PROJECT TITAN
Building a PC That Lives Below Freezing
A sealed, dried and actively refrigerated extreme-overclocking platform engineered to keep an entire high-performance computer operating in a controlled sub-zero environment.
Project Titan operating as a complete sealed sub-zero PC prototype.
Project Titan is not an exercise in making one processor cold. It is an attempt to engineer the entire environment around a computer - coolant, air, moisture, power delivery, instrumentation and thermal interfaces - so the machine can benchmark for extended periods at temperatures where a conventional open PC would rapidly encounter condensation and frost.
During a recent test period spanning roughly 2.5 days, the prototype accumulated about 35 hours of benchmarking with coolant generally in the -30 to -35 C range. Internal case air reached approximately -15 C, while the monitored dew point reached roughly -25 to -30 C.
Project support: Project Titan currently receives sponsorship/support from Seasonic, Splave One, Thermal Grizzly, and Patriot Viper memory. Their products and support are identified where relevant; benchmark results and engineering observations remain presented as recorded project data.
PROJECT TITAN | Sealed Sub-Zero PC Environment
The Problem With Going Below Zero
At ordinary temperatures, PC cooling is mostly a heat-transfer problem. Below freezing, the atmosphere becomes part of the thermal system. Water vapor can condense or freeze on cold tubing, blocks, fittings, PCB hardware and connectors. Titan therefore attacks the problem at the enclosure level instead of relying only on local insulation around the CPU or GPU.
The case is built from a 20/20 extruded-aluminum skeleton with temporary ABS sheet panels sealed to the frame. Custom 3D-printed pieces form mounts and cable interfaces. Permanent CNC/laser-cut aluminum panels are planned as the enclosure evolves.
The working enclosure: insulated coolant plumbing, internal circulation hardware and sealed component space.
Dry the air, then measure it
Multiple homemade packs of 3A molecular sieve are distributed through the enclosure. The sealed structure limits the arrival of new humid room air while the sieve adsorbs water vapor already trapped inside. A dedicated environmental monitor places its sensor inside the chamber and displays relative humidity, air temperature and calculated dew point externally.
Large molecular-sieve packs provide passive moisture capacity inside the enclosure.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Cooling the Computer - and Its Atmosphere
Titan uses a Watercool HEATKILLER D5 pump/reservoir combination together with a Cloudray P24100 laser-chiller pump. The Cloudray unit is nominally a 24 V pump and is currently operated from a 300 W, 20 V laptop-style supply. A remotely controlled 120 V outlet allows external switching.
Before coolant leaves the enclosure for the external refrigeration system, it passes through a 240 mm radiator with two 120 mm fans. In Titan that radiator is not rejecting heat to room air; it is using the sub-zero liquid loop as an internal air heat exchanger, actively pulling the sealed case atmosphere downward in temperature.
The internal radiator and insulated loop make the case itself part of the refrigeration strategy.
Coolant path
HEATKILLER glass D5 pump/reservoir -> Cloudray P24100 pump -> internal 240 mm radiator -> external refrigeration/chiller assembly -> return pass-through -> Aqua Computer temperature/flow monitor -> Alphacool ES Distro Plate C5 -> GPU / CPU / RAM / VRM / PCH cooling circuits -> distro plate -> reservoir.
Alphacool ES Distro Plate C5 (12995), used as Titan's compact hydraulic distribution center.
PROJECT TITAN | Sealed Sub-Zero PC Environment
The Hardware Inside Titan
The current platform centers on an ASUS ROG Maximus Z790 Apex Encore motherboard and Intel Core i9-14900KS. The CPU is cooled by a Watercool HEATKILLER IV full-nickel block. System memory is 32 GB of Patriot Viper Xtreme 5 DDR5-8200 fitted with aftermarket heat spreaders and an Iceman Cooler RAM water block.
The distribution network also feeds adapted HEATKILLER VRM blocks and a PCH/chipset block coupled to the motherboard's factory heatsink structure. The objective is to refrigerate the motherboard environment rather than leave major heat sources at ordinary case temperature.
