Author: Greg Gutmann
Affiliation: Tokyo Institute of Technology
The Process of Solving Thermal Issues
I started by installing the four ZOTAC GAMING GeForce RTX 2080 Ti Blower (ZT-T20810A-10P) cards in an X299 ROG Rampage VI Extreme motherboard. Then I proceeded to run my own VR simulation on two GPU as a thermal test, within a minute the two GPU reached 91C.
My first reaction was to download Firestorm, Zotac’s GPU tuning application, and manually set the fan speeds higher. This resulted in nearly no change even when setting the fan speed to 75%, higher fan speeds were not tested for noise reasons. This result was not a very big surprise though, as the cards were literally pressed together. It was actually a bit curious how there was any airflow. Next, I took apart the GPU to better understand the cooling design. As well as to send PCB pictures to EKWB, a liquid cooling company, to see if they had any compatible GPU water blocks.
Over the Top Idea (optional reading):
Since the issue was the airflow, my first idea was cutting holes in strategic locations to create an air intake while not greatly impacting the blower’s ability to force air down the inside of the card and out the back of the case. After a bit of googling it seemed like this hasn’t been tested. Results could be hypothesized though based on past computational fluid dynamics(CFD) simulations and knowledge of pressure around these types of fans.
After this I actually got excited, my reaction was I can simulate this with my code and test to see if it is a viable solution, as well as test where new air intakes would provide the greatest benefit. While I may try this if I have time in the future, I decided to take a simpler approach.
To increase the amount of air flow near the blower fan intake I removed all of the backplates from the GPU, then reinstalled the GPU into the desktop.
This created about 4.5mm of spacing between each card. I used the plastic HDMI caps from the GPU, with some electric tape for friction, as spacers to ensure that all of the GPU were evenly spaced. (not necessarily recommended, but it works and it is free) After this modification, I reran my simulation on two GPU for ten minutes using default GPU settings. The result of this test was temperatures between 74C and 86C with the fans running at about 50% (very quiet but not silent).
Removing the backplates appears to be the best solution and sufficient for reducing the temperatures to reasonable levels. With this modification,the Zotac 2080 Ti blower cards become a good option if four 2080 Ti are needed in a single workstation. I was actually surprised at how well the minor change worked.
- Before: 91C at 75% fan speed
- After: ~80C at ~50% fan speed
These results show that water cooling or other drastic modifications are not needed to run four RTX 2080 Ti in a workstation, assuming that room temperatures are low enough. The room was 21C during the test. Might need to retest when summer comes.
Later I will run my multi-GPU simulation to stress all of the GPU and update this post with those results. I will also check for a standardized multi-GPU stress test, to provide some results that others can reproduce. I ran FurMark, but that only tested the top GPU plugged into the monitor.
I have also tested two triple fan open-air cooled 2080 Ti pressed together, more or less out of curiosity. I do not recommend this as they became very hot. If there were four in a case the airflow would be fairly non-uniform around the cards, not an ideal circumstance, and that is assuming they are still able to move air. Using that assumption, it might be feasible if the PC case has very high airflow coming into the case and out of the case. However, by using blower style cards there is a more uniform and predictable flow of air, in the case and past the GPU heat sink, which is typically beneficial when cooling.
The post’s contents can be used as long as the source and URL are cited. Thanks 🙂