RTX 3070 Paste Swap Drops VRAM 24C And Core 13C On Aging Card

By Indie Kings | October 3, 2026

Updated October 3, 2026: A Reddit user report relayed by PC Central says VRAM fell from 99C to 75C and the core fell from 85C to 72C on one Gigabyte RTX 3070 Eagle OC V2. The core to hotspot delta narrowing from 21C to 11C is presented here as a reading that points to contact failure, with the single card and single game limit kept in view.

From 99C to 75C on the Memory

The subject is the Gigabyte GeForce RTX 3070 Eagle OC V2. That full card name matters because cooler designs differ by brand and revision.

The report comes from an unnamed Reddit user and reaches readers via PC Central. PC Central published the account on Sep 21, 2026 23:06 UTC under the title RTX 3070 VRAM Temp Drops 24C After Thermal Paste Swap.

PC Central attributes the piece to Marek Novák. PC Central also supplies its own summary line for the event.

PC Central summarizes the case as a Reddit user replacing thermal paste on a Gigabyte RTX 3070 Eagle OC V2, dropping VRAM temps by 24C and core temps by 13C using phase change pads. That summary is PC Central language and it sets the scope.

The stated goal was to address excessive fan noise at 100 percent speed due to high operating temperatures. The fan report is tied to heat in the source account.

The Reddit user reported fan noise became manageable after temperatures stabilized. That outcome is a user report relayed by PC Central.

The modification targets both the GPU core and the surrounding memory modules. Power delivery modules were also included in the material swap.

The user replaced aged paste with Thermalright PhaseSheet PTM on the GPU core. That placement is specific in the PC Central account.

They applied Thermalright Putty Advance to the VRAM and power delivery modules. That second placement is also specific in the PC Central account.

Testing was conducted while running Overwatch 2. No other workload is named in the PC Central report.

PC Central does not specify if other titles were benchmarked. That limit is stated plainly here because it shapes what can be claimed.

PC Central says the results were immediate and significant across all measured temperature points. That characterization belongs to PC Central.

PC Central frames degraded thermal paste as a silent killer for older graphics cards, often causing fans to scream and performance to throttle. That framing is PC Central language and it is attributed here rather than adopted.

PC Central says swapping materials can restore factory like cooling efficiency. That statement is attributed to PC Central and it is not repeated here as advice.

PC Central says upgrading the thermal interface offers a practical way to extend the life of existing hardware without buying new components. That view is attributed to PC Central and it is not adopted here as a recommendation.

The measurement block is narrow and it is worth quoting in full before interpretation begins. Each number below comes from the PC Central report of one Reddit user test.

VRAM reduction was 24C, from 99C to 75C. Core reduction was 13C, from 85C to 72C.

Hotspot reduction was 23C, from 106C to 83C. Core to hotspot delta moved to 11C after being 21C before.

Each figure describes a single test on one card in Overwatch 2. No broader sample is stated by PC Central.

The table below restates the measurements with that standing attached to every row. It exists so a reader can check the numbers without relying on prose alone.

The pictured card in this article is an MSI RTX 3070 Ventus 3X OC, which is an RTX 3070 but a different brand and cooler from the tested Gigabyte Eagle OC V2. A different cooler is exactly the variable that decides whether stock interfaces work, so the photo cannot show the tested contact or the tested result.

Image: MSI GeForce RTX 3070 Ventus 3X OC. Credit: Wikimedia Commons, CC BY 4.0, TheStriker.

The card under test uses NVIDIA silicon in a Gigabyte board and cooler assembly. The photo shows different board partner hardware around similar NVIDIA silicon.

That distinction matters for the rest of this article. Cooler mounting pressure and pad contact vary by design.

The Delta Is the Diagnosis

The most useful number may not be the 24C VRAM drop. It may be the delta change from 21C to 11C.

Core to hotspot delta describes the gap between average core reading and the hottest reported point. A wide gap can point to uneven contact.

Before the swap, the gap was 21C. After the swap, the gap was 11C.

That narrowing is read here as a sign of contact failure rather than cooler failure. The reading is an interpretation of one reported test and it is not proof.

A cooler failure reading would look different in principle. It would often leave temperatures high but more uniform if heat could not leave the heatsink.

A contact failure reading looks like localized heat that improves once gaps are filled. The reported pattern fits that second shape.

The core fell from 85C to 72C while the hotspot fell from 106C to 83C. The hotspot fell further than the core average.

That uneven improvement is why the delta gets attention. Large drops at the hottest point can indicate restored transfer at that point.

PC Central says these materials fill microscopic gaps better than standard paste and eliminate the pump out effect. That material claim comes from PC Central.

PC Central says phase change pads provide superior thermal transfer compared to traditional pastes. That comparison also comes from PC Central.