A hardware-modified Gigabyte RTX 5090 MASTER
The graphics card is a Gigabyte GeForce RTX 5090 MASTER fitted with an all-copper water block and modified specifically for extreme-overclocking experimentation. The PCB carries 1 milliohm shunt modifications and a second native-style 12V-2x6 power input. Wiring has also been added to the card's I2C interface for external voltage-control and monitoring work.
RTX 5090 PCB during modification and thermal-interface preparation.
The added second 12V-2x6 input provides a second direct native-style GPU power connection.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Sub-Zero GPU Thermal Interfaces
The GPU does not use conventional thermal pads. Thermal Grizzly TG Putty Pro is used in place of the pad stack for secondary components. At the GPU die, Thermal Grizzly KryoSheet is used instead of conventional thermal paste, liquid metal or PTM7950.
In Titan testing, KryoSheet has provided the most repeatable GPU temperature behavior across the unusually wide operating range encountered by the project - from roughly -40 C sub-zero operation through temperatures approaching +80 C. That consistency is especially valuable when the same card moves repeatedly between cold benchmarking and warmer service or setup conditions.
Thermal Grizzly TG Putty Pro used instead of conventional GPU thermal pads.
Thermal Grizzly KryoSheet inventory used for repeatable dry-contact die interfaces.
External tuning and control
An ElmorLabs EVC2SE was used during some benchmark sessions through the card's I2C interface. The newer MellonVolt tool has also been incorporated into the tuning workflow. These tools belong to the experimental control side of the project and complement the physical power and shunt modifications on the card.
PROJECT TITAN | Sealed Sub-Zero PC Environment
What the Telemetry Actually Shows
The supplied HWiNFO captures are important because they document the system under real benchmark activity rather than only showing a cold idle screenshot. They also expose multiple power telemetry channels that should not be conflated.
Benchmark telemetry capture: sub-zero GPU temperature, clock data and both GPU Power and GPU Rail Powers channels.
Across the provided captures, the standard GPU Power channel records peaks around 509.9 to 528.3 W. The separate GPU Rail Powers channel reports much larger peak values, including 1,069.209 W, 1,292.206 W and 1,315.649 W. Those rail-power readings are retained here exactly as reported by the monitoring software; they are not discarded simply because they differ dramatically from the conventional GPU Power field.
At the same time, the captures show maximum GPU clocks in the approximately 3.47 to 3.57 GHz range and maximum GPU temperatures around 11 to 12.5 C during the recorded sessions, while minimum GPU temperatures extend below -30 C. One capture records -33.95 C minimum GPU temperature and a 3,570 MHz maximum GPU clock.
A second telemetry capture preserves the separate rail-power reading while documenting a 3,472 MHz maximum GPU clock.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Why the Hot-Spot and Memory Readings Need Context
Some sensor fields in the captures are clearly abnormal. The GPU memory-chip temperature field displays impossible multi-million-degree values, and hot-spot behavior can be inconsistent at deep sub-zero temperatures. Those fields should be treated as sensor/telemetry artifacts rather than physical temperatures. This does not invalidate the other simultaneously reported channels; it simply means each sensor has to be interpreted on its own merits in an operating regime far outside normal desktop conditions.
Power Delivery Outside the Chamber
A Seasonic PRIME TX-1600 is mounted externally on a custom upper case panel so its heat does not enter the controlled atmosphere. Cable extensions are sealed into panel cutouts, forming an electrical interface to the enclosure while leaving the native PSU largely unmodified.
The GPU power feeds are the exception. Both 12V-2x6 connections use native Seasonic cables routed directly from the external PSU to the modified RTX 5090, avoiding GPU power adapters in the high-current path.
Instrumentation at the PC
Returning coolant passes an Aqua Computer temperature/flow monitor at the computer, providing a measurement of what actually reaches the enclosure rather than relying solely on chiller-side instrumentation. A second coolant-temperature sensor is positioned in the Alphacool distro plate before the GPU cooling path.
External environmental instrumentation keeps the chamber condition visible without opening the case.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Measured Environmental Performance
The current enclosure has reported internal dew points in approximately the -25 to -30 C range while case air has fallen to approximately -15 C. Those numbers must be interpreted against the coldest exposed surface: a -30 C dew point does not automatically make a -35 C coolant surface condensation-safe. The engineering objective is therefore not a single low humidity number, but sufficient dew-point margin everywhere that matters.
