to the very awaited and anticipated, NVIDIA’s Maxwell architecture. So now we now get to know the details of the first high end GPU that would replace the high end Kepler counterparts.
GM107, the first GPU based on NVIDIA most efficient GPU micro-architecture that is the Maxwell, featured on the budget oriented cards like GTX 750 and GTX 750 Ti which really amazes the users with its performance and makes us hard to believe that many OEM cards don’t even require a 6 pin PCI-E power connector and the typical board consumption ducks right below the 75W ceiling provided by the PCI-E x16 slot itself. But enthusiasts were eagerly waiting for the 20nm Maxwell GPUs with the blings that Maxwell promises like UVM (Unified Virtual Memory) and Integrated CPU. Sources have it that 20nm GM204, a GPU expected to power the GTX 880, would be replacing the GK104 chip (first time featuring on GTX 680) but it would be faster than the GK110. It might seem to amuse many readers, but few readers who are acquainted with Fermi series cards would not be surprised since GK104 surpassed GF104 but was faster than even GF110. Likewise, the scenario remains same. The chip is DirectX 12 ready and sports Stream Multiprocessor Maxwell (SMM) SIMD design same as GTX 750 Ti, with just 3 more SMMs compared to 750 Ti spread across the GPCs (Graphics Processing Clusters) and loathed with large L2 cache. Even more is said GM204 and GM210 would be integrated with “Project Denver” ARM CPU architecture based 64-bit core processor.
The GTX 880 (based on fully unlocked GM204) is rumored to have the following specs:
- GM204 GPU based on 20nm fabrication processes with 3200 CUDA cores and 200 TMUs and 32 ROPs
- 4GB GDDR5 memory with 238 GB/s bandwidth across 256 bit wide bus @ 7400 MHz
- 7.9 Billion transistors with 5.7 TeraFlops of single precision FP32 performance and 230W of typical board consumption.
|Card name||GTX 880||GTX 680||GTX 780 Ti|
|# of CUDA cores||3200||1536||2302|
|Transistor count||7.9 Billion||3.5 Billion||7.1 Billion|
|GPU clock base/boost||900/950 MHz||1006 MHz||870/933 MHz|
|Effective memory clock||7400 MHz||6000 MHz||7100 MHz|
|Base Memory buffer||4GB GDDR5||2GB GDDR5||3GB GDDR5|
|Memory width||256 bit||256 bit||384 bit|
|Max effective bandwidth||238 GB/s||192 GB/s||288 GB/s|
|FP32 precision||5.7 TFlops||3 TFlops||4.8 TFlops|