Windows 11 only has 45% of random write speed in Nvme SSD

Any chance Microsoft is listening and working on this? The holiday code freeze is over, folks :)

I recently upgraded my motherboard to a much better model (the new Asus TUF Gaming X570 Pro Wifi II), one effect is that my 3DMark PCIe score jumped from 22.3 to 27.2 GB/s, a crazy +22%. But the 980 Pro's random write iops are still almost the same, perhaps just a small lift due to the motherboard (159K->163K) but still a fraction of the performance due.

I suspect not, as some people/tech sites INACCURATELY described a recent MS update as containing a fix: Microsoft Has a Fix For Slow SSD Speeds in Windows 11 | Digital Trends

Not true. What MS addressed was for ALL types of drives, hard and otherwise (basically a "journaling" function), and did not significantly change the SSD NVMe random write issue (when comparing Win 10 to Win 11 performance AND where the SSD contains the operating system). Unfortunately, MS "FIX" for slow NVMe SSDs was blasted all over the Internet. So I suspect MS is now thinking it is a-okay -- nothing more to do here.

This is the more accurate (but less press attention getting) version of the alleged "fix": Microsoft's NVMe slow-down fix for Windows 11 didn't work according to our testing - Neowin

Yeah I've seen those equivocated reports about the fix. Meanwhile, today I just updated my 980 Pro's to the new firmware (5B2QGXA7) and before anyone hopes that's the fix, nope still the same 162K iops random write. Random reads are up a bit, from 759K to 820K for me (+8%), but the previous result was already pretty good and this is the kind of improvement you would expect from driver optimizations but the massive underperformance for random writes can only be a deeper Windows I/O subsystem problem.

Yeah I've seen those equivocated reports about the fix. Meanwhile, today I just updated my 980 Pro's to the new firmware (5B2QGXA7) and before anyone hopes that's the fix, nope still the same 162K iops random write. Random reads are up a bit, from 759K to 820K for me (+8%), but the previous result was already pretty good and this is the kind of improvement you would expect from driver optimizations but the massive underperformance for random writes can only be a deeper Windows I/O subsystem problem.

The latest officially released Win11 build (build 22000.469), which came out on 1/25/2022, has made some incremental improvements with this issue (maybe unintentionally, because it's not mentioned in the notes). But others have noticed the same.

I noticed about a 15% improvement on random write speeds vs the prior build.

The latest officially released Win11 build (build 22000.469), which came out on 1/25/2022, has made some incremental improvements with this issue (maybe unintentionally, because it's not mentioned in the notes). But others have noticed the same.

I noticed about a 15% improvement on random write speeds vs the prior build.

I also have the latest Windows update (specifically KB5008353) so maybe that was the cause of my improvement in random reads, not the Samsung firmware update. There was also KB5007262 a month before which also claimed to fix SSD performance (and hyped by many tech sites as the big fix we needed), that one had absolutely no effect for me. Both fixes seem to be focused on the USN journal, and maybe that recovered some performance for some users, many factors can be involved including the NVMe controller, chipset, Windows configuration etc. But the big stinker should be something else, that Microsoft didn't fix yet.

is this issue still being fixed? I still have the problem even with the latest update KB5011563