Environmental monitor mounted outside the enclosure while sensing the sealed internal atmosphere.
35 Hours Below Freezing
The most important milestone so far is duration. Over approximately 2.5 days of testing, Project Titan accumulated roughly 35 hours of benchmarking with coolant consistently around -30 to -35 C. That makes the prototype more than a one-run demonstration. It is being developed around repeatability, controllability and the ability to remain operational for extended benchmarking sessions.
The external industrial refrigeration system is its own engineering project and will be covered separately. Titan's case project is focused on what happens after sub-zero coolant arrives: distribute it, exploit it, keep the surrounding air dry enough to survive it, and instrument the system well enough to know when conditions are no longer safe.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Next: Automatic Desiccant Regeneration
The current 3A molecular-sieve packs offer substantial passive capacity but eventually require regeneration. A MechMaxx molecular-sieve desiccant drying system has already been purchased for future integration. The goal is to move from a passive dry enclosure toward a continuously managed atmosphere: Measure -> Dry -> Regenerate -> Verify -> Maintain.
A Working Prototype, Not a Render
Titan remains visibly experimental. ABS panels are temporary, seals and penetrations continue to evolve, and the permanent aluminum enclosure is still ahead. That is also what makes the project useful: each revision is driven by what actually happens when a complete, heavily modified high-end PC spends hours operating below freezing.
The external environmental display is part of a system designed to be watched continuously during operation.
Engineering an Environment, Not Just a Cooler
Refrigeration controls the liquid. Two pumps move it. The distro plate routes it. Water blocks refrigerate the GPU, CPU, memory and motherboard. The internal radiator cools the surrounding air. The enclosure isolates that air. Molecular sieve removes moisture. Sensors measure both liquid and atmosphere. The modified RTX 5090, its power delivery, KryoSheet/TG Putty Pro interfaces, EVC2SE and MellonVolt tuning tools turn that environment into a benchmarking platform.
Project Titan has not reached its final form. Lower temperatures, permanent panels, regenerative drying and further instrumentation remain ahead. But the core concept is already operating: build a PC that does not merely survive a short sub-zero run, but can live and benchmark there.
PROJECT TITAN | Sealed Sub-Zero PC Environment
PROJECT TITAN — PART II
Engineering the Cascade Cooling Plant
From flash-frozen coolant to an isolated, buffered, multi-fluid sub-zero system
The DLSB-30/80 cascade refrigeration fluid chiller that became the primary refrigeration plant for the next phase of Project Titan.
The second phase of Project Titan moved beyond the original custom single-stage refrigeration system and into cascade refrigeration. The new primary plant is a DLSB-30/80 low-temperature circulating chiller, a two-stage cascade machine intended to operate with circulating-fluid temperatures approaching -80 C. That capability provided far more temperature headroom than Titan's earlier system, but it also exposed a problem that did not appear at ordinary sub-zero temperatures: the coldest point in the refrigeration plant can be much colder than the bulk fluid temperature shown on the controller.
A Cascade System, Not Just a Colder Thermostat
The DLSB-30/80 uses two mechanically distinct refrigeration stages. The first, or high-temperature stage, creates the conditions needed for the second, low-temperature stage to operate. The low-temperature stage then provides the ultra-low-temperature refrigeration used to pull the circulating fluid toward the machine's approximately -80 C design range.
The temperature-control strategy is crucial to understanding Titan's early coolant problem. The chiller's control system regulates the second-stage compressor on and off according to the measured circulating-fluid temperature. It does not simply hold the second-stage evaporator/brazed-plate heat exchanger at the same temperature displayed for the bulk fluid.
Consequently, while the controller may show circulating fluid at -40, -50 or -56 C, the active low-stage refrigerant circuit can expose the internal evaporator/BPHE to substantially colder local conditions. The refrigeration hardware must be cold enough to continue pulling heat from the fluid toward an eventual -80 C operating point. Bulk fluid temperature, evaporating refrigerant temperature and heat-exchanger surface temperature therefore cannot be treated as the same measurement.
Observed operation: the controller displays approximately -56.0 C process temperature with a -60.0 C setpoint. This is documented bulk-fluid operation, not a claim that the coolant in this photograph had reached -80 C.
