I'm feeling pretty confused about my recent MemTest results. I ran MemTest from HCI Design on two different systems with the same memory kits and about 98% of available memory used. The first system has 2x16GB RAM running at 3200MT with a 5700G CPU using 16 threads. The second system has 4x16GB RAM running at 3400MT with a 5950X CPU using 32 threads. Each CPU core is processing around 4GB, so I expected both systems to finish at roughly the same time. However, the first system is completing the test about 2.25 times faster than the second one, reaching 900% completion overnight while the second only got to 400%. I thought the second system would be slightly faster due to its RAM overclock and better benchmarking in AIDA64. Is there something I'm missing with the second system?
3 Answers
Memory access speed can heavily depend on the paths to and from the memory, particularly the memory controller itself. Both CPUs might actually have similar IMCs, which could limit their performance. Plus, it could be that the 5700G is already maxing out at what it can handle with 3200MT RAM, whereas adding more cores doesn't always yield more performance.
Don’t forget that just having more cores doesn’t necessarily mean better performance for every task. With multithreaded applications, gains can diminish after a certain point, especially considering you have different amounts of memory. The 64GB in the second system might not necessarily translate to faster speeds in MemTest.
Exactly! The CPU types are different, and that probably plays a huge role too. The design of the integrated memory controller (IMC) on the 5700G and 5950X might be where the bottleneck is happening. Running a one-instance test on both systems could help clarify things.
You might be overlooking a few details. In your setup, the first system is using a dual-channel configuration with 4 channels effectively running across double the memory chips, which can really stress the memory controller. The second system is still in dual-channel mode, so it might not be utilizing its resources as efficiently as the first one.
That makes sense! I'm thinking of running a single instance of MemTest on both to compare how they perform with the same load. Maybe it's a Windows issue too since Sys 1 is on Win10 and Sys 2 on a fresh Win11 install.
Good point! You also have to consider that consumer CPUs typically don’t truly support quad channel memory. You might have two DIMMs per channel, but that essentially gives you the bandwidth of a dual-channel configuration.
You’re spot on. The architecture of the 5950X with its two chiplet design might be why you’re seeing that 2.25X difference instead of a straight 2X. It gets complicated!