A living document detailing the current and retired test platforms used by GamersNexus

The Highlights

  • This document will be updated as we add new benches
  • Old test benches will be kept on this page, but moved to the bottom
  • Remember that testing regularly requires changes to test platforms to accommodate the DUT

Table of Contents

  • AutoTOC
Visit our Patreon page to contribute a few dollars toward this website's operation (or consider a direct donation or buying something from our GN Store!) Additionally, when you purchase through links to retailers on our site, we may earn a small affiliate commission.

Intro

This page is a living document that will be updated as we revolve test platforms through. The page will be organized by type of test bench and year. We regularly cycle benches into retirement as we update for new components. We have had numerous manufacturers inform us that they try to keep tabs on our bench platforms for their own internal research and performance analysis, so we hope that this may help manufacturers with their product development when attempting to isolate performance concerns we present.

And to reiterate: This page is NOT COMPLETE. This is under construction. There is a lot of information still missing here.


Credits


Test Lead, Writing

Steve Burke

Writing

Patrick Lathan


Steve & Stone working side-by-side at the GPU test benches. We have two benches simultaneously deployed for rapid bring-up and testing. They are 100% mirrored and are calibrated against each other for every single review to ensure consistent data between machines, which also serves as an error check by duplicating data in separate environments. In this photo, Stone is handling GPU power transient testing.

Objectives

This page will primarily serve as a landing page for test benches that we commonly use. This page is still being built and is under construction. The objective is to provide a single stop to answer questions about what hardware we use.

Our philosophy is to review each component in a vacuum, eliminating as many variables and bottlenecks imposed by other components as possible. This means that GPU testing for GPU performance will use a relatively high-powered CPU and RAM, thus hopefully limiting bottlenecks (to the extent possible) outside of the GPU; however, for that same GPU review, we would use a different test bench for acoustics. That's because the acoustics testing does not require unbound performance to eliminate variables, but instead requires a total (within reason) elimination of noise from non-GPU components instead. In this example, we would use the GPU test bench for thermals and gaming, but the GPU acoustic bench for noise levels and frequency response. Likewise, we use a different bench entirely for GPU power consumption measurements, as the host platform is irrelevant for these and only the power consumption matters (and it allows us to asynchronously test power consumption while our primary benches remain occupied for testing gaming or thermals, which greatly improves logistics to allow more testing in a shorter time period).

A lot of this has to do with a general internal approach of specializing within the team. Typically, the same person will handle GPU power consumption tests for every review, but a different person will handle GPU gaming performance and analysis, and a different person still might handle acoustics. Splitting benches like this not only isolates for variables, but also allows the team's experts in each test method to operate without blocking their teammates, as the benches are not singularly needed for every test in a suite.

Caveats & Limitations

We sometimes have to change-out single components (or revert to older benches) for certain tests. For example, if we are running a suite of CPU tests that requires us to go back to AM4 or Z270 or similar, we obviously won't be using the standardized Intel or AMD platform listed as the current primary, but would instead use one of the prior platforms. We will try to keep the older (retired) platforms listed further down the page for such events, but it will take us some time to comb through the site history and catalog them all in this format.

These test benches might not perfectly represent each individual content piece. Some pieces require a singular variable change to these benches as part of the piece. For example, if we're testing something like Intel APO, we will use a different boot drive (to avoid cross-contamination with the primary bench drive) and will use different driver packages and a potentially different BIOS configuration (to enable the dynamic tuning). Likewise, if we're testing PCIe bandwidth scaling and impact on an NVIDIA *90-class card, we will change the PCIe slot's generation for that test. The DUT (Device Under Test) or feature under test will deviate from the fixed bench as part of the test.

Sometimes we miss listing a detail since there are so many, but we will do our best to explain these caveats in each individual content piece.

Consistency is What Matters

Consistency is all that really matters. For comparative reviews, which is how we conduct our reviews, we just need an unchanging and fixed platform that allows us to reliably see scaling and differences between components. In the above example then, having APO toggling or PCIe generation switches being the new variable on an otherwise unchanging platform allows us the most comparative data against our "baseline" for that platform.

