Windows 10 1809 Benchmarks. The Spectre patch kills performance.

Kol12 · 16 Oct 2018

Disabling both Spectre and Meltdown patches smoothed my gaming benchmark out considerably. The previous GPU and CPU wild frametime spiking smoothed out to almost flat. Higher frames score and minimum FPS increased by a whopping 28 FPS.

I hope Microsoft can improve the performance impact in future patches but I don't know whether it's possible..

Kol12 · 17 Oct 2018

TairikuOkami said:

Some people solved it by disabling HPET (High Precision Event Timer) and noticed 30% performance increase.

So are people still fiddling with HPET as a means to increase performance? It's something I don't know too much about. Would appreciate any info...

CountMike · 17 Oct 2018

Kol12 said:

So are people still fiddling with HPET as a means to increase performance? It's something I don't know too much about. Would appreciate any info...

It's sketchy, somewhere it helps, somewhere not, mostly no difference on or off.

Kol12 · 17 Oct 2018

TairikuOkami said:

Thanks. Great tool ,when I changed it to 0.5ms, the latency went down by ~1000, I guess that is good?!

Does the lower latency help with performance anywhere that you've noticed? Does the tool have to be left open for to continue working?

Kol12 · 17 Oct 2018

CountMike said:

It's sketchy, somewhere it helps, somewhere not, mostly no difference on or off.

From what I've read it has to be enabled in both the BIOS and OS to work, is that correct? What does HPET look like in the BIOS? Can't say I recall seeing any HPET, does get it named something else?

CountMike · 17 Oct 2018

Kol12 said:

From what I've read it has to be enabled in both the BIOS and OS to work, is that correct? What does HPET look like in the BIOS? Can't say I recall seeing any HPET, does get it named something else?

Some BIOS versions don't have that option at all, for my MB only in some hacked/modified BIOS version so I can do only SW part anyway.
HPET Support - The BIOS Optimization Guide - Tech ARP

Cliff S · 17 Oct 2018

Kol12 said:

So are people still fiddling with HPET as a means to increase performance? It's something I don't know too much about. Would appreciate any info...

On this thread start here: 2017 Hardware Thread - Page 32 - Windows 10 Forums

Also on this thread start here: 2018 Hardware Thread - Page 65 - Windows 10 Forums

Kol12 · 17 Oct 2018

Cliff S said:

On this thread start here: 2017 Hardware Thread - Page 32 - Windows 10 Forums

Also on this thread start here: 2018 Hardware Thread - Page 65 - Windows 10 Forums

Cheers thanks.

TairikuOkami · 17 Oct 2018

Kol12 said:

Does the lower latency help with performance anywhere that you've noticed?

I have not noticed any difference really.

Kol12 said:

Does the tool have to be left open for to continue working?

Yes

Geneo · 17 Oct 2018

craker said:

Everyone run for their AMD machines, this Spectre patch in 1809 is crucifying machine performance.

This is a small portion of benchmarking numbers from our MFC CAD application. It's getting on in years and doesn't use any GPU acceleration. Even an 1809 machine without the Spectre patch performs badly against a 1709 build.

OUTSIDE LIMITS test "adviser\analysis\analyseAllRules"
8 ( 4.55 ) secs in "Win10_1709"
13 ( 7.55 ) secs in "Win10_1809_NoSpectre"
22 ( 11.25 ) secs in "Win10_64_1809_WSpectre.htm"

OUTSIDE LIMITS test "adviser\analysis\analyseOneRule"
8 ( 4.80 ) secs in "Win10_1709"
11 ( 8.02 ) secs in "Win10_1809_NoSpectre"
20 ( 11.98 ) secs in "Win10_64_1809_WSpectre.htm"

OUTSIDE LIMITS test "adviser\analysis\filename_invalidForPred"
8 ( 4.69 ) secs in "Win10_1709"
11 ( 7.80 ) secs in "Win10_1809_NoSpectre"
22 ( 11.86 ) secs in "Win10_64_1809_WSpectre.htm"

OUTSIDE LIMITS test "adviser\analysis\report"
8 ( 4.59 ) secs in "Win10_1709"
11 ( 7.63 ) secs in "Win10_1809_NoSpectre"
20 ( 11.50 ) secs in "Win10_64_1809_WSpectre.htm"

OUTSIDE LIMITS test "adviser\decouplingFiles\AN-Pins_NoPart"
7 ( 4.45 ) secs in "Win10_1709"
11 ( 7.59 ) secs in "Win10_1809_NoSpectre"
20 ( 11.83 ) secs in "Win10_64_1809_WSpectre.htm"

OUTSIDE LIMITS test "adviser\decouplingFiles\analyseBrowse"
8 ( 5.25 ) secs in "Win10_1709"
11 ( 8.06 ) secs in "Win10_1809_NoSpectre"
19 ( 11.66 ) secs in "Win10_64_1809_WSpectre.htm"

My machine is performing better than ever. In fact, the GDI performance on 1809 with the Spectre patch enabled is better than 1803 with it disabled, which translates to a better overall desktop experience. And other computational/multimedia benchmarks are only a few % down with Spectre enabled vs. disabled in 1809.

I think Microsoft ought to be given some credit for stepping up to the plate and tackling performance problems that were not of their making.