DESIGN CAPABILITY: approximately -80 C fluid operation | DOCUMENTED TEST POINT: approximately -56 C fluid at a -60 C setpoint
That distinction is preserved throughout this article. The -80 C figure describes the intended operating capability of the DLSB-30/80 architecture. Photographs and telemetry are used separately to document temperatures actually observed during Project Titan testing.
Inside the Refrigeration Plant
Removing the panels makes the cascade architecture tangible. The machine contains separate compressor and heat-exchange hardware, insulated low-temperature refrigerant plumbing and the internal fluid heat-exchange assembly. These components form a refrigeration plant whose local refrigerant-side temperatures can be substantially lower than the temperature of the circulating bath.
Lower refrigeration compartment: compressor, condenser/air-moving hardware and copper refrigerant plumbing inside the DLSB chassis.
Another view of the compressor and refrigeration hardware. The cascade system contains substantially more mechanical infrastructure than a conventional PC water chiller.
Upper internal section showing heavily insulated low-temperature piping and heat-exchange hardware associated with the cascade system.
Internal fluid/refrigeration plumbing. The insulated sections are essential because the low-temperature stage operates far below ambient conditions.
Why the Original Coolant Failed
Titan had previously used OMTech laser-chiller fluid with useful operation around -35 C. In the earlier refrigeration system that was adequate. In the DLSB, it was not.
When the second stage was active, the coolant passing through the internal brazed-plate heat exchanger encountered local temperatures cold enough to begin freezing or forming slush even though the bulk fluid temperature had not necessarily reached the DLSB's ultimate -80 C capability. The restriction reduced coolant flow. Lower flow increased residence time in the extremely cold exchanger, while the loss of circulation simultaneously reduced heat transport from the PC. The result was the counterintuitive condition of a powerful refrigeration plant running while the PC-loop temperature began to rise.
Engineering lesson: freeze protection must be selected for the coldest local heat-transfer condition in the system, not merely the bulk temperature displayed by the controller.
Separating the Refrigeration Fluid From the PC Fluid
The solution was to stop asking one coolant to satisfy two very different sets of requirements. Titan was divided into a primary refrigeration circuit and a chemically separate secondary PC circuit.
For the DLSB primary loop, a five-gallon pail of Sunnyside denatured alcohol was sourced locally. The purpose was straightforward: give the cascade refrigeration circuit a fluid with substantially more low-temperature margin than the previous laser-chiller coolant.
Sunnyside denatured alcohol selected for the DLSB primary refrigeration circuit.
Straight denatured alcohol was deliberately kept out of the computer itself. PC water blocks, pumps, reservoirs and their internal seals were not selected around continuous service with that fluid. Instead, the computer received its own coolant formulation.
Prime Guard RV & Marine antifreeze was used as the secondary-loop base. The product contains a water/alcohol/glycol mixture, and Titan's test fluid was further modified with laboratory-grade methanol. Three one-gallon batches were prepared independently to make the fill repeatable and to avoid a total-volume estimate changing the composition of the entire loop.
Prime Guard RV & Marine antifreeze used as the base for the isolated PC-side coolant.
Batch target used for the one-gallon PC coolant mixture: approximately 2.24 L base RV antifreeze plus 1.54 L methanol, for approximately 3.78 L total.
The Unlikely Heat Exchanger
With two chemically independent loops, the next problem was transferring heat between them. The first workable answer came from a homemade wort chiller previously purchased through Facebook Marketplace.
The exchanger consists of 3/8-inch copper tubing routed inside blue PEX tubing. Titan sends the DLSB's denatured-alcohol primary fluid through the inner copper tube. The PC-side RV-antifreeze/methanol mixture travels through the outer PEX passage, surrounding the copper tube without mixing with the primary fluid. In effect, the repurposed assembly became a concentric liquid-to-liquid heat exchanger.
The repurposed wort chiller: 3/8-inch copper tubing runs inside the blue PEX, allowing the DLSB and PC coolants to exchange heat without mixing.
Initial testing proved the concept, but it also exposed the exchanger's limitation. The temperature difference between the primary chiller fluid and the PC loop remained large - roughly 25 to 30 C under the observed conditions. The refrigeration plant could go colder, but heat transfer through the improvised exchanger had become the new bottleneck.