Likewise, this means that components being "old" or back by a couple generations becomes largely irrelevant. It only starts mattering if it's meaningfully limiting either the performance scaling of the product or the technology that can be used for the product. For instance, for purposes of generating heat for a cooler review, the CPU itself doesn't matter as much as its power consumption (and IHS shape, to be fair).

Methodology Pieces

This contains a list of our recently published or current test methodologies. Note that we have not yet gone back to upfit all of the older pieces for this website revision, so some links are to the videos.

This section is under construction. We still have to build this section out fully. It is missing a lot of information.

Recent Game Test Methodologies & Research
Hardware Test Methodologies & Research

Test Benches

Mini-ITX Case Variable Test Bench (2023-TBD)

PartStandard Component(s)Provided By
CPUIntel i5-13600K
@ Fixed clocks & voltages (clocks/voltages determined per case; we may settle on permanent settings once we get past our first few ITX case reviews)
Bought by GN
GPUNVIDIA RTX 4070 FENVIDIA (review unit)
MotherboardASUS ROG Strix Z790-I Gaming WiFiBought by GN
PSUCorsair SF1000L 1000W SFX-L
Phanteks Revolt SFX 850W
Bought by GN
Provided by Phanteks
RAMGSkill Ripjaws S5 2 x 16GBBought by GN
SSDSamsung 960 Pro M.2 NVMe SSDBought by GN
CoolersVARIABLE
Standardized options:

240mm CLC: Fractal Design Lumen S24 v2 RGB
120mm CLC: Arctic Liquid Freezer II 120
"Standard" Tower: Scythe FUMA 3 or Deepcool AK400
Short Single Tower: Noctua NH-U9S
Mini-Tower: Noctua NH-D9L
Downdraft Low-Profile: Deepcool AN600 67mm
Alt. Downdraft: Noctua NH-C14S
Fractal Design, Arctic, Deepcool
Additional Fans (Optional)VARIABLE
Standardized options:

Arctic P14 Slim PWM PST 140mm
Arctic P12 Slim PWM PST 120mm
Arctic P8 Max 80mm
Noctua NF-A6x25 PWM 60mm
Noctua NF-A9x14 PWM 92mm
Arctic
Noctua

GPU Acoustics Test Bench (2023-TBD)

The below test bench details our platform used for GPU acoustics. This is deployed for collecting data such as the below. The GPU Acoustics Bench is not used for performance analysis (other than acoustic performance), and so the core components do not matter. For this bench, the most important aspect is controlling the noise: We use passive cooling, like from the Noctua NH-P1, to ensure no noise interference from other parts of the system. The power supply is used in a 0-RPM mode and validated during testing. The GPU does not perform work, but rather has its fan speed manually adjusted to meet the automatic fan speed we find during load testing (from our GPU FPS & thermal test platform, also on this page). We also adjust the speed in increments of 5% to capture a dBA profile across the RPM range.

Unless otherwise stated, video card coolers are tested at an industry-standard distance of 1m from the center fan (aligned on the horizontal and vertical center of the video card cooler). The video card is situated in a horizontal Thermaltake Core P3 platform with the glass removed, so it is basically just a flat "case."

PartComponentProvided By
CPUIntel i7-8086K CPUBought by GN
MotherboardASUS Maximus X Hero Wi-Fi ACASUS
RAMCrucial Ballistix 2x8GB DDR4-3200 16-18-18-36Crucial
CoolerNoctua NH-P1 Passive CoolerNoctua
PSUCorsair HX1500i PSU with Passive Fan ModeCorsair
OSWindows 11 ProBought by GN
CaseThermaltake Core P3 in Horizontal 'Mode'Thermaltake

GPU Test Bench (2022-2023)

The below test bench details our two GPU review platforms that are used for performance evaluation and thermals. The GPU test benches are overclocked, which helps significantly limit the bottlenecks potentially imposed by a CPU. The platforms also run tests at 1080p, 1440p, and 4K, with the higher resolutions further distancing us from CPU scaling limitations. Presently, our platform is primarily GPU-bound, but 1080p is starting to become constrained for high-end cards. This platform will be retired from service at the end of 2023 or early 2024.

