GPU Performance takes hit when installed as MSIX vs MSI (Sandboxed vs Unsandboxed)

[debraspicher]

This was discussed in another related thread and tested by a number of users using the “unzip to desktop” unofficial install method for the MSIX package in order to negate sandbox. It was learned that there is a significant performance increase when using the same package un-sandboxed versus sandboxed. (i.e. sandboxed == slower) It seems to impact higher end hardware more, but it isn’t clear for sure without more machines being tested. I tested an AMD card in the other thread and it showed a hit, but that card is not that powerful compared to the others. (Feel free to add your own tests below with higher end AMD cards)

These tests are using the official 2.0.4.1701 MSIX installation vs MSI installations.

230127_2.0.4.1701_Sanded-vs-Unsanded-Benchmarks_ALL-MACHINES.zip

I tested with 3 machines using 3x benchmarks for averaging each installation method:

Machine 1 (Main):

Microsoft Windows 10 Home (Build 19045)
AMD Ryzen 7 5800X @ 3.8Ghz (-30 all core +200mhz PBO); Mobo: Asus X470 Prime Pro
32GB DDR4 (3600Mhz)
EVGA NVIDIA GeForce GTX 3080 X3C Ultra 12GB
Monitor 1&2 4K @ 150%
 

Perf Variance: 16 ~ 20%

Sanded:


Unsanded:


 

Machine 2 (Hub’s):

Microsoft Windows 10 Home (Build 19044)
AMD Ryzen 5 5600X 6 core @ 3.7Ghz
Asus ROG Strix B450-F Gaming
16GB DDR4 (3200Mhz)
Gigabyte RTX 3070 Gaming OC 8GB Rev 2.0
Monitor 1 1440p @ 225%


Perf Variance: 15 ~ 21%

Sanded:


Unsanded:

Machine 3 (Living Room):

Microsoft Windows 10 Home (Build 19045)
AMD Ryzen 7 3600 6-core @ 3.8Ghz (-30 all core +200mhz PBO)
ASRock A520M-HDV
32GB DDR4 (3000Mhz)
Asus TUF GTX 1660 TI
Monitor 1 4K @ 225%
 

Perf Variance: 12 ~ 14%

Sanded:

Unsanded:

[cgidesign]

Microsoft Windows 10 Pro 22H2
Intel 13700KF @ base 3.4 Ghz / during benchmark ~ 5 GHz (no overclock, but undervolted)
64GB DDR4 (3200Mhz)
NVIDIA GeForce RTX 3090 (undervolted and limited to max. 1800 GHz)
Monitor 2K @ 125%

Benchmark results are varying from run to run but sandboxed is always a bit slower except from "combined Multi GPU".

[winfried_z]

Also here a visible advantage of the MSI-EXE (left) version over MSIX (right).

Win11 pro, 22H2

AMD Ryzen 7 5800X 8-Core, 3.80 GHz

32GB RAM

NVIDIA GeForce RTX 3060Ti, studio driver 528.24

Display 3840 x 2160, 150%

[walt.farrell]

For me (on Win10, system specs in my signature) I'm seeing some measurements consistently better in MSI, but some consistently better in MSIX.

Three runs of each.

[debraspicher]

Thank you for these. I may come back later and reformat. But just wanted to get the variances calculated:

@cgidesign

Microsoft Windows 10 Pro 22H2
Intel 13700KF @ base 3.4 Ghz / during benchmark ~ 5 GHz (no overclock, but undervolted)
64GB DDR4 (3200Mhz)
NVIDIA GeForce RTX 3090 (undervolted and limited to max. 1800 GHz)
Monitor 2K @ 125%

Raster Single GPU: +8.8% MSI
Raster Multi-GPU: +2% MSI

Combined Single GPU: +4.8% MSI
Combined Multi-GPU: +4.5% MSIX

@winfried_z

Win11 pro, 22H2
AMD Ryzen 7 5800X 8-Core, 3.80 GHz
32GB RAM
NVIDIA GeForce RTX 3060Ti, studio driver 528.24
Display 3840 x 2160, 150%

Raster Single GPU Variance: +14% MSI
Raster Single GPU Combined: +6% MSI

@walt.farrell

Windows 10 Home, version 21H2 (19044.1706)
32GB memory
Intel Core i7-10750H @ 2.60GHz
GPU1: Intel UHD Graphics Comet Lake GT2
GPU2: NVIDIA GeForce RTX 3070 Laptop GPU

Raster Single GPU: 6% ~ +14% MSI
Raster Multi-GPU: 1.3% ~ +4% MSI

Combined Single GPU:  +10 ~ 11% MSIX
Combined Multi-GPU: +12% ~ 18.3% MSIX

[winfried_z]

Debra: Thanks for this insightful topic.

If it is of interest, here is another comparison with my rather old, but still useful equipment, MSI-EXE (left) version vs. MSIX (right):

Win10 home, 22H2

Intel(R) Core(TM) i7-6700 CPU (4 cores) @ 3.40GHz

32GB RAM

NVIDIA GeForce GTX 1080, studio driver 528.24

Display 2560 x 1440, 100%

 

Raster Single GPU: +13% MSI

Combined Single GPU: +3.6% MSI

[Nomad Raccoon]

any developer insight on this issue? Has anyone acknowledged the big performance disparity? And perhaps some insight on why there are only 1-2 cpu core threads being used by the app?

[Komatös]

19 hours ago, Nomad Raccoon said:

any developer insight on this issue? Has anyone acknowledged the big performance disparity? And perhaps some insight on why there are only 1-2 cpu core threads being used by the app?

It is not surprising that the versions that run in the sandbox are more sluggish. The sandbox is a virtual container in which the "doors" are opened and closed with every read and write operation. In addition, a "guard" checks whether the data is valid.