Win10 vs Win7 - Same Hardware - Much Lower Performance

[GWild2020]

Hi - I have a PC that I have upgraded from Windows 7 Pro to Windows 10 Pro.

I did a CLEAN INSTALL, not an upgrade. The only difference I can see is the installer changed drive C from MBR to GPT and I now boot UEFI.

My Passmark score was 7500+ in Windows 7. My Passmark scroe is 6000 in Windows 10. Anyone else see this sort of huge drop in performance? And yes, I have installed all the latest drivers, and that's made no improvement.

The places it is radically different is SSD (25k down to 18k) and GPU (14k down to 11k). Oddly, CPU benchmarks are about the same (about 23k).

Here are some of my tests.

[David (PassMark)]

Drop in 2D and Disk result are likely the result of security patches. Spectre & Meltdown class of security issues causes all API calls to be slower. Which effects disk and 2D the most. Then there was the more recent Win10 KB4512578 patch, which caused another (up to) 30% drop. Maybe there are also some driver differences for your hardware.

3D result might be due to different tests being run. In Win10 you have access to DirectX12 at 4K resolution. Which isn't available on Win7.

[GWild2020]

In other tests, M.2 disk performance is virtually the same between 7 and 10. Samsung Magician reporting 3200MB/s disk reads; Passmark is reporting only 1200 or so... I can't see why there's such a large discrepancy unless Magician is somehow bypassing OS I/O routines.

As for updates, my Win7 install was current, with all of the latest patches up to Dec 23, 2019; including the Spectre and other patches, and I doubt that's the reason for the drop.

GPU 3D with Dx12 might be part of that drop, but I wonder. Too bad the app doesn't let me choose to run a subset for comparison.

Frankly, I'm about to revert to Win7 ... a 30% drop in performance is a bit hard to accept.

[GWild2020]

After hours of research, trials and errors, I've resigned myself to the facts Windows 10 is much slower than Windows 7, despite all the benchmarks that suggest otherwise. That or my GTX1080 FTW2 is a dog and my EVO 970 SSD is too (they aren't).

Also, I have come to realize Passmark is also part of the problem. Tests are NOT compatible if run on Windows 7 vs Windows 10. There is no way to equalize OS differences in any meaningful way, nor is there a way to filter test results to do reasonable comparisons (compare this video card against only Windows 10 results -- not Win7+Win8+Win8.1+Win10). With Windows 7 results my hardware was reporting about 97th percentile. With Windows 10 my results are now 69th percentile. Only change being the OS. So the benchmark is broken -and there is also probably something in Windows 10 that is also slowing PCIe data flow compared to Win 7.

[David (PassMark)]

Samsung Magician is pretty rubbish software. But aptly named.

You can compare the result of individual tests in PerformanceTest.
Linear read speeds aren't really impacted by the patches. But doing lots of seeking and small disk reads is.
I don't have a list, but I was pretty sure some of these patches were Win10 only. But we haven't done a deep analysis.

Just because you don't like the benchmark results doesn't mean the benchmark is broken.

See also
https://www.passmark.com/forum/perfo...4532#post44532

[GWild2020]

The results are broken if the same hardware has radically different results for something unrelated to the hardware.

If all GTX1080's are 30% slower with Windows 10, it would mean something entirely different than just my system being slower.

Right now, if I compare all GTX1080 benchmarks in the web pages, it is a meaningless comparison because OS versions are masked. If users could filter a specific card to a specific OS, then the results become comparable. e.g., I am comparing my results today against my earlier results reported under Win 7 when I look at the GPU benchmarks. Yet, my hardware is the same. If the differences between Win7 and Win10 are due to DX12, fine, but let folks know there are differences and what to expect.

It's like comparing the same tire - but one is mounted on a truck and another on a sports car. Sure - same tire - but vastly different results.

So allow users to filter results: I want to see all GTX1080 cards on an i9-7900 paired with an X299 chipset and 2666DRAM and WIN10. I really doubt a comp against a GTX1080 in a Celeron with 1GB or DDR2 and WINxp is helpful. But it is possible such a system in with the current web page data here.

[David (PassMark)]

You can load up individual results to compare (called Baselines) from within the PerformanceTest software. If you just want to look at Win10 results then you can.

92% of our collected results are now from Win10. So Win7 doesn't have much influence anymore.

Doing a pure hardware benchmark of video cards isn't possible. 20 years ago this was possible, but nowadays there is no way to use the video card without also using it's device driver, the operating system and an API on top of that (e.g. OpenGL, DirectX 9, etc..). Then there are heaps of other software factors. Device driver settings, Windows power settings, thermal throttling, monitor resolutions, adaptive VSync, installed patches, CPU used, overclocking, etc... So you can't separate the software from the hardware for video card benchmark. You benchmark both the software, the hardware and the configuration setting and that can't be avoided.

[GWild2020]

I understand the complexity of bench marking stuff (I started writing code 50 years ago - seems so ancient now). But it should be easy enough to partition results so the numbers are comparable rather than not...

I wasn't aware the baseline could be selective by OS. So I dug into the baseline tools, and figured out how to filter by OS. It also lead me to see many Windows 10 systems are reporting lower numbers like I am seeing. A few aren't but they appear to be heaving tweaked systems (4900MHz for their 3.3GHz CPU, lol). But in most cases my factory tweaked FTW2 GPU results are still much lower than all the other baselines: my latest 2D and 3D results are the lowest in all of the comparable systems - by 25% - though all of the results are earlier Win10 builds pre 18262.

Even then, it seems something is broken.

[David (PassMark)]

But it should be easy enough to partition results so the numbers are comparable rather than not..

When Win8 & Win10 initially came out there wasn't much difference in terms of performance to Win7. So no need to separate them. But after years of patches to the O/S and drivers they are diverging a bit now. But as mentioned 92% of submissions are now Win10. So breaking out the Win7 and Win8 results doesn't really make sense either.

But in most cases my factory tweaked FTW2 GPU results are still much lower than all the other baselines

Which would point to it being a configuration issue.

[GWild2020]

Which would point to it being a configuration issue.

Except it isn't. I've been exploring other benchmark tools. They seem to agree my hardware is fine. I am benching well ahead of comparable systems running Windows 10. It seems only Passmark 9.0 is rating my system poorly. Even EVGA (the hardware vendor) says the Passmark numbers shown look wrong.

It seems in the 2D arena, Passmark is just out to lunch. And the 3D numbers seem to be because I really do have a 4k monitor connected. A penalty for having a good monitor seems a lame way for a benchmark to work. Even setting the display down to 1920x1080, Passmark bumps me back to 4k for the tests, so I can't get a result that is comparable to the other Win10 systems that can't do 4k.

ps: My nVidia / EVGA driver is for Win7, 8, 10. I can't agree the driver is the issue, entirely.

[GWild2020]

A couple of 'other' benchmarks - that unlike Passmark - show my system is performing reasonably well compared to similar systems. Yet Passmark says my system is the lowest ever tested... go figure.

Firestrike is DX11, TimeSpy is DX12.

Firestrike: Better than 88% of all results



Time Spy: Better than 71% of all results

[David (PassMark)]

Yet Passmark says my system is the lowest ever tested..

Rubbish.

Here is the actual distribution for the GTX 1080 for the PassMark 3Dmark.



Your result of 11,800 is within a couple of percentage points of the average (12,455 at the moment).

The worst ever tested is in fact a result of zero (typical of having no device driver installed). But we filter that for the distribution graph. So the lowest meaningful result is around 5000.