Accelsius Achieves Industry-Leading Thermal Milestones for Next-Generation AI Computing
Accelsius has announced two major breakthroughs in thermal testing for AI computing cooling solutions. Their proprietary NeuCool two-phase, direct-to-chip technology successfully tested at 4,500W per GPU socket, setting an industry record for direct-to-chip liquid cooling capacity.
The company demonstrated its two-phase CDU's ability to cool a 250kW AI server rack while maintaining GPU temperatures below thermal limits, even with 40°C facility water. This performance enables 25% cooling energy savings and increased free cooling hours. The system maintained GPU junction temperatures below NVIDIA's 87°C thermal throttle limit under maximum load.
These achievements address critical cooling challenges for next-generation AI infrastructure and support sustainability goals through improved energy efficiency. The company will showcase these capabilities at Data Center World in Washington, DC, and present cold-plate research for vertical servers at the OCP EMEA Summit in Dublin.
Accelsius ha annunciato due importanti innovazioni nei test termici per le soluzioni di raffreddamento dell'IA. La loro tecnologia proprietaria NeuCool, a due fasi e diretta al chip, ha superato con successo il test a 4.500W per socket GPU, stabilendo un record di settore per la capacità di raffreddamento a liquido diretto al chip.
L'azienda ha dimostrato la capacità del suo CDU a due fasi di raffreddare un rack di server AI da 250kW mantenendo le temperature delle GPU al di sotto dei limiti termici, anche con acqua di impianto a 40°C. Questa prestazione consente un risparmio energetico del 25% nel raffreddamento e un aumento delle ore di raffreddamento gratuito. Il sistema ha mantenuto le temperature di giunzione delle GPU al di sotto del limite di throttling termico di 87°C di NVIDIA sotto carico massimo.
Questi risultati affrontano