Nvidia.com

For a year and a half, every owner of an NVIDIA GeForce RTX 50-series graphics card ran their card without access to the single most useful diagnostic reading on the chip. The hotspot sensor — which measures the actual peak temperature on the GPU die, not the comfortable-looking average — was physically present on every Blackwell card sold since January 2025. NVIDIA's own factory technicians used it to diagnose defective hardware. Consumers could not read it at all.

That changed on July 14, 2026, when CPUID released HWMonitor 1.65. The update restored hotspot temperature readings for GeForce RTX 50 cards through an ordinary Windows application — no internal tools, no bootable Linux environments, no repair-shop credentials required. Within 72 hours, reports of alarming readings began surfacing, and by July 17, at least one board partner had officially directed customers toward after-sales warranty inspection based on the new data.

The sequence is not coincidental. NVIDIA's internal diagnostic software, called MODS (Modular Diagnostic Software), could read the hotspot sensor throughout the entire 18-month period that consumer tools could not. The asymmetry was structural: any RTX 50 card returned for warranty service would be diagnosed by NVIDIA's technicians using a sensor its owner had never been permitted to see.

What Is a Hotspot Sensor and Why Does the Delta Matter

A modern GPU die contains a distributed network of thermal sensors embedded directly into the silicon. The "GPU temperature" reading that appears in Task Manager, MSI Afterburner, or on gaming overlays is a blended average across this network — a figure that smooths out local temperature spikes by design. The hotspot sensor, also called junction temperature, reports the single highest reading from the entire network: the actual peak temperature at the most stressed point on the die.

Under normal, healthy conditions on a well-cooled card, the hotspot typically runs 12–20°C above the average GPU temperature. That delta is the key diagnostic signal: when it widens substantially — to 35°C or 40°C — it indicates that one area of the die is dramatically hotter than the surrounding silicon, usually because the thermal interface material (TIM) between the die and the heatsink is not making proper contact at that location. The RTX 5070 Ti repair case documented by Tom's Hardware confirmed exactly this failure pattern.

Thermal interface materials — typically thermal paste — can fail in two ways relevant here. The first is dry-out, where paste loses conductivity over time. The second, more immediately relevant, is pump-out: as the GPU heats and cools through repeated gaming sessions, the die package and heatsink expand and contract at slightly different rates. This mechanical cycling physically migrates paste from the center of the die toward the perimeter, eventually leaving the hottest region of the silicon nearly dry. Academic literature on TIM reliability documents this as a well-established failure mode in high-power semiconductor packaging.

When pump-out or poor initial paste application is the problem, the average GPU temperature often looks normal — the rest of the die stays cool — while the affected region hits temperatures high enough to trigger NVIDIA's thermal protection system. For Blackwell GPUs, that ceiling is 107°C.

NVIDIA's Technicians Had the Data. Consumers Did Not.

When the GeForce RTX 50 series launched on January 30, 2025, GPU-Z developer W1zzard confirmed to VideoCardz that NVIDIA had removed hotspot temperature data from the public NVAPI interface — the official Windows SDK that third-party monitoring tools use to query GPU sensors. Early builds of monitoring software initially returned a nonsensical 255°C reading before developers pulled the field entirely, recognizing that the sensor was simply inaccessible through official channels.

What the developer community did not know at the time was precisely how deep the block ran. NVIDIA had removed the hotspot sensor not just from its publicly documented NVAPI, but from the NDA-protected private NVAPI interfaces available to partner software developers as well. MSI Afterburner developer Alexey "Unwinder" Nicolaychuk explained this in a post on Guru3D forums that became the clearest public statement on the situation: NVIDIA keeps VRAM and hotspot temperature monitoring interfaces for its private internal software only, not open even to partners making NVIDIA-based graphics cards, and every tool now displaying those readings on the NVIDIA side relies on reverse-engineered private interfaces or direct GPU register access — approaches that do not make the GPU maker happy.

Because MSI Afterburner is officially licensed by NVIDIA and commercially tied to supported API interfaces, Unwinder cannot add the feature without NVIDIA's authorization. Afterburner will not display Blackwell hotspot temperatures until NVIDIA adds the sensor back to its public API — which, as of July 17, NVIDIA has not committed to doing and has not publicly addressed at all.

The tools that have bypassed the restriction are doing so through direct access to the GPU's MMIO (Memory-Mapped I/O) registers — a low-level pathway that communicates with the hardware beneath the driver layer entirely. This method works and is not obviously unstable, but it also means that a future NVIDIA driver update could close the access path, returning RTX 50 owners to the pre-July 14 diagnostic blind spot.

NVIDIA has not said whether it will restore documented API support, tolerate the current workarounds, or eventually block them.

A Repair Shop Showed What the Sensor Catches

The practical stakes of the sensor omission became concrete in early July, when Brazilian repair specialist Paulo Gomes published diagnostic footage from his repair shop using NVIDIA's MODS tool to examine an RTX 5070 Ti that a customer had brought in with performance complaints.

