The MSI MEG Ai1300P PCIE5 1300W PSU Review: The ATX 3.0 Era Has Begunby E. Fylladitakis on December 8, 2022 11:00 AM EST
Our First Foray Into ATX 3.0 PSU Testing: A High Hurdle to Clear
In light of the new ATX 3.0 standard, we took a shot at adding power excursion compliance testing into our articles. Given that this is the big addition to the ATX 3.0 specification– and indeed its very reason for being – it's where we would like to see if PSUs are truly living up to the very high standards set by the new specification.
Unfortunately, the testing requirements of the new standard have proven too high for our equipment - and that of the majority of small laboratories - to meet. The power excursion specifications suggest electrical current-to-time figures that are extremely short. For example, taking the MEG Ai1300P of this review into consideration, for the single case of the 200% power excursion testing, we would need to test that it can handle 2600 Watts for 0.1 ms. Assuming a starting load of about 800 Watts and 65 A on the 12 V rail, the 12 V load would have to get up to 215 A and back down to 65 A within 0.1 ms. Furthermore, according to Intel's testing guide, this would have to continue for at least a minute, which means at least five hundred cycles in this scenario.
In an ideal world, we would just enter the current and time figures into the software and our electronic loads would run the test, instantly getting the load up to 215 A for 0.1 ms and then immediately back down to 65 A for 1.9 ms, according to the guide's requirements.
In the real world, however, there is no such thing as "instantly". Electronic loads, like any other device that is bound by the laws of physics, require time to react. The speed at which an electronic load can increase its amperage is called Ramp (or Slew) Rate and our larger loads have an ideal Slew Rate of 0.5 A/μS. Assuming that they operate linearly and exactly as specified, which no electronic load does for a variety of reasons, our two primary electronic loads in parallel would require at least 0.15 ms (150 μS) just to get the load up at 215 A. They would also require time, albeit less than half of it, to get the load back down to 65 A. When the test dictates a test time of 0.1 ms and the testing equipment requires at least twice that much time just to react, it goes without saying that testing results are highly unreliable.
Nevertheless, we took a shot at testing the power excursion capabilities of the MSI MEG Ai1300P PCIE and of the few ATX 3.0 compliant units that we currently have available. We took two approaches: one by assuming that our electronic loads are "ideal" and programmed the exact duty cycle figures that Intel dictates in their guide, and one by trying to take into account the real slew rate times of our loads and calculate the RMS equivalent duty cycle.
Both of our approaches ultimately failed, as all of the PSUs we currently have available would shut down at most tests above 120% power excursion - therefore we need not worry about our loads being insufficient to test the MEG Ai1300P at 200% excursion (we are also currently limited to 2400 Watts on the 12V line). Theoretically, testing with the RMS-equivalent duty cycle times should work and the PSUs should not be shutting down, yet we cannot claim that the units are not technically capable of meeting their specifications when our equipment is not meant to be running such tests.
Intel requires the PSUs to have a slew rate of at least 5 A/μS, so an electronic load must be at least as fast as that figure to be able to perform ATX 3.0 compliance testing. From a professional's point of view, proper testing would require the testing equipment to be at least 30% faster than the absolute minimum required. This requires a highly advanced (and expensive) electronic load with multiple modules, like the Chroma Mainframe and High-Speed modules Intel themselves is using, which has a total slew rate of 8 A/μS and it would need only 0.02 ms to get the load from 65 A all the way up to 215 A - and that still is 20% of the test's required 0.1 ms time in our example, a figure that many experts would find far too great for precise measurements.