Turning the Old Freezer Into a Thermal Buffer
Rather than abandon the exchanger, Titan reused another piece of the previous cooling system: the mini deep freezer that had served as the fluid bath for the earlier custom refrigeration chiller.
The entire copper-in-PEX exchanger was submerged in a bath made primarily from concentrated propylene-glycol coolant. The freezer can pull that bath toward approximately -40 C. Its job is not to replace the DLSB or independently absorb the PC's entire benchmark load. Instead, the cold bath surrounds and insulates the exchanger, reduces environmental heat gain, adds thermal mass and provides a stabilizing buffer around the PC-side fluid while heat moves into the colder primary circuit.
The resulting system is no longer a simple chiller connected to a computer. It is a multi-stage thermal network in which refrigeration temperature, primary-fluid temperature, buffer-bath temperature and PC-loop temperature are deliberately allowed to be different.
DLSB CASCADE -> denatured-alcohol primary loop -> submerged copper/PEX heat exchanger -> RV-antifreeze/methanol PC loop -> Titan water blocks
What the Combined System Can Sustain
With the buffered heat exchanger in place, testing has supported approximately -45 to -55 C primary chiller-loop temperatures, a freezer bath near -40 C, and approximately -35 to -40 C coolant delivered on the PC side. Before benchmark loading, that has placed the RTX 5090 core in approximately the -30 to -35 C range.
Those numbers are intentionally presented as separate measurements. A -55 C primary-loop reading does not imply a -55 C GPU, and the DLSB's -80 C design capability does not mean every Titan benchmark is performed with -80 C coolant. Each temperature belongs to a different point in the thermal chain.
Why the Temperature Delta Matters
The 25 to 30 C difference observed across the early exchanger configuration is useful engineering data rather than an embarrassment to hide. It identifies where the next gains are available. Once refrigeration capacity is sufficient, exchanger conductance, coolant viscosity, flow rate, surface area, insulation and thermal mass determine how much of that low-temperature capability actually reaches the silicon.
Project Titan's evolution therefore illustrates a recurring rule in extreme cooling: achieving an impressive evaporating temperature is only the beginning. The engineering problem is transporting that temperature through every interface while maintaining flow, chemical compatibility, electrical reliability and repeatable benchmark behavior.
From Refrigeration Plant to Silicon
The refrigeration-side work connects directly to the hardware described in Part I. Titan's RTX 5090 uses Thermal Grizzly TG Putty Pro instead of conventional thermal pads and a Thermal Grizzly KryoSheet at the GPU die. In the project's testing, the KryoSheet interface has produced the most consistent GPU temperature behavior across the unusually wide operating range, from deep sub-zero conditions through positive-temperature operation.
Some benchmark sessions have also used an ElmorLabs EVC2SE for external control and monitoring, while the newer MellonVolt tool has been incorporated into the tuning workflow. The supplied telemetry is treated as recorded data: the standard GPU Power channel and the separate GPU Rail Powers channel are both preserved rather than discarding the rail-power readings simply because they differ substantially.
That combination - cascade refrigeration, isolated fluids, a buffered heat exchanger, a sealed and dried PC environment, modified GPU power delivery and repeatable thermal interfaces - is what turns Project Titan from a cold-water experiment into a complete extreme-overclocking platform.
The Engineering Objective
The goal of Project Titan is not to produce the lowest unloaded temperature on a display. It is to sustain useful sub-zero conditions at the CPU and GPU while the computer is actually generating benchmark-level heat.
The DLSB-30/80 supplies the low-temperature refrigeration headroom. Denatured alcohol allows the primary circuit to exploit more of that range. The isolated methanol/RV-antifreeze secondary loop protects the PC-side hardware from direct exposure to the primary fluid. The repurposed counterflow exchanger transfers heat between them, and the -40 C freezer bath adds insulation and thermal buffering.
Every stage introduces a temperature difference, and every difference represents both a loss and an engineering opportunity. The current system already delivers approximately -35 to -40 C PC coolant while keeping the ultra-low-temperature cascade machinery chemically separated from the computer. Future exchanger and control refinements can focus on shrinking that delta rather than rebuilding the entire refrigeration plant.