Under standard Windows monitoring, the card showed an average GPU temperature of roughly 67–68°C — readings that would not raise concern on any standard threshold guide. MODS revealed that the hotspot was repeatedly hitting 107°C, NVIDIA's thermal protection ceiling for Blackwell. During a single test session, the card triggered thermal throttling multiple times: the GPU's clock speed dropped to reduce heat, the temperature briefly declined, then immediately climbed back to 107°C and throttled again.

When the repair team opened the card, the cause was immediately visible. The factory-applied thermal paste had migrated outward from the die center, accumulating as a thick ring around the die's perimeter while the center — directly above the hottest portion of the silicon — was nearly dry. Replacing the paste reduced the hotspot peak to approximately 100°C, eliminated the throttling events, and restored the card's full performance.

Gomes told his audience that his shop had diagnosed several RTX 50 cards with this same symptom pattern before the MODS access became publicly discussed. Average temperatures looked fine. Hotspot was at the throttle limit. The owner had no way to know.

Three Days After Monitoring Was Restored, Colorful Directed a Customer to Service

When CPUID released HWMonitor 1.65 on July 14, it did not explain how the software bypasses NVIDIA's sensor block. The update simply works, surfacing the hotspot reading in a standard Windows monitoring app through what TechPowerUp and Hardware Busters confirmed is the same sensor that MODS had been reading internally.

HWiNFO followed with its own update restoring hotspot support for RTX 50 cards, though with a caveat: its developers identified two candidate sensors on the Blackwell die and have not yet confirmed which one represents the GPU hotspot specifically, with current theory suggesting one tracks the GPU hotspot and the other the VRM hotspot. AIDA64 beta 8.30.8337, published July 15, also added explicit Blackwell hotspot support, according to ThinkComputers.

RTX 50 owners began stress-testing their cards with these newly enabled tools over the following 72 hours, and the first publicly reported problem case emerged quickly. One owner of a Colorful iGame GeForce RTX 5080 Vulcan White found a hotspot reading near 100°C in HWMonitor and contacted Colorful's customer service through the company's Chinese-language support platform. According to screenshots of the support exchange that circulated online, Colorful's representative stated that the company considers GPU temperatures between 50°C and 90°C normal for its cards. A hotspot that remains at or above 95°C for more than 10 minutes under heavy load, the representative said, should be treated as overheating: first check case airflow and dust accumulation, and if temperatures remain elevated under good cooling conditions, apply for an after-sales inspection to rule out a cooler fault or other hardware defect.

This is the first publicly documented instance of a Blackwell board partner directing a customer to warranty service specifically on the basis of a hotspot reading that had been invisible to consumers for 18 months.

A separate case on the NGA forum involved a Colorful iGame GeForce RTX 5090 D Advanced OC that reportedly reached a 112°C hotspot with an approximately 35°C gap between the standard GPU reading and the hotspot sensor, with fans running near 3,000 RPM. A 35°C delta is substantially wider than what healthy cards have shown in early community testing, where users with properly cooled systems have reported deltas of 12–20°C.

What the Numbers Mean — and the Tool That Still Cannot Show Them

The two publicly surfaced alarm-level cases do not establish that the RTX 50 series has a broad or systematic thermal defect. Community data from early adopters using HWMonitor 1.65 shows significant variance: a water-cooled, shunt-modified RTX 5090 pulling more than 900W during a 3DMark stress run peaked at approximately 68°C hotspot with a 17°C delta — consistent with a healthy card under extreme load. Stock air-cooled RTX 5070 Ti results from the same community testing window show deltas in the 12–15°C range.

What the early reports do illustrate is precisely why hotspot monitoring exists as a distinct sensor category in the first place. The diagnostic value is not in the absolute temperature but in the delta. A wide delta on a card that otherwise appears healthy signals poor cooler-to-die contact — exactly the condition that TIM pump-out or uneven factory paste application produces, and exactly the condition that the RTX 5070 Ti repair case confirmed.

Meanwhile, the most widely used GPU monitoring tool in the enthusiast community remains unable to show this data. MSI Afterburner, which Unwinder described as restricted by its commercial relationship with NVIDIA to using only sensors exposed through supported API interfaces, has no clear path to Blackwell hotspot support unless NVIDIA acts. GPU-Z developer W1zzard has stated, according to VideoCardz reporting, that the GPU-Z team is waiting for NVIDIA's response before deciding whether to implement their own hotspot reading for RTX 50 — an understandable posture given that all current implementations are one driver update away from being broken.

A community-developed plugin, BlackwellHotspot.dll, has emerged as an unofficial extension for MSI Afterburner that reads the hotspot reading through direct register access, but it is unsupported and carries the same driver-dependency risk as every other current implementation.

