PerformanceTest V12 Alpha Release

[Simon (PassMark)]

We are pleased to announce the Alpha release of V12 of PerformanceTest for community testing and feedback.

Download Links [Alpha8]:
Windows x86-64: Link
Windows ARM64: Link
Linux x86-64: Link
MacOS (M-series): Link
Android: Link
iOS: Link
Linux ARM64 later on

System Requirements:
OS: Windows 10 1709+ / Ubuntu 24.04+ / MacOS 13.3+ / Android 13+ / iOS 16.3+
Memory: 4GB
Storage: 700MB available space

Please contact us to report any issues found in these builds.

Licence Requirements:
Old keys from V11 will not work in V12.
You should be able to have V11 and V12 installed at the same time (if you select different installation folders).

Free Upgrades:
When the final V12 release is complete anyone who bought V11 in the previous 12 months will get a free upgrade to V12.



What's New:

Alpha 1 [12.0.16157] - 7/July/2026
A new foundation
V12 separates the benchmarks from the user interface. This means the interface can be shared across platforms, and benchmark sets can be added or updated independently of the software itself.

This also means older benchmark sets can stay available for longer, so results stay comparable across a wider range of machines and hardware generations, rather than being tied to whatever ships with a given version of the software.

The interface has been completely redesigned and is now shared across Windows and Linux, both with the same graphical UI and the same underlying tests. A macOS version is in the works, with Android and iOS versions to follow after that.

Benchmark sets are downloadable, with more sets available from the PassMark store.

Interface
- Complete UI overhaul, shared across platforms
- Linux now has a full graphical UI, matching Windows
- Same tests run on both Windows and Linux
- macOS support in development, with Android and iOS to follow
- 'Recommendations' tab (Windows only), flags configuration issues on your machine
- 'Manage Sets' tab lists all installed benchmark sets, additional sets can be downloaded here

CPU Marks:
- Multi-threaded and single-threaded results are now split, using the same underlying tests
- Added Matrix Small, Matrix Large, JSON Parsing, and XML Parsing tests
- Split Encryption into separate AES encryption and SHA256 hashing tests, removed ECDSA encryption
- Removed Extended Instructions test

2D Mark:
- Added HTML Scroll, Video Decode, HTML WebRes, and a Javascript benchmark based on Octane
- Renamed Image Rendering to Stretch Image
- Removed Simple Vectors, Fonts and Text, Windows Interface, Direct 2D, and Direct 2D SVG tests

3D Mark:
- Added Space Battle Unity, replacing all DirectX tests
- Split GPU Compute into Mandelbrot and N Body Gravity tests
- Removed Julia and Particles sub-tests

Added NPU Mark and an optional Matrix Multiplication test

Alpha 2 [12.0.16234] - 10/July/2026
- Fixed some invalid help links
- Added NPU info to System Information tab
- Display in status bar ID of last submitted baseline, clicking on ID will open the display page
- Attempt to fix PCIe max speed if GPU is idle

Alpha 3 [12.0.16320] - 21/July/2026
- Fixed CPU information display issue on Linux
- Added inspect_cpu_info.sh to help check if cpu information is collected correctly
- Attempt to fix memory info when running PerformanceTest in VM

Alpha 4 [12.0.16439] - 31/July/2026
- Added 'warning' level to recommendations, e.g. if PerformanceTest fails to gather GPU temperatures
- Improve RAM slot recommendation

Alpha 5 [12.0.16811] - 1/September/2026
- Added MacOS version (M-series)
- Added option import existing results to be viewed within client
- Revamped user interface

Alpha 6 [12.0.17161] - 9/October/2026
- Added Android & iOS versions
- Further adjustments to user interface
- Fixed up incorrect tested disk being shown on comparison page
- Added check to prevent users from downloading/running incompatible binaries
- Fixed memory voltage readings in Linux
- Fixed the button not opening the benchmark directory location

Alpha 7 [12.0.17180] - 14/October/2026
- Consolidate disk recommendations into per-issue cards with per-disk rows
- Recommendation Tab, Fixed reporting temperatures against the wrong disk
- [Android] Fixed benchmark parse errors not appearing on the main screen
- [Android] Updated memory test to use standard error reporting system
- Fix missing QR code after upload