PartComponentProvided By
CPUIntel Core i7-12700KF Overclocked
(4.9GHz P-Cores, 3.9GHz E-Cores)
Bought by GN
MotherboardMSI Z690 UnifyMSI
RAMDDR5-6000 G.Skill Trident Z (manually tightened timings)G.Skill
CoolerArctic Liquid Freezer II 360 @ 100% Fan SpeedBought by GN
PSUEVGA 1600W T2 Supernova
Corsair AX1600i
EVGA
Corsair
OSWindows 11Bought by GN
Additional parameters include: Hardware-accelerated GPU scheduling and ReBAR both enabled. Power plan set to High Performance. Note: Acoustic testing uses a bench with 0 fans, so passive PSU + coolers.

Test Applications & Games

We are constantly changing the games that get tested on this platform, and as such, maintaining a list of what we currently show is difficult. Instead, we'll list everything that the platform runs as of November 2, 2023. This list will likely be moved to the GPU methodology piece once we get it into the new website.

GameResolution & SettingsResearch & Methods Piece
Starfield1080p/Low
1080p/Medium
1080p/High
1440p/High
4K/High
4K/Ultra
* Dynamic resolution, upscaling, and inherently variable options are disabled unless otherwise noted, such as VRS.
- Starfield GPU Benchmarks & Research (Video)
- Starfield CPU Benchmarks & Research (Video)
- Starfield Graphics Optimization Guide (Video)
Cities Skylines 21080p/Very Low
1080p/Low
1080p/Medium
1080p/High
1440p/Low
4K/Medium
4K/High
* Dynamic resolution, upscaling, and inherently varaible options are disabled unless otherwise noted.
- Cities Skylines 2 GPU Benchmarks & Research (Video)
Baldur's Gate 31080p/Medium
1080p/Ultra
1440p/Ultra
4K/Ultra
Baldur's Gate 3 GPU Benchmarks & Research:
- Article
- Video
Total War: Warhammer 31080p/High-Custom
1440p/High-Custom
4K/High-Custom
Tom Clancy's Rainbow Six: Siege1080p/Ultra
1440p/Ultra
4K/Ultra
(Dx11)
Strange Brigade (Vulkan)1080p/Ultra
1440p/Ultra
4K/Ultra
Strange Brigade (Dx12)1080p/Ultra
1440p/Ultra
4K/Ultra
Shadow of the Tomb Raider1080p/High
1440p/High
4K/High
Resident Evil 41080p/Prioritize GFX
1440p/Prioritize GFX
4K/Prioritize GFX
Horizon Zero Dawn1080p/Quality
1440p/Quality
4K/Quality
GTA V4K/Custom (mix of Very High & Ultra)GTA V Testing Mystery
- Video
GTA V Graphics Optimization Guide:
- Article
- Video
FFXIV: Endwalker1080p/Maximum
1440p/Maximum
4K/Maximum
F1 20221080p/Ultra-Custom
1440p/Ultra-Custom
4K/Ultra-Custom
Dying Light 21080p/High-Custom
1440p/High-Custom
4K/High-Custom
Cyberpunk 2077: Phantom Liberty(Pending)
(Ray Tracing) Shadow of the Tomb Raider1080p/Highest (RTX)
1440p/Highest (RTX)
4K/Highest (RTX)
(Ray Tracing) F1 20221080p/Ultra-Custom + RT
1440p/Ultra-Custom + RT
4K/Ultra-Custom + RT
(Ray Tracing) Cyberpunk 2077: Phantom Liberty(Pending)
Bolded settings are those which are run regularly. Unbolded settings are those which have been tested or for which we have an internally defined SOP (Standard Operating Procedure), but that we might not run once the initial game research is done. If no bold options are shown, all listed are used. Dynamic resolution, upscaling, and inherently varaible options are disabled unless otherwise noted.

Additional tests are run, but are not explicitly listed as they may be more variable (or we just haven't publicly documented them at this time).

