Palit GTX 750 Ti StormX Dual 2GB Review

1. Introduction2. Palit GTX 750 Ti StormX Dual 2GB3. Palit GTX 750 Ti StormX Dual 2GB (Super High Res Gallery)4. Testing Methodology5. Unigine Heaven Benchmark6. Unigine Valley Benchmark7. 3DMark Vantage8. 3DMark 119. 3DMark10. Alien V Predator11. Sleeping Dogs12. Tomb Raider13. Total War: ROME 214. Dirt ShowDown15. Metro: Last Light16. GRID 217. Splinter Cell: Blacklist18. Battlefield 419. Thermal Dynamics20. Acoustics Performance21. Power Consumption22. Overclocking23. Closing Thoughts24. View All Pages

Earlier this month Nvidia released their new Maxwell based GTX 750 Ti graphics card, and we were on hand to review the excellent MSI Twin Frozr solution on launch day. Today we take a look at another Nvidia partner card, the Palit GTX 750 Ti StormX Dual, featuring a substantial core and memory clock increase and custom dual fan cooler. Is it worthy of a shortlist?

The new Palit GTX 750 Ti is a modified design featuring dual fans and an extended plastic shroud.


Palit have decided to overclock their solution extensively, increasing the core clock speed from 1,020mhz to 1,202mhz. The GDDR5 memory also receives a large clock bump, increased from the reference card speeds of 1,250mhz to 1,500mhz (6Gbps effective). This should give a noticeable speed increase over both the Nvidia reference solution, and the MSI Twin Frozr card, which was clocked at a more modest 1,085mhz.

The Nvidia reference card is using 2GB of Hynix GDDR5 memory, but like MSI, Palit have opted for Samsung branded GDDR5.

The Maxwell SM architecture delivers improved efficiency and 35% more performance per CUDA core on shader limited workloads. Nvidia have changed the architecture with the SM scheduler architecture and algorithms having been rewritten to avoid stalling and further reducing the energy per instruction required for scheduling. Maxwell SM architecture enabled Nvidia to increase the number of SM’s to five in GM107, compared to two in GK107 – and all with only a 25% increase in die area.

Nvidia have explained that the organisation of the SM has been changed. Each SM has been partitioned into four separate processing blocks, each with their own instruction buffer, scheduler and 32 CUDA cores. GK107 Kepler could have a non power of two number of CUDA cores, with some shared. The new partitioning simplifies the design and scheduling logic which reduces computation latency and saves area and power demand.

Pairs of processing blocks share four texture filtering units and a texture cache. The L1 compute cache function is now combined with the texture cache function and shared memory is a separate unit, shared across all four blocks. Each Maxwell SM is smaller than a Kepler SM but delivers around 90% of the performance. The smaller area means that Nvidia can incorporate many more SM’s per GPU. The GM107 (v GK107) has 25 percent more texture performance, 1.7 times more CUDA cores and 2.3 greater shader performance.

Nvidia have improved Video capabilities with the latest hardware. Maxwell incorporates an improved NVENC block to deliver faster encode performance over Kepler – 6-8 times real time compared against 4 times. Nvidia report a 8-10 times faster decode rate too. Maxwell also has a new GC5 power state designed to drop the GPU power demand under light workload situations – such as when playing back video.

Become a Patron!