A Windows tuning and monitoring tool for NVIDIA, AMD and Intel Arc GPUs. Adjust supported clocks, voltage, power, fans and memory timings, with live telemetry, five profile slots and light/dark modes. No HWiNFO dependency.
Intel IGCL support adds core offsets, memory speed, voltage/power/temperature limits, editable V/F curves, fixed fan duty and hardware fan curves. Controls use driver-reported ranges. Memory tuning is shown in Mbps (19,000 Mbps = 19 Gbps); Intel's temperature limit is a percentage, not °C. Existing NVIDIA and AMD controls keep their units. Intel editor increments are 5% voltage, 15 MHz core offset, 100 Mbps memory speed, and 5% power/temperature limits. Loading existing settings preserves off-grid values.
Verified on Arc Pro B60 24 GB / driver 32.0.101.9030 with small write/readback/restore tests, including fan modes and profile Apply. The original settings and curves were restored. This is functional verification, not a stress-stability result. See Intel support and validation. Intel support introduced in the 1.0.9 source is included in 1.1.0. Build this checkout or download the matching 1.1.0 executable from Releases.
- Download
RochGPU.exefrom Releases. - Put it in a folder you own and run it. Allow administrator access when requested.
- Open Telemetry to monitor the card, or edit a setting and press Apply.
Windows x64 and a compatible vendor graphics driver are required. The self-contained executable needs no separate .NET installation.
- Intel Arc Pro B60 tuning: voltage limit, core offset, memory speed, power/temperature limits, V/F editing, and fixed or hardware-curve fans through the installed Intel driver.
- Core clock range: optional minimum/maximum limits, profile persistence, verified writes, and restoration of the driver's factory range.
- Fan settings stay selected: applying power or clocks keeps fan changes made in the Fan window. Software fan-curve updates are serialized with fan-mode changes.
- Expanded Battlemage telemetry: VR temperatures, memory voltage monitoring, GPU/memory power, memory bandwidth, VRAM usage, PCIe link and throttle indicators when reported. Memory voltage is read-only.
- Graphics details fixed: ASRock board identity, Battlemage chip, Xe cores, GDDR6/bus width, VBIOS and PCI address; matching Intel driver date; scrollable details.
- Review fixes: Intel stock profiles use the driver's voltage-limit default. UI reads and voltage-cap edits share the telemetry lock. A stalled poll cannot silently unload its driver library. Vendor DLL discovery is restricted to System32.
- Build checks: both
build.ps1and setup run core, telemetry and app regression suites before publishing.
The supported Intel editor steps are 5% voltage, 15 MHz core, 100 Mbps memory, 5% power and 5% temperature. Existing off-grid settings remain intact when loaded. Temperature percentage is not degrees Celsius.
See 1.0.9 changes and validation. B60 functional tests verify requests/readback/restoration; they do not establish stress stability or support on every Arc model. Earlier NVIDIA fixes are documented in 1.0.8 notes.
Compatibility depends on the GPU, board firmware and installed driver; an available control is not a guarantee that every card supports it.
Version 1.0.7 keeps disabled NVVDD and MSVDD rail limits under the NVIDIA driver's control, so Voltage Boost can update both live ceilings by the amount reported for that specific card. It does not hard-code the 20 mV behavior observed on the development RTX 5070 Ti. Version 1.0.6 introduced the NVIDIA OCP compatibility fix below.
| GPU / driver | Verification status |
|---|---|
| GeForce RTX 5070 Ti — 616.92 | Power, core and memory offsets passed write/read-back/restore tests. Both NVVDD and MSVDD OCP rails passed a scoped hardware test; the updated app was subsequently confirmed working by its owner, including OCP. |
| NVIDIA 616.56 | Original reported failure version. The modern OCP path is implemented, but this exact driver has not been tested directly. |
| Other NVIDIA drivers / GPUs | Feature-dependent; not covered by the 616.92 test. Older drivers retain the legacy OCP path. R615+ changed-limit requests require validation of the modern OCP layout; unknown driver identity blocks changed OCP writes. |
| Intel Arc Pro B60 — 32.0.101.9030 | Native tuning, hardware fan curves, V/F editing and core clock range passed scoped write/readback/restore checks. Latest identity and telemetry checks are read-only. Other Arc models are not hardware-validated. |
| AMD | Uses ADL/ADLX rather than NVIDIA's OCP path. See RX 9070 XT support for verified features; this NVIDIA fix does not establish additional AMD driver compatibility. |
Newer drivers can accept the old OCP read structure while rejecting writes using that structure. Roch GPU now uses the modern control layout for changed OCP limits on R615+, validates the reported rail types and bounds, and checks both the requested rail and the untouched rail after writing. Matching live limits are accepted without unnecessary writes; a failed modern write is not retried with the legacy layout.