CPU Cooler Test Bench (2020-2024)

PartComponentProvided By
CPUAMD Ryzen 5 3600 - Used for lower ~68W heat loads for small coolers.
AMD Ryzen 7 3800X (2 chiplets active) - Used in all cooler benchmarks.
AMD Ryzen 9 3950X (3 chiplets active) - Used for higher heat loads to show scaling on big coolers.
AMD
MotherboardMSI X570 MEG ACEGN Purchase
RAMGSkill Trident Z Royal DDR4-3600 CL16GSkill
GPUEVGA NVIDIA GeForce GT 1030 SC (passive cooler, second slot down)GN Purchase
PSUEVGA 1600W P2EVGA
OSWindows 10GN Purchase
We have published a full, in-depth Test Methodology piece about our CPU cooler benchmarks and reviews. To learn how this testing is done, check that link!

CPU Test Bench (2024-TBD)

We are still working to fully add our modernized testing methodology for CPU benchmarks and reviews. Often, the review itself will contain verbal explanations of each test and methodological notes. We are slowly compiling those as we continue to build-out the living doc on the website. The below tests are typically run on every CPU; however, we sometimes remove tests for reasons that could relate to time with the product before embargo lift, compatibility, performance being too low (e.g. a CPU that struggles to maintain 30 FPS might just see that test eliminated), game updates that invalidate prior data (and need to be recollected), or if we believe the charts in a review are enough to illustrate the pattern without lengthening the review further.

Our CPU tests include these games and settings:

Production Workloads
ApplicationUnits & Direction of ImprovementReason
Cinebench R23 NTCB Marks. Higher is betterUsed for external validation and cross-reference to ensure the CPU is functioning as intended
Cinebench R23 1TCB Marks. Higher is betterUsed for external validation and cross-reference to ensure the CPU is functioning as intended
7-Zip CompressionMIPS. Higher is betterEvaluates performance when working under prolonged compression tests, represented as Millions of Instructions Per Second (MIPS)
7-Zip DecompressionMIPS. Higher is betterAs above, but decompression
Chromium Code CompileMinutes. Lower is betterCode compiles cannot be represented by a single test. For a number of reasons, mainly being practical, we only currently test compilation speed with Chromium. This test helps show longer workload performance, which can often shuffle the stack based on boosting behavior. Shows up most on CPUs such as those with Tau limits (like older Intel CPUs)
Blender 3DMinutes. Lower is betterEvaluates heavy all-core loading performance where clocks tend to be at their lowest, with power and thermal at its highest. Prolonged test helps show real-world fall-off from thermal & power (if limits exist)
Adobe PhotoshopAggregate points (Puget)Evaluates a mix of higher single-core or limited-thread count performance. This shows best on CPUs with higher IPC or single-core performance. Mixes multi-threaded workloads via Puget Suite to create an aggregate score.
Adobe PremiereAggregate points (Puget)Evaluates functions such as filters, time warps, transforms, scales, and some rendering capabilities. Newer versions of this test have eliminated scrubbing due to variance, per Puget.
SpecWSAggregate points (Spec)This is experimental for GN's testing right now, meaning we are still vetting it. Currently, we use SpecWS for CFD testing (Computational Fluid Dynamics) on CPUs via Rodinia CFD and biomedical modeling via the LAMMPS test.
Gaming Workloads
GameYear ReleasedResolution & Settings
Final Fantasy XIV: Dawntrail20241080p/Maximum
1440p/Maximum
Dragon's Dogma 220241080p/High
F1 2420241080p/High
1440p/High
Baldur's Gate 320231080p/Medium
Starfield20231080p/Low
* Dynamic resolution, upscaling, and inherently variable options are disabled unless otherwise noted, such as VRS.
Cyberpunk: Phantom Liberty20231080p/Medium
Total War: Warhammer III20221080p/Medium
* Battle benchmark
Stellaris20161080p/High
* Simulation time (seconds)
Rainbow Six Siege20151080p/Very High
1440p/Very High
CPU Miscellaneous Tests
SoftwareTests
Blender + HWINFOFrequency
Thermal
Power
Cinebench 1T + HWINFOFrequency
External Validation
Blender + EPS12V MonitoringPower
Efficiency
Limited Games SuiteEfficiency

We will add the test bench hardware information for the next round of reviews, but the vast majority of critical information is listed in the chart subtitles.

Check back soon: We still need to add our case bench, our new cooler bench, and our CPU benches, among other platforms!