Temperature and Overclocking Testing

Cooler Testing Methods

To best gage the quality of the GPU cooler under review, GPU temperature was taken with the graphics cards idle and under load. To replicate GPU idle conditions, the system was rebooted and allowed to sit idle for 30 minutes. To replicate a stress graphics load, EVGA OC Scanner X was run over a 30 minute period using the Furry E (GPU memory burner::3072MB) 3D Test, a 1280×1024 resolution, and an 8x MSAA Antialiasing setting. After each run, the system was shut down and allowed to rest for 30 minutes to cool down. This procedure was repeated a total of 12 times – six times for the stock speed runs and six times for the overclocked speed runs.

Temperature measurements were taken directly from the GPU thermistors using TechPowerUp GPU-Z v0.8.3. For both the idle and load temperatures, the highest recorded value in the application were used for the run. Note that the temperature values are reported as deltas rather than absolute temperatures with the delta value reported calculated as GPU temperature – ambient temperature.

To adequately measure the EVGA GTX 970 SC's cooler performance in SLI, performance testing was done for all scenarios with the cards liquid cooled using a modified configuration with the HeatKiller GPU-X3 GPU water block. During the tests with the modified card, it was found that a minimal amount of airflow was needed blowing across the VRM heat sinks to prevent card instabilities. Results are reported for both single card and SLI configurations for comparison purposes.

Note that the temperature values are reported as deltas rather than absolute temperatures with the delta value reported calculated as GPU temperature – ambient temperature. For all tests, room ambient temperature was maintained between 24-27C.

Stock Temperature Testing

The graphics card temperature testing was conducted at stock speeds with air-based and liquid-based cooling.

With the two EVGA GTX 970 SC cards cooled in parallel, an interesting temperature trend emerges. While the cards remain very close in proximity temperature-wise, the primary card exhibits slightly cooler temperatures than the secondary card. This trend was seen with both cards running at stock speeds and the processor running at stock and overclocked speeds. The temperature delta between the two cards running in tandem was a mere 1C with the temperatures scaling as expected in comparison to the system running with a single card. Even with the CPU overclocked (causing more heat to be dumped into the cooling loop), the secondary card was measured to be 14C above ambient when under load.

Overclocked Temperature Testing

Using the EVGA Precision X 16 v6.3.5 tweaking software, the graphics card was overclocked to its highest stable settings using air-based and liquid-based cooling. For details on the overclocked settings used for testing and benchmarks, please see the Manual Overclocking section below.

With the GPU and video card memory overclocked, temperatures only increased slightly over those measured at default stock settings. The same temperature difference between the two cards was seen with the secondary card running slightly warmer than the primary.

Stock settings

With dual cards running under SLI, we began to see oddities with the GPU voltage being supplied to the primary card. THe primary video card showed a load voltage of only 1.5000V while the secondary card showed a load voltage of the expected 1.2000V. According to information from several support forums and NVIDIA itself, there is a known issue when operating multiple cards in SLI affecting the voltage applied to one of the cards in the SLI set. This voltage difference did not have any negative affects on card performance or operating stability at stock settings, but was found to affect card stability during the overclocking runs.

Manual Overclocking

For the overclocking tests, EVGA's Precision X 16 tweaking software (v6.3.5) was used to enable the settings with TechPowerUp GPU-Z (v0.8.3) used to validate the settings properly took effect. Graphics card stability was tested by performing a full run through the FutureMark 3DMark Fire Strike benchmark with a 1920×1080 resolution without crash or artifacting. Once the 3DMark Fire Strike benchmark run stabilized, card stability was checked using the Unigine Heaven, Grid 2, and Metro: Last Light in-game benchmark tests. The settings were further refined until no artifacting or crashes occurred in any of those applications. To further ensure card operational reliability at the configured settings, the card was torture tested over an extended period with EVGA OC Scanner X using the Furry E (GPU memory burner::3072MB) 3D Test, a 1280×1024 resolution, and an 8x MSAA Antialiasing setting.

Running dual GTX 970 SC cards in SLI with the modded configuration and the HeatKiller GPU-X3 GPU water block placed in-line with the Raystorm CPU block, we were able to increase the GPU boost clock and memory speeds to 1468 MHz (+100) and 1977 MHz (+475). As mentioned above, we saw an oddity with the GPU voltage measurements between the two cards with the primary card being undervolted by 0.05V and causing stability issues during our overclocking attempts. According to several forum posts concerning this issue, upping the GPU clock speed on the unaffected card (in our case, the secondary card) between 10-25MHz will cause the GPU voltage on the primary card to approach the expected values. Increasing the GPU offset on the secondary card to +115MHz increased the voltage for the primary card to 1.1870V from the 1.1500V is was sitting at previously. This voltage bump was enough to stabilize the dual card. Though not as aggressive as we were able to achieve with the single card setup, the overclock achieved remains respectable.

EVGA Precision X 16 profile settings

  • GPU Clock Offset, card 1 – +100MHz
  • GPU Clock Offset, card 2 – +115MHz
  • Memory Clock Offset – +500MHz
  • GPU Voltage Overvoltage – +37mV (Max)
  • Power Target – 110% (Max)
  • GPU Temperature Target – 91C (Max)
  • Fan Preset – N/A

Performance numbers

  • GPU Boost Clock Speed – 1468MHz
  • Memory Speed – 1977MHz
  • GPU voltage, card 1 – 1.1870V
  • GPU voltage, card 2 – 1.2000V
« PreviousNext »