On the tested RTX 5070 Ti / 616.92, NVVDD 300 → 290 → 300 A and MSVDD 120 → 110 → 120 A were verified. These were temporary reductions followed by restoration—not tests of above-default current limits. Higher OCP limits reduce protection headroom; driver-reported bounds are not safe tuning recommendations. Interrupted-write recovery and all possible settings are not exhaustively validated.
After replacing the executable or updating the driver, fully exit the previous Roch GPU window and tray instance before opening the new build. If Apply fails, retain the recovery journal and collect diagnostics with RochGPU.exe diag, along with the GPU model, driver, vBIOS and app version. See driver compatibility details.
- NVIDIA, AMD and Intel Arc in one app: supported core/memory clocks, voltage controls and power limits, with controls adapted to the detected card.
- Live telemetry: temperatures, voltages, clocks, utilization, power, fans, GPU memory and PCIe readings where available. Summary cards and current/minimum/maximum/average columns make changes easy to track.
- Hardware-read GPU identity: the main header shows driver and vBIOS versions, VRAM type/vendor, maximum PCIe generation and lane width, and Resizable BAR state. The read-only Graphics window adds board, silicon, memory and installed-driver details.
- Native AMD monitoring: ADL and ADLX read the installed AMD driver directly—no HWiNFO, shared-memory feed or extra monitoring driver.
- Fan control: manual speed and fan curves; AMD Zero RPM lives in the Fan window. NVIDIA software curves require the app to keep running.
- Five saved profiles: save tuning setups and choose a startup profile. Startup uses a frozen copy; toggle Startup off/on after saving changes to refresh it.
- NVIDIA clock and voltage controls: V/F Curve, XOC and Fan editors sit together above the main controls. Supported XBAR, SYS and video offsets appear directly below Core Offset and apply without separate enable switches; potentially damaging rail, OCP and clock-range controls remain individually gated in XOC.
- AMD memory timings: select the modes your driver exposes, alongside voltage offset and memory-clock tuning. Unsupported timing modes are not invented.
- Light/dark interface and CLI: grouped tuning controls, a separate telemetry window, diagnostics and command-line monitoring/tuning from the same executable.
- Social links: YouTube | X | Discord in the bottom-left footer, matching Roch CPU and Viewer.
- Profile recovery: full-profile applies keep a recovery journal and attempt rollback after errors or interrupted writes. There is no Keep/Revert countdown, and recovery is not a stability test.
Verified native readings include GPU/hotspot/memory temperatures, core voltage, core/memory clocks, board power, fans, utilization, dedicated/shared GPU memory and PCIe link information. FCLK and SoC rows require driver support; the tested RX 9070 XT does not expose them through the public interfaces. Effective core clock and effective FCLK are not implemented. See native AMD telemetry.
AMD's memory-clock slider sets the maximum memory clock, with limits supplied by the installed driver. On the tested RX 9070 XT, the driver reports 2650–3000 MHz with a 2650 MHz default. This lower slider limit is not the minimum live memory clock; ranges can differ between cards and drivers.
| NVIDIA tuning | AMD RX 9070 XT tuning | Telemetry |
|---|---|---|
 |
 |
 |
The AMD tuning and telemetry screenshots show Roch GPU 1.0.4 on an RX 9070 XT. NVIDIA, fan and V/F editor screenshots are from earlier builds. Features depend on the GPU; displayed settings are examples, not tuning recommendations.
Install the .NET 10 SDK, then run in PowerShell:
.\build.ps1Open dist\RochGPU.exe. The current source version is 1.1.0. The build runs all three regression suites before publishing.
Overclocking can cause crashes, data loss or hardware damage. Test changes carefully. Controls and sensor readings depend on what your driver exposes.
- Soroush Falahati — NvAPIWrapper: the vendored NVAPI binding, modified and retargeted to .NET 10.
- dumbie — RadeonTuner: the Overdrive 8 calling-convention reference that helped unblock the AMD backend; no RadeonTuner code is included.
- AMD ADL and ADLX: public driver-interface documentation for AMD tuning and telemetry.
- Intel Graphics Control Library (IGCL): public interfaces used by the independent Intel backend; only the installed driver DLL is loaded. No Arc Power source or binaries are bundled.
- NVIDIA NVAPI and NVML: vendor driver interfaces used for NVIDIA tuning and monitoring.
Created by Roch Studio / @MateoPCTech. Licensed GPL-3.0-or-later; third-party components retain their own licenses. See third-party notices.
Detailed reference · Native AMD telemetry · Profile recovery · License