GTX 1070 Thermal Benchmark – Founders Edition / Reference Temperatures
Thermal benchmarking is our “thing.” It's what we most enjoy doing, and have spent the most time refining. For the GTX 1070, we sent the card through the usual “thermals over time” and peak-load analysis. Here's what we've got for the averaged peak load temperatures:
As shown, the GTX 1070 Founders Edition – equipped with the nVidia stock cooler – performs about equally to the GTX 980 reference with regard to thermals. The GTX 1070 sits at ~53.86C (delta tempearture over ambient); that's a fair bit cooler than the GTX 1080 FE, operating at 57.51C. Our custom-made version of the GTX 1080, the GN Hybrid, had an operating temperature of around 18.66C delta T, just for reference. AMD's liquid-cooled R9 Fury X runs at ~36.39C (with an impressively low idle temperature), but uses more expansive cooling than the “hybrid” solution – the Fury X has no VRM blower fan, and instead uses liquid to cool all of its components. This is in contrast to the “hybrid” solutions, which use a VRM blower fan and only liquid-cool the GPU proper.
The GTX 1070 is about 6.55% cooler than the GTX 1080.
GTX 1070 Thermals Over Time
Here's an over-time presentation of the data. The results from the above peak average load chart are pulled from the stabilized portion of this chart (what we sometimes call “equilibrium”). The data and test passes are the same, just stretched over time rather than displayed in easy-to-read bars.
All cards begin heating at the same time, following a 120-second idle period. Our in-house test automation ensures that these lines align correctly.
The GTX 1080 and GTX 1070 follow a similar ramp curve. The GTX 1070, in pink, keeps a consistent temperature with few dips or raises, unlike some other cards on the bench.
The thermal performance for nVidia's “Founders Edition” cooler is reasonable. As we said with the GTX 1080, it's about what you'd expect from a “reference card” – or one made by the GPU supplier. AIB partner cards, like those from MSI, EVGA, Gigabyte, and others, will push lower temperatures as a result of more refined cooling solutions. It is sometimes the case that these devices cost more – like the EVGA Hybrid cards or Gigabyte Waterforce cards – but they often ship for about the same (or slightly lower) price as the reference design.
NVidia's GTX 1070 doesn't excite us from a thermal control standpoint. It does what it should – keeps the card at a reasonable temperature and with a VRM fan noise level just under 50dB (though not shown in this review, because of time issues from brushing up against the Taipei/Computex trip). The FE sets a benchmark for the AIB partner cards, for which we're most interested in performing thermal benchmarks.
GTX 1070 Thermals versus Frequency (And Thermal Throttling)
This endurance test was conducted over a (shorter than normally, due to constraints) 1-hour test period. We use real-world games to analyze performance and look for points of throttle during execution, usually coinciding with temperature spikes. This time, we used Metro: Last Light at 2560x1440 with Very High quality and High tessellation. This torture tests the card.
Temperature is represented as an absolute value, unlike our previous charts; it is not a delta value. This is to more clearly illustrate the point of clock throttling, should one exist.
Of the hour-long period, four major dips occurred on the GTX 1070. The card fell down to the low hundreds for its frequency, but quickly shot back up. Regardless, these instances will create frametime latency spikes and can cause a user-observable “lag spike” in gameplay for a brief time. Four for an hour isn't bad, but this would be resolved by an AIB cooler. Each spike coincides with temperature jumps. Here is the excerpted data from the most noteworthy dip:
Catch it? The clock only dips down for one second, but it's enough to recover an additional 4C from the GPU. That hard slam to 139MHz creates an observable “lag spike” in gameplay for less than one second of time. Whenever the GPU hits 77C, we observed this throttle.
More-or-less a non-issue, but most AIB coolers should not be hitting the same high temperatures. We'll see when they're out, though.
GTX 1080 Thermals versus Frequency (And Thermal Throttling)
Just for a look at our previous testing, here's the GTX 1080 excerpt:
The frequency is fairly stable here, but shows dips in frequency which grow in consistency as the test continues. Let's zoom in on that data; the scale is a little wide to properly see what's going on. Here's a blown-up version of those spikes, truncating the data from ~5500s to ~7500s:
Frequency fluctuations show a range of approximately ~60MHz each time the GPU diode hits ~82C absolute. This can trigger a slight latency increase or framerate fluctuation at the exact moment of frequency fluctuation, but is basically imperceptible.
Here's a look at the most exaggerated spike:
|Time (s)||Core MHz||Abs. T. (C)|
A few hits later, the temperature falls pursuant to this frequency throttle (which is the point of the momentarily throttle):
|Time (s)||Core MHz||Abs. T. (C)|
With the frequency restored and the game continuing to crunch numbers and graphics, temperatures eventually return to 82C and drop again.