The developers of the GNU MP (GMP) library, which provides functions for performing arithmetic operations with arbitrary precision, have warned users about an incident with repeated failure of AMD Ryzen 9000 series processors based on the Zen 5 microarchitecture during testing of the library. It is assumed that the damage to the processors was caused by overheating that occurs during intensive execution of GMP functions. Until a full analysis of the problem is completed, users are advised not to actively use GMP on systems with AMD Zen 5 CPUs.
Concerns are caused by repeated failure of the AMD Ryzen 9950X (AMD Zen5 X16 4300MHz) CPU on server in the project's test infrastructure. The first failure occurred in February 2025, after which the processor was replaced under warranty. At the end of August, the situation repeated itself, and the CPU, replaced under warranty and installed in a system with a different power supply and motherboard, also stopped functioning, showing the same symptoms indicating overheating (a discolored area approximately 1 centimeter in diameter appeared on the chip).
Preliminary data indicates that the GMP library makes excessive use of MULX instruction loops, which results in AMD Zen 5 systems consuming significantly more power than stated and failing to compensate for the heat with a cooling system selected according to specification. During both corruptions, a test was performed with manual assembler loops with a single MULX instruction, creating the maximum load.
The systems used Asus Prime B650M-K and Asus Prime B650M-A WIFI II motherboards, as well as Corsair SF450 and Corsair RM650 power supplies. Both configurations used the Noctua NH-U9S cooling system, but in the first case the radiator was installed in the center, and in the second, as recommended by the manufacturer, it was shifted to the side with greater heating.
It is noted that in the second case, the thermal paste was unevenly distributed (the layer was thinner on one side), and it is entirely possible that the second failure was caused by improper heatsink installation. The Ryzen 9950X CPU is specified to require a 170-watt thermal design power (TDP, Thermal Design Power), while the cooling system used provided 165 watts of heat dissipation. The missing 5 watts were compensated for by additional coolers in the case and a low room temperature of around 20Β°C. Both processors failed after several months of operation under high load. Meanwhile, another processor has been in use for other tasks in the project's infrastructure for a long time. server in the same configuration and with the same load level, and there are no problems with it.
Source: opennet.ru
