Here it is, NVIDIA's GeForce GTX 750 Ti, one of the first GPUs based on its next-generation "Maxwell" GPU architecture. First reports of the chip being based on "Maxwell" hit us by surprise as we presumed that a new micro-architecture is invariably pegged to NVIDIA's foundry partner, TSMC, to launch its next-generation 20 nanometer silicon fabrication process. The fact that TSMC's 20 nm node isn't in a position to ship out such complex ASICs as graphics processors right now and that NVIDIA wouldn't want its new architecture to in any way be hit by unforeseen issues arising out of the move to a new process may have led to the decision to use the existing 28 nanometer process.
The GeForce GTX 750 Ti and the GTX 750 (also launched today) are based on the 28 nm GM107 (GeForce-Maxwell 107) silicon. This 1.87 billion-transistor GPU features a basic component hierarchy that's not too different from that of "Kepler," but sees some under-the-hood changes in the design of its key parallel processing sub-unit, the streaming multiprocessor "Maxwell" (SMM). The GTX 750 Ti is positioned between the GTX 760 and GTX 660, which is a sizable gap to fill. NVIDIA is promising some huge performance-per-watt gains with "Maxwell." Given that the GM107 is based on the same 28 nm process as the GeForce Kepler series, there's only one way NVIDIA can deliver on its promise - by developing a better overall architecture.
MSI's GeForce GTX 750 Ti Gaming comes with the company's mighty TwinFrozr dual-fan cooler, which provides plenty of cooling performance for the GTX 750 Ti GPU. The card is also overclocked to 1085 MHz out of the box, which should provide a quick and easy performance boost.
MSI states that their GTX 750 Ti Gaming will retail at $200, which is a hefty $50 (or 33%) price increase.
MSI GTX 750 Ti Market Segment Analysis
GeForce GTX 650 Ti
Radeon HD 7790
Radeon HD 7850
GeForce GTX 750
Radeon R7 260X
GTX 650 Ti Boost
GeForce GTX 750 Ti
MSI GTX 750 Ti Gaming
Radeon R7 265
GeForce GTX 660
Radeon HD 7870
Radeon R9 270X
GeForce GTX 660 Ti
GeForce GTX 760
Shader Units
768
896
1024
512
896
768
640
640
1024
960
1280
1280
1344
1152
ROPs
16
16
32
16
16
24
16
16
32
24
32
32
24
32
Graphics Processor
GK106
Bonaire
Pitcairn
GM107
Bonaire
GK106
GM107
GM107
Pitcairn
GK106
Pitcairn
Pitcairn
GK104
GK104
Transistors
2540M
2080M
2800M
1870M
2080M
2540M
1870M
1870M
2800M
2540M
2800M
2800M
3500M
3500M
Memory Size
1024 MB
1024 MB
2048 MB
1024 MB
2048 MB
2048 MB
2048 MB
2048 MB
2048 MB
2048 MB
2048 MB
2048 MB
2048 MB
2048 MB
Memory Bus Width
128 bit
128 bit
256 bit
128 bit
128 bit
192 bit
128 bit
128 bit
256 bit
192 bit
256 bit
256 bit
192 bit
256 bit
Core Clock
925 MHz
1000 MHz
860 MHz
1020 MHz+
1100 MHz
980 MHz+
1020 MHz+
1085 MHz+
925 MHz
980 MHz+
1000 MHz
1050 MHz
915 MHz+
980 MHz+
Memory Clock
1350 MHz
1500 MHz
1200 MHz
1250 MHz
1625 MHz
1502 MHz
1350 MHz
1350 MHz
1400 MHz
1502 MHz
1200 MHz
1400 MHz
1502 MHz
1502 MHz
Price
$100
$115
$150
$120
$130
$130
$150
$200
$150
$190
$190
$200
$260
$240
Architecture
NVIDIA's primary design goal with the GM107 is to up the performance-per-watt game. If NVIDIA achieves a significant performance-per-watt gain with "Maxwell" over "Kepler," it can trade that gain for performance on bigger "Maxwell" chips bolstered by the 20 nm process. It's this fact that makes the GK107 academically interesting.
As we mentioned earlier, the GM107 essentially features the same component hierarchy as previous generation "Kepler" GPUs, but introduces changes to the design of the streaming multiprocessor (SMM), the parallel-processing sub-unit of the GPU. At its outermost ring, the GM107 features the GigaThread Engine, a component that marshals data and instructions between the graphics processing cluster (GPC), the raster operations processors (ROPs), the L3 cache, the memory controllers, the bus interface, and the display I/O.
Several GPCs can typically be routed to a GigaThread engine, but being a mid-range GPU, the GM107 features just one. This GPC features a raster engine that handles high-level assembly of data and instructions and five streaming multiprocessors (SMMs), which is where the number crunching takes place. Unlike the streaming multiprocessors (SMXs) of "Kepler" GPUs, which feature an incoherent group of 192 CUDA cores, the SMM features four groups of 32 cores each, which totals 128 per SMM. The SMM shares a Polymorph Engine that features components such as the tessellator, fetch, setup, transform, and stream output with the four groups. The four groups of 32 CUDA cores, each, feature dedicated warp schedulers and registers, with a texture cache cushioning transfers between the groups and TMUs. The GM107 hence features a total of 640 CUDA cores and 48 TMUs. At a higher level, the chip features 16 color ROPs and a 128-bit wide GDDR5 memory interface.