The delta reading does not prove the cooler was adequate in every case. It only describes what changed on this one card in this one test.

One card cannot establish a rule for all RTX 3070 cards. One run in Overwatch 2 cannot establish behavior in other workloads.

The value of the delta is therefore diagnostic rather than general. It gives a way to read this report with care.

A reader checking a similar card would need to check their own card contact first. No remote reading can replace that direct check.

PC Central says this material combination appears to offer a reliable solution for cards suffering thermal degradation. That assessment is attributed to PC Central.

This article does not adopt that assessment as advice. It records the assessment as part of the source record.

The delta from 21C to 11C remains the cleanest single line summary of restored contact in this account. It is still only one measurement pair from one user.

Future reports with repeated runs would be needed to test whether the reading holds. No such repeats are stated in the PC Central account.

The diagnosis stays provisional for that reason. Contact failure is the working reading, not a proven cause.

What Went Where

The material split is explicit in the PC Central report. Core and memory areas received different materials.

The GPU core received Thermalright PhaseSheet PTM. That is stated as used by the Reddit user.

The VRAM modules received Thermalright Putty Advance. That is also stated as used by the Reddit user.

The power delivery modules received Thermalright Putty Advance. That third placement completes the reported swap.

Paste and pads are not interchangeable. That caution is important because the two product types serve different gaps and surfaces.

This article gives no step by step for applying either material. No placement sequence is described beyond the location list.

PC Central says these materials fill microscopic gaps better than standard paste and eliminate the pump out effect. The claim covers gap filling and material stability.

Pump out refers in PC Central language to loss of interface material over thermal cycles. PC Central says the chosen materials avoid that effect.

PC Central says phase change pads provide superior thermal transfer compared to traditional pastes. That statement is recorded as a source claim.

The table below maps each location to its reported material. Standing is included so the table cannot be mistaken for a general guide.

The table names Thermalright products because the source names them. Naming is not an endorsement.

The table names Reddit because Reddit is the stated origin of the user report. No thread is linked and no thread is described beyond that origin.

The table names Overwatch 2 because Overwatch 2 is the only stated workload. Other titles are listed as not specified.

No thickness is stated in the source for the putty. No quantity is stated in the source for either material.

No source statement supports generalizing this exact split to other boards. Different boards use different stack heights and contact layouts.

A reader should check their own card contact first before drawing any parallel. Visual inspection by the owner is the only relevant check for that owner.

The material story is therefore narrow by design. It answers what went where on this card and nothing more.

That narrow scope is a strength for accuracy. It avoids turning one build into a universal formula.

Overwatch 2 and Nothing Else

Testing was conducted while running Overwatch 2. That sentence carries the full workload description available from PC Central.

PC Central does not specify if other titles were benchmarked. That absence is stated plainly and it limits interpretation.

A single title means a single load shape. Different titles stress core and memory in different proportions.

Overwatch 2 is a live service shooter with variable scene load. That general description does not add test detail beyond the title name.

No resolution is stated. No settings preset is stated.

No frame rate before or after is stated. No clock speed before or after is stated.

No test duration is stated. No ambient temperature is stated.

No fan curve is stated. The only fan detail is 100 percent speed from heat before and manageable noise after stabilization.

That fan detail comes as a user report via PC Central. It is qualitative rather than logged.

A single run also leaves run to run variation unmeasured. Repeated runs would be needed to show stability of the new numbers.

The source does not state whether the card was repasted before. Prior service history is therefore unknown.

The source does not state the age of the card. Service life cannot be inferred from the report.

The source does not state VRAM part numbers. Memory vendor and revision cannot be inferred from the report.

The result is best read as a dated snapshot. It shows what one card reported on one day in one title after one material change.

That snapshot still has value for readers tracking thermal degradation reports. It does not have value as a benchmark.

Any comparison to other titles would need separate data. No such data is present in the PC Central account.

The standing line in both tables repeats this limit on purpose. Single test, one card, Overwatch 2 is the boundary.

Prior Site Precedent in One Sentence

This site previously reported an RTX 3080 paste and pad modification where new pads alone did almost nothing.

That prior report is mentioned here only to note that interface contact has come up before on this site. No further detail from that report is restated in this section.

The parallel is limited to the theme of contact and interface choice. Different cards and coolers prevent direct comparison.

The current RTX 3070 account stands on its own numbers. It does not borrow support from the earlier item.

What Opening a Card Means

Opening a graphics card requires disassembly. That fact needs plain emphasis before any further discussion.

This article gives no step by step. No procedure is described and no procedure is implied.

Opening the card will very likely affect any warranty. A reader should assume warranty impact and verify their own coverage status.

