- Joined
- Mar 7, 2008
https://support.microsoft.com/en-nz/help/4497165/kb4497165-intel-microcode-updates
This test kicked off from a post on another forum. The poster claimed a small boost in performance after installing the Windows update which supplies new(er) microcode than might be provided by a BIOS, especially on older systems where it hasn't been updated by the manufacturer in a long time. I did testing on a Haswell system, and there might be 1% in it for a patched up Win10 system, comparing without to with the microcode update patch. I don't have an easy way to do a test with pre-Spectre/Meltdown era software. It has to be assumed modern Windows has those baked in to some degree regardless of hardware support.
Tests were run at least 3 times, and the best result presented on the theory there are things that can slow a benchmark down, but none that will speed it up. So as you do more runs, you converge towards the maximum possible.
Test system has Win 10 1909 with Feb updates installed. Hardware is an i5-4570S, and the mobo bios was last updated in 2015, so no chance for the bios supplied microcode to have updates for Spectre/Meltdown.
Before
Cinebench R20: 1195
P95 29.8b6
4096k FFT: 122.92
128k FFT: 7948.86
After
Cinebench R20 1209 (+1.2%)
P95 29.8b6
4096k FFT: 122.62 (-0.3%)
128k FFT: 8016.93 (+0.9%)
There is a small change detectable, and because of the multiple runs, not all of this can be put down to run to run variations. Nothing really practical but you can't say it isn't there. I would add, for the two cases where there was around 1% improvement, these are NOT ram bandwidth sensitive tests. Cinebench has never shown much dependence on ram performance, and Prime95 128k FFT is small enough to easily fit inside the L3 cache of most modern CPUs. Prime95 4096k FFT is too big to fit in mainstream consumer Intel CPU caches so is more easily ram bandwidth limited for all but the slowest CPUs (e.g. dual cores are probably fine). That might be the bigger limit, and/or the update might make ram access fractionally worse than before.
I also ran the MS powershell tool Get-SpeculationControlSettings to see if anything changed. I note the following line was different after applying the patch. I have no idea if this is significant.
Windows OS support for MDS mitigation is enabled: False before, True after
This test kicked off from a post on another forum. The poster claimed a small boost in performance after installing the Windows update which supplies new(er) microcode than might be provided by a BIOS, especially on older systems where it hasn't been updated by the manufacturer in a long time. I did testing on a Haswell system, and there might be 1% in it for a patched up Win10 system, comparing without to with the microcode update patch. I don't have an easy way to do a test with pre-Spectre/Meltdown era software. It has to be assumed modern Windows has those baked in to some degree regardless of hardware support.
Tests were run at least 3 times, and the best result presented on the theory there are things that can slow a benchmark down, but none that will speed it up. So as you do more runs, you converge towards the maximum possible.
Test system has Win 10 1909 with Feb updates installed. Hardware is an i5-4570S, and the mobo bios was last updated in 2015, so no chance for the bios supplied microcode to have updates for Spectre/Meltdown.
Before
Cinebench R20: 1195
P95 29.8b6
4096k FFT: 122.92
128k FFT: 7948.86
After
Cinebench R20 1209 (+1.2%)
P95 29.8b6
4096k FFT: 122.62 (-0.3%)
128k FFT: 8016.93 (+0.9%)
There is a small change detectable, and because of the multiple runs, not all of this can be put down to run to run variations. Nothing really practical but you can't say it isn't there. I would add, for the two cases where there was around 1% improvement, these are NOT ram bandwidth sensitive tests. Cinebench has never shown much dependence on ram performance, and Prime95 128k FFT is small enough to easily fit inside the L3 cache of most modern CPUs. Prime95 4096k FFT is too big to fit in mainstream consumer Intel CPU caches so is more easily ram bandwidth limited for all but the slowest CPUs (e.g. dual cores are probably fine). That might be the bigger limit, and/or the update might make ram access fractionally worse than before.
I also ran the MS powershell tool Get-SpeculationControlSettings to see if anything changed. I note the following line was different after applying the patch. I have no idea if this is significant.
Windows OS support for MDS mitigation is enabled: False before, True after