Alpha 8 [12.0.17315] - 4/November/2026
- Added a dark mode theme option in Settings
- Added an "Upload usage information to PassMark" option in Settings. It is off by default for registered users and always on for the shareware version
- Added crash reporting. PerformanceTest now records errors and unexpected shutdowns and can report them to PassMark. If a session crashes, the report is sent the next time you launch the app
- Added the SysInfo and DirectIO versions to the System Information tab
- Added sleep prevention on mobile during benchmark runs. If the OS locks the device anyway, the run is cancelled and you're told why
- Changed Core Benchmarks to a new benchmark set, results from the new set can't be compared directly with results from the previous set
- Changed the NPU Matrix Multiplication scoring so that mid-range NPUs score around 1000. Previously, scores were close to zero
- Changed the interface: new logo, help and settings icons, better spacing on tabs and header buttons, and restyled help buttons. The Settings window now matches the configuration window
- Changed percentile bars to show only for benchmarks that haven't run yet or are running
- Changed error notifications on mobile so they can be swiped away
- Fixed startup appearing to hang on protected builds
- Fixed some settings not being saved on Linux, Android and iOS
- Fixed the wrong error being shown when a benchmark set failed to load
- Fixed memory manufacturer names of more than one word being parsed incorrectly in the System Information tab
- Fixed cached benchmark data always being downloaded again instead of reused
- Fixed the tested drive not being identified for Disk Mark results on macOS
- Fixed swiping changing tabs in the wrong direction on mobile
- Fixed the iOS settings radio buttons and a small glitch in the iOS app icon
- Fixed the Linux dependency check so it runs during package installation

[zenmota]

Anything on what's new or fixed with V12? Also, can you install V12 with V11 installed and have or run both?

Thanks

[zenmota]

Under Recommendations, if idle it shows the GPU PCIe Interface not at Maximum (like not right). This is because the Video Card is Idle, so it drops to a lower speed. If the video card is doing something like running BurnIn Test, it shows right speed and says it is OK. It kind of make you think there is something wrong with video card when there isn't, it's just idle. Screenshots attached. Thanks...

[Harry Han (PassMark)]

Thanks for letting us know. The PCIe interface issue in Recommendations should be fixed.

[zenmota]

Yes, Alpha 2 fixed the PCIe Interface in Recommendations problem. Thanks

[daria22]

Great to hear the PCIe Interface issue is sorted in Alpha 2 — thanks for the quick fix, Harry! Looking forward to seeing how the shared Windows/Linux UI and new benchmark sets (Matrix, JSON/XML parsing, Space Battle Unity) perform once more testers get a chance to try them. Any rough timeline yet for the macOS alpha?

[zenmota]

The download link for Alpha 3 (12.0.16320) for Windows x86-64 downloads the older version Alpha 2 (12.0.16234), not the newer Alpha. Not sure about other links.

Thanks

[Simon (PassMark)]

Links fixed now

[zenmota]

I might be missing something or just not understand how it works, but after PerformanceTest V12 Alpha 4 gets through running all the Benchmarks it says Baseline uploaded successfully but there is no link, button or anything to go to a webpage or whatever to compare benchmarks.

Thanks

[Simon (PassMark)]

Was a issue with a script on the server side of things (baselines was still uploaded), fixed now.

[zenmota]

Yes, after running benchmarks it goes to webpage. I guess maybe if there was a button or link that I could click to go right to the benchmark/comparison page without having to run any benchmarks might be nice, if that makes any sense. Would make things easier.

Thanks

[Simon (PassMark)]

You can click on the PerformanceTest logo at the top which will take you to a PT index page (though maybe a button would be better)

Sidenote, you also can click on the footer area of PT where 'run id: xxx' (should appear when baseline is uploaded), which will also take you to the same page:

(need to make it more obvious)

[Doug Barrett]

Running Alpha 5, should it recognize my Hailo8 (in a Windows 11 system) as an NPU?

[Richard (PassMark)]

Running Alpha 5, should it recognize my Hailo8 (in a Windows 11 system) as an NPU?

There is no support for Hailo 8 in PerformanceTest V12.

PerformanceTest V12 will detect devices that are registered as GUID_DEVCLASS_COMPUTEACCELERATOR (typically listed as Compute Accelerator or Neural Processors in Device Manager) or NPUs that are returned from DXCore (DXCORE_ADAPTER_ATTRIBUTE_D3D12_CORE_COMPUTE). It looks like the Hailo 8 registers the device as a System Device (GUID_DEVCLASS_SYSTEM) and as such is not returned in the detection.