Paste and pads are not interchangeable. Using the wrong interface type in the wrong place can worsen contact.

A reader should check their own card contact first. General reports cannot show whether a specific card has the same issue.

No tools are listed here. Tool choice is outside the scope of this article.

No thicknesses are listed here. Stack height is board specific and no value is given in the source.

No quantities are listed here. Amount and coverage guidance is not provided.

No guide is linked here. No forum thread is linked here.

No video is linked here. No store listing is linked here.

No product page is linked here. Thermalright and Gigabyte appear as names in prose only.

The Reddit origin is named in prose only. No thread link is provided and no username is stated in the source.

The caution also covers expectations about noise. Manageable fan noise after stabilization is one user qualitative report.

Qualitative noise reports do not translate across cases and rooms. Fan perception depends on conditions not stated in the source.

PC Central frames degraded thermal paste as a silent killer for older graphics cards. That language may sound urgent but it remains a source framing.

This article does not recommend opening any card. It documents a reported outcome and its limits.

A careful reader will treat the temperature table as information. Information alone does not create a safe plan for hardware work.

Anyone uncertain about disassembly should leave the card closed. Uncertainty is itself a reason not to proceed.

What Is Not Established

Many natural follow up questions have no answer in the PC Central report. Listing them prevents silent guessing.

Ambient temperature is not stated. Room conditions can move every temperature reading.

Test duration is not stated. Short and long sessions can produce different stabilized values.

Fan curve is not stated. Fixed and automatic control can produce different noise and temperature pairs.

Whether the card was repasted before is not stated. Prior work can change baseline contact.

The age of the card is not stated. Service time can change paste condition but no age is given.

Putty thickness is not stated. Gap filling depends on geometry that is not documented.

VRAM part numbers are not stated. Memory components cannot be identified from the report.

Clock speeds are not stated. Frequency behavior cannot be compared before and after.

FPS before or after is not stated. Performance change cannot be quantified from the report.

Results on any other card are not stated. Cross card claims would therefore be unsupported.

The table below collects those gaps in one place. Each row cites its standing as not stated in the source.

The gaps do not invalidate the reported numbers. They limit how far those numbers can travel.

A future report could close several gaps at once with logs and repeats. This report does not attempt that.

Until such detail exists, the responsible use of this story is narrow. It is a case note with a diagnostic hint.

The diagnostic hint is the delta change from 21C to 11C. That hint is stated as a reading and not as proof.

Readers should keep single test, one card, Overwatch 2 attached to every figure. Context prevents overreading.

FAQ

What card was tested in the PC Central report?
The card was the Gigabyte GeForce RTX 3070 Eagle OC V2, as named by PC Central in its Sep 21, 2026 report.

What temperatures changed after the material swap?
Per PC Central, VRAM moved from 99C to 75C, the core moved from 85C to 72C, and the hotspot moved from 106C to 83C in one Overwatch 2 test.

What materials were used and where did each go?
Per PC Central, the Reddit user applied Thermalright PhaseSheet PTM on the GPU core and Thermalright Putty Advance on VRAM and power delivery modules.

Why does the core to hotspot delta matter here?
The delta narrowed from 21C to 11C, which this article reads as restored contact rather than proof of cooler failure, based only on this single reported test.

Was anything else tested besides Overwatch 2?
PC Central names only Overwatch 2 and does not specify if other titles were benchmarked, so no other workload can be claimed.

Does this article tell readers to open their cards?
No, opening a card requires disassembly, this article gives no step by step, opening the card will very likely affect any warranty, paste and pads are not interchangeable, and a reader should check their own card contact first.

Bottom Line

One Reddit user test relayed by PC Central reports large drops on one Gigabyte RTX 3070 Eagle OC V2 in Overwatch 2. VRAM is stated at 99C to 75C and core at 85C to 72C.

The hotspot move from 106C to 83C completes the picture. The delta move from 21C to 11C supplies the diagnostic reading.

That reading favors contact failure over cooler failure for this specific case. It remains an interpretation of one data pair.

Materials are mapped without instruction. Thermalright PhaseSheet PTM on the core and Thermalright Putty Advance on VRAM and power delivery is the reported split.

Workload coverage stops at Overwatch 2. PC Central does not specify other titles and this article claims none.

Safety stays central because disassembly is involved. This article gives no step by step and it does not recommend the procedure.

Sources: PC Central report by Marek Novák titled RTX 3070 VRAM Temp Drops 24C After Thermal Paste Swap, published Sep 21, 2026 23:06 UTC, available at https://pccentral.net/rtx-3070-vram-temp-drops-24c-after-thermal-paste-swap/. Reddit origin, Thermalright materials, Gigabyte board, NVIDIA silicon and Overwatch 2 workload are named in prose as reported by that source.