RTX 2080 Ti reboots my computer during Tensorflow executions

For some Tensorflow scripts, the computer freezes and then reboots, the same codes executed on a GTX 1080 Ti have no problem. The codes that I execute just use 200MB and 10% of the RTX 2080 Ti.

After a long search I found this answer:


Were they suggested running two independent cables from the PSU to the card, but I can’t do this by myself or I’ll lose the computer’s warranty, so I send it back to the repair service and they weren’t able to fix the problem. (I have a Seasonic Focus 80 plus gold 1000FX fully modular PSU of 1000W)

After tinkering in the BIOS I switched on the ‘GAME BOOST’ and ‘XMP’ buttons and I haven’t had any problems yet. But a question arises to me; Is it safe to permanently keep these buttons ON? is there going to be an update on drivers that fix this problem?

Did you discard the possibility of having a faulty card by testing it in another computer and running the same TF job to see what happens?
Are you overclocking the card and/or monitoring to see if it is overheating?
The settings you changed in BIOS are related to CPU/memory system. Were you overclocking/monitoring these too?

Is there any confirmation that this is related to a driver problem?

This seems like sound advice. Given that the RTX 2080 Ti has two PCIe power connectors, how do you currently supply power to it? [Later:] Looking up your PSU at https://seasonic.com/focus-plus-gold, I find that it comes with PCIe power cables that have two connectors, which I find unusual. Hooking up your RTX 2080 Ti to the two connectors on the same cable would be a Bad Idea™. Presumably those two connectors are for connecting to two 6-pin plugs in daisy-chain fashion, on a strand designed to supply 150W.

Depending on what exact model of the RTX 2080 Ti you have, specified power draw could be 250W to 270W. Consequently this GPUs uses two 8-pin power connectors, as specified by NVIDIA: https://www.nvidia.com/en-us/geforce/graphics-cards/rtx-2080-ti/.

Generally speaking, your symptoms (sudden reboot under GPU load) are consistent with an overloaded power supply. GPUs often draw power from multiple sources:

(1) The PCIe slot can supply up to 75W, although for most NVIDIA GPUs, maximum power draw through the slot seems limited to between 40W and 50W
(2) Each 6-pin PCIe power cable can supply up to an additional 75W
(3) Each 8-pin PCIe power cable can supply up to an additional 150W

Inside the PSU the PCIe power cables are wired according to the maximum current they need to supply per the list above. So if you were to use a 6-pin to 8-pin converter on a 6-pin PCIe power cable because your PSU lacks two 8-pin PCIe power cables, you could easily overload that strand. Same for hooking up two 8-pin plugs to a strand designed for 150W.

For a typical PC configuration where the RTX 2080 Ti is the only GPU in the system, a 650W PSU should be sufficient, unless you also have a power-hungry high-end CPU plus a lot of high-speed system memory, in which case a 750W PSU would be preferred. Your 1000W power supply therefore seems more than generous. 80 PLUS Gold compliance is what I would consider the minimum standard for such a machine, so you are good there as well.

I have no idea what the BIOS switches are that you mention. What specific model of the RTX 2080 Ti do you have installed (vendor name)? Are there any other GPUs in this system?