PROJECT TITAN
Extreme Sub-Zero Refrigeration Overclocking Platform
PROJECT TITAN | Sealed Sub-Zero PC Environment
PROJECT TITAN - VALIDATION
From Cascade Engineering to HWBOT Results
The purpose of Project Titan is not simply to make coolant cold. Its engineering target is repeatable, sustained sub-zero operation under real benchmark load. The September 2026 HWBOT submissions provide the clearest validation of that objective: the same platform developed around the sealed enclosure, cascade refrigeration plant, isolated coolant circuits and modified RTX 5090 produced multiple first-place benchmark submissions while also sustaining GPU clocks in the mid-3.5 GHz range.
Documented Benchmark Results
Benchmark
Score
HWBOT rank*
Submitted GPU / memory
Steel Nomad Vulkan
19,994
1st of 324
3517 / 2250 MHz
Solar Bay
335,399
1st of 515
3555 / 2250 MHz
Speed Way
19,258
1st of 610
3570 / 2250 MHz
Steel Nomad DX12
20,099
1st of 554
3525 / 2250 MHz
Fire Strike Ultra
42,152
1st of 2,580
3472 / 2125 MHz
Time Spy
54,082
2nd of 5,735; 1st RTX 5090
3547 / 2250 MHz
Fire Strike
81,001
12th of 5,631
3472 / 2125 MHz
Fire Strike Extreme
66,503
8th of 3,929
3472 / 2125 MHz
Port Royal
51,599
8th of 1,112
3472 / 2125 MHz
*Ranks shown are the HWBOT page values verified during preparation of this article on September 9, 2026. Rankings can change as new submissions are added.
Steel Nomad Vulkan: 19,994 marks at 3517/2250 MHz. The supplied HWBOT capture records the submission at the top of the world-record ranking.
Solar Bay: 335,399 marks at 3555/2250 MHz, documented as a first-place HWBOT world-record submission.
Speed Way: 19,258 marks at 3570/2250 MHz. HWBOT records the submission first overall at the time documented.
Steel Nomad DX12: 20,099 marks at 3525/2250 MHz, another first-place submission in the documented Titan run set.
Fire Strike Ultra: 42,152 marks at 3472/2125 MHz. The supplied submission capture records first place overall.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Time Spy: Second Overall, First Among RTX 5090 Submissions
Time Spy requires a more precise description. Titan scored 54,082 overall with a 61,923 Graphics Score and 31,489 CPU Score. HWBOT recorded the run second in the unrestricted world-record ranking while simultaneously placing it first in the 1-core GeForce RTX 5090 hardware ranking. The separate leaderboard capture shows the higher overall result ahead of Titan was produced with an RTX 5090D. The article therefore does not mislabel the 54,082 run as the unrestricted Time Spy world record.
Time Spy submission: 54,082 overall, 61,923 Graphics and 31,489 CPU, submitted at 3547/2250 MHz. HWBOT records second overall and first in the RTX 5090 hardware category.
Time Spy world-record ranking capture supplied with the project. Titan is shown second overall behind an RTX 5090D result.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Telemetry: Temperature, Frequency and Rail Power
The benchmark screenshots also connect the ranking results back to Titan's thermal and electrical engineering. The sensor captures are retained as recorded evidence rather than simplified into a single power number. In particular, HWiNFO exposes a conventional GPU Power channel and a separate GPU Rail Powers channel. Those channels report very different maxima on this modified card and should not be treated as interchangeable.
Port Royal telemetry: 51,599 points, approximately -31 C GPU temperature at capture, a 3,570 MHz maximum GPU clock, 528.302 W maximum on the conventional GPU Power channel and 1,069.209 W maximum on GPU Rail Powers.
The Port Royal capture is especially useful because both power channels are visible in the same frame. The conventional GPU Power maximum is 528.302 W while GPU Rail Powers reaches 1,069.209 W. Project Titan therefore preserves the rail-power reading exactly as reported by HWiNFO instead of discarding it because it exceeds the conventional board-power channel.
Solar Bay telemetry: 335,399 points with GPU temperature around -31.7 C at capture. HWiNFO records a 3,583.2 MHz maximum measured GPU clock and 989.133 W maximum GPU Rail Powers.