An Earlier Igor's Lab Investigation Adds Context

The July 2026 reports of high hotspot deltas on the GPU die follow a separate RTX 50 thermal investigation that Igor Wallossek of Igor's Lab published in April 2025. That investigation used thermal imaging cameras to examine the power delivery systems on multiple RTX 50-series AIB partner cards from PNY, Palit, MSI, and others.

Wallossek's findings were that all RTX 50-series AIB cards exhibited high-temperature zones in the power delivery area — the region of the PCB where voltage converters (FETs, coils, and drivers) supply current to the GPU supply rails. These components are densely packed on Blackwell reference designs, and insufficient thermal cooling of the power delivery section causes localized temperatures that, in the PNY RTX 5070, reached 107.3°C in that zone while the GPU core read a much lower temperature. Wallossek demonstrated that adding thermal pads to the power delivery area reduced temperatures by approximately 10–12°C across the affected models.

This investigation is distinct from the July 2026 GPU-die hotspot reports: the Igor's Lab findings concern power delivery PCB hotspots visible via thermal camera, while the current reports concern the GPU silicon die's junction temperature sensor. Both represent thermal management concerns on Blackwell hardware, but they are different physical locations and different failure modes.

RTX 50 Owners: How to Check Your Card Now

RTX 50 owners who want to know their hotspot temperature can download one of three tools as of July 17, 2026. HWMonitor 1.65 (free, from CPUID) is the most widely tested and the first to restore the reading. HWiNFO's latest beta release also supports RTX 50 hotspot monitoring, though with the caveat that the team has not yet confirmed sensor identity. AIDA64 beta 8.30.8337, published July 15, lists Blackwell hotspot support in its release notes.

Run a sustained stress test — at least 15–20 minutes at full GPU load — and note both the standard GPU temperature and the hotspot reading. An enthusiast community baseline based on early reports suggests deltas in the 12–20°C range for healthy cards under stock air cooling. A delta above 25°C, or a sustained hotspot reading above 95°C, warrants further investigation: check case airflow first, and if temperatures remain elevated under good cooling conditions, the experience of at least one board partner suggests that contacting your card's manufacturer for warranty service may be appropriate.

What remains unanswered — and what will determine whether this story has a clean resolution or a messy one — is whether NVIDIA will restore documented API support for the hotspot sensor, tolerate the current workaround implementations, or issue a driver that closes the MMIO register pathway. The sensor exists. NVIDIA's technicians use it to diagnose the cards that get returned to them. Whether the people who paid for those cards will be able to use it too is, as of today, NVIDIA's choice to make, and NVIDIA has said nothing.

Frequently Asked Questions

Why did NVIDIA remove RTX 50 hotspot temperature monitoring in the first place?

NVIDIA has not publicly explained the decision. When the Blackwell generation launched in January 2025, the sensor was removed from the public NVAPI interface and — as later confirmed by MSI Afterburner developer Alexey Nicolaychuk — even from the private API layer available to licensed partner software developers. The sensor remained physically present on every Blackwell die and accessible to NVIDIA's internal MODS diagnostic tool throughout the entire period. NVIDIA has not responded to press inquiries asking whether the removal was intentional, architectural, or temporary.

Is my RTX 50 GPU silently throttling right now?

Possibly, and until July 14, 2026 there was no consumer-accessible way to find out. If your card has shown unexplained performance dips, higher-than-expected fan noise during gaming, or inconsistent frame rates, a faulty hotspot reading may be the cause. Download HWMonitor 1.65, HWiNFO's latest beta, or AIDA64 beta 8.30.8337, run a 15–20 minute stress test, and watch the hotspot reading alongside the standard GPU temperature. A delta above 25°C between the two readings, or a sustained hotspot above 95°C, is worth investigating further.

What does the hotspot "delta" actually indicate about my GPU?

The delta — the gap between the average GPU temperature and the hotspot reading — is primarily a proxy for cooler-to-die contact quality. A card with properly applied thermal paste and good cooler mounting contact typically shows deltas of 12–20°C under sustained load. A wider delta, particularly above 25–30°C, suggests that one area of the die is far hotter than the rest, usually because the thermal interface material has migrated away from the center of the die through pump-out or was unevenly applied at the factory. The repair case documented in July 2026 showed a roughly 40°C delta on an RTX 5070 Ti that was confirmed to have factory paste squeezed to the perimeter with the die center nearly dry.

If my hotspot is high, is it safe to repaste the card myself?

Re-pasting a GPU requires opening the card, which may void the manufacturer warranty depending on the board partner's policy. Given that at least one board partner — Colorful — is now officially directing customers to apply for after-sales inspection when hotspot temperatures exceed 95°C under good cooling conditions, owners experiencing elevated readings should first contact their board partner's warranty or support team. If the card is out of warranty, re-pasting with a quality thermal compound is a well-documented enthusiast repair procedure, but it should follow disassembly guides specific to your card's cooler design.