However, even if detected by PerformanceTest, we do not support Halio hardware currently. The NPU Test will initially only support Intel (OpenVino), AMD (XRT), Qualcomm (QNN) and Apple (ANE) NPUs.

[NightmareBlocks]

PerformanceTest V12 Linux Alpha: Drive benchmark fails with LUKS + LVM root filesystem

Summary


PerformanceTest V12.0 Build 16811 Alpha fails to run the Drive Performance benchmark on my Pop!_OS 24.04 system.

The system uses an NVMe SSD with the following storage stack:

NVMe → LUKS → LVM → ext4 → /

PerformanceTest detects the NVMe device and the LVM logical volume, but does not associate the / mount with the underlying physical disk. It then attempts to run the disk benchmark with the literal target TEMPFOLDER, which does not exist.

The CPU, GPU, and Memory benchmarks run successfully. The failure appears to be specific to Drive Performance / storage-device detection. System information


Operating system: Pop!_OS 24.04
Architecture: x86-64
Kernel:
7.1.5-76070105-generic
#202607241434~1787955023~24.04~38a419f


PerformanceTest:
V12.0 Build 16811 [Alpha]


CPU:
AMD Ryzen 7 250 w/ Radeon 780M Graphics
8 cores / 16 threads


GPU:
AMD Radeon 780M Graphics (RADV PHOENIX)
Driver Version: 26.1.6 Storage configuration


Output of:
lsblk -o NAME,TYPE,FSTYPE,SIZE,MOUNTPOINTS
NAME TYPE FSTYPE SIZE MOUNTPOINTS
loop0 loop squashfs 1.5G /snap/android-studio/247
loop1 loop squashfs 1.5G /snap/android-studio/244
loop2 loop squashfs 216.1M /snap/brave/690
loop3 loop squashfs 4K /snap/bare/5
loop4 loop squashfs 217.9M /snap/brave/692
loop5 loop squashfs 55.5M /snap/core18/2999
loop6 loop squashfs 55.5M /snap/core18/3084
loop7 loop squashfs 63.8M /snap/core20/2866
loop8 loop squashfs 63.8M /snap/core20/2922
loop9 loop squashfs 74M /snap/core22/2437
loop10 loop squashfs 74M /snap/core22/2955
loop11 loop squashfs 66.8M /snap/core24/1643
loop12 loop squashfs 66.8M /snap/core24/2124
loop13 loop squashfs 140.4M /snap/gimp/560
loop14 loop squashfs 140.9M /snap/gimp/561
loop15 loop squashfs 164.8M /snap/gnome-3-28-1804/198
loop16 loop squashfs 531.4M /snap/gnome-42-2204/247
loop17 loop squashfs 531.5M /snap/gnome-42-2204/263
loop18 loop squashfs 614.5M /snap/gnome-46-2404/164
loop19 loop squashfs 614.5M /snap/gnome-46-2404/168
loop20 loop squashfs 91.7M /snap/gtk-common-themes/1535
loop21 loop squashfs 402M /snap/mesa-2404/1781
loop22 loop squashfs 402M /snap/mesa-2404/1839
loop23 loop squashfs 1.4M /snap/nvtop/344
loop24 loop squashfs 50.3M /snap/snapd/27738
loop25 loop squashfs 44.7M /snap/snapd/28254
loop26 loop squashfs 175.5M /snap/spotify/97
loop27 loop squashfs 87.8M /snap/whattsapp/7
loop28 loop squashfs 143.8M /snap/spotify/99
zram0 disk 16G [SWAP]
nvme0n1 disk 953.9G
├─nvme0n1p1 part vfat 1022M /boot/efi
├─nvme0n1p2 part vfat 4G /recovery
├─nvme0n1p3 part crypto_LUKS 944.9G
│ └─cryptdata crypt LVM2_member 944.9G
│ └─data-root lvm ext4 944.8G /
└─nvme0n1p4 part swap 4G
└─cryptswap crypt swap 4G [SWAP]


The root filesystem is confirmed with:
findmnt /
TARGET
SOURCE FSTYPE OPTIONS
/ /dev/mapper/data-root ext4 rw,noatime,errors=remount-ro