Fire Strike telemetry: 81,001 points. This screenshot ties the benchmark result to approximately -29 C GPU operation and nearly 1 kW maximum on the separate GPU Rail Powers channel.
Sub-zero operation also exposes sensor channels that are not trustworthy. The multi-million-degree GPU memory-chip temperature values visible in the captures are clearly invalid telemetry artifacts. They are not used as physical temperature measurements. That does not justify rejecting unrelated channels that remain internally useful; each sensor is interpreted according to what it actually reports and whether the value is physically credible.
PROJECT TITAN | Sealed Sub-Zero PC Environment
The Engineering Chain Behind the Scores
These results close the loop on the refrigeration work described earlier in the article. The DLSB-30/80 is a cascade machine intended to drive its low-temperature stage toward the conditions required for approximately -80 C fluid operation. Its control system cycles the second-stage compressor according to circulating chiller-fluid temperature; it does not directly regulate the second-stage refrigerant-side BPHE surface to the displayed bulk-fluid setpoint. That architecture is why the earlier coolant could locally freeze inside the exchanger even while the displayed fluid temperature was substantially warmer than -80 C.
The solution - a denatured-alcohol primary circuit, a chemically isolated methanol/RV-antifreeze PC circuit, a concentric copper-in-PEX liquid-to-liquid exchanger, and a roughly -40 C propylene-glycol freezer bath used as an insulating thermal buffer - converted raw refrigeration capability into controllable coolant delivery at the computer. Integrated testing has produced roughly -45 to -55 C primary-loop temperatures, approximately -40 C buffer-bath conditions and approximately -35 to -40 C PC-loop coolant, with pre-benchmark GPU temperatures around -30 to -35 C.
The benchmark captures show the practical consequence: the RTX 5090 can begin and complete high-load 3DMark runs while remaining deeply sub-zero and sustaining clocks around 3.5 GHz. The refrigeration plant, sealed dry enclosure, coolant chemistry, heat exchanger, power-delivery modifications and tuning workflow are therefore parts of one system rather than isolated modifications.
PROJECT TITAN | Sealed Sub-Zero PC Environment
Project Partners & Sponsors
Project Titan currently receives sponsorship/support from four performance-hardware partners whose products are represented directly in the platform: Seasonic, Splave One, Thermal Grizzly, and Patriot Viper memory. This support helps sustain the repeated hardware development, thermal experimentation and benchmark validation documented in this article.
Seasonic
Power-delivery partner. Titan uses an externally mounted Seasonic PRIME TX-1600, with native Seasonic GPU power cables routed directly to the modified RTX 5090.
Splave One
Extreme-overclocking partner/supporter. Splave One thermal interface material is used on the CPU side of the Titan platform.
Thermal Grizzly
Thermal-interface partner. Titan uses TG Putty Pro on GPU secondary components and KryoSheet at the GPU die for repeatable operation across the project's unusually wide temperature range.
Patriot Viper memory
Memory partner. Titan uses Patriot Viper Xtreme 5 DDR5-8200 memory, integrated into the refrigerated loop with an aftermarket heat-spreader/Iceman Cooler RAM-block arrangement.
Sponsor acknowledgment does not change how results are reported: HWBOT rankings are identified from the documented submission pages, while temperatures, clocks and power channels are retained as measured or displayed in the supplied telemetry.
Primary HWBOT Submission References
The following links are the submission pages supplied for Project Titan. They are included as primary references so the article can be checked against the live benchmark records. Because HWBOT rankings are dynamic, the score and submission settings are permanent evidence while ordinal rankings may change over time.
Time Spy (GPU) - supplied submission
At this stage, Project Titan has moved beyond proof-of-concept refrigeration. Its significance is the integration: an ultra-low-temperature cascade plant, moisture-controlled sealed enclosure, chemically separated coolant circuits, modified high-power GPU and repeatable benchmark workflow operating as a single platform.
PROJECT TITAN SPONSORS / SUPPORTING PARTNERS
Seasonic | Splave One | Thermal Grizzly | Patriot Viper memory
PROJECT TITAN | Sealed Sub-Zero PC Environment