And:
df -h /
Filesystem Size Used Avail Use% Mounted on
/dev/mapper/data-root 929G 143G 740G 17% / Relevant PerformanceTest log


PerformanceTest detects the GPU correctly:
GPU 0 found: AMD Radeon 780M Graphics (RADV PHOENIX)
Driver Version: 26.1.6
Vendor ID: 0x1002, Device ID: 0x1900
Device Type: Integrated GPU


It then encounters the storage problem:
Storage device has no model info, skipping
Storage device data-root skipped: type 'lvm' is not physical disk
Storage device cryptswap skipped: type 'crypt' is not physical disk
Storage mount / not found in lsblk output
Storage device nvme0n1p1 skipped: path '/boot/efi' is not user-writable
Storage device nvme0n1p2 skipped: path '/recovery' is not user-writable


When the Drive Performance benchmark starts, it executes:
Executing '[52] Standard Read /home/gustav/PerformanceTest_V12_crossplatform/BinaryPackages/00000036_PT_Disk_x86-64_linux/PT_DiskTest --type SeqRead --duration 5 --drive TEMPFOLDER '


This fails with:
failed to execute benchmark_id 52, name: Standard Read, error: Attempting to write to drive: "TEMPFOLDER" which does not exist


The subsequent Drive tests fail in the same way:
Executing '[53] Standard Write ... --drive TEMPFOLDER'
failed to execute benchmark_id 53, name: Standard Write, error: Attempting to write to drive: "TEMPFOLDER" which does not exist

Executing '[54] IOPS 32KQD20 ... --drive TEMPFOLDER'
failed to execute benchmark_id 54, name: IOPS 32KQD20, error: Attempting to write to drive: "TEMPFOLDER" which does not exist

Executing '[55] IOPS 4KQD1 ... --drive TEMPFOLDER'
failed to execute benchmark_id 55, name: IOPS 4KQD1, error: Attempting to write to drive: "TEMPFOLDER" which does not exist Expected behavior


PerformanceTest should detect the writable root filesystem at /dev/mapper/data-root, or otherwise correctly resolve the underlying physical NVMe device and provide a valid benchmark target.

If this storage configuration is unsupported, PerformanceTest should ideally report that the Drive benchmark cannot be performed rather than attempting to execute PT_DiskTest with the invalid target TEMPFOLDER. Actual behavior


PerformanceTest does not recognize / as a benchmarkable storage target.

It then invokes PT_DiskTest with:
--drive TEMPFOLDER


which does not exist.

This causes the Drive Performance tests to fail and prevents completion of the full benchmark run. Reproduction

1. Install/run PerformanceTest V12.0 Build 16811 Alpha on Pop!_OS 24.04.
2. Use a system whose root filesystem is an ext4 logical volume inside LVM, with the LVM physical volume inside a LUKS-encrypted NVMe partition.
3. Start the full benchmark / Drive Performance benchmark.
4. PerformanceTest fails to resolve / to a physical storage device.
5. The Drive benchmark is invoked with --drive TEMPFOLDER.
6. All Drive tests fail because the target does not exist.

Additional information


The relevant log file is:
~/.local/share/PassMark/PerformanceTest12/logs/PerfTestLog.txt


The issue is reproducible. It occurred across multiple runs, with the same TEMPFOLDER target appearing each time.

The CPU, GPU, and Memory benchmark components otherwise detect the hardware successfully. Possible area to investigate


The storage-device discovery appears to reject:
data-root
type: lvm


because it is not considered a physical disk, while the physical NVMe device is not successfully associated with the / mount.

The following message may also be relevant:
Storage device has no model info, skipping


It may be worth checking how the Linux storage detection handles the relationship between:
/dev/nvme0n1
↓
/dev/nvme0n1p3
↓
LUKS / cryptdata
↓
LVM / data-root
↓
ext4 /


and what value is substituted as TEMPFOLDER when no valid benchmark target is found. Requested outcome


Please investigate the Drive benchmark's storage-device detection for LUKS + LVM configurations and prevent PT_DiskTest from being invoked with the invalid TEMPFOLDER target.

I would also appreciate confirmation of whether this storage configuration is currently supported by the Linux V12 Alpha and, if so, whether there is a recommended workaround for benchmarking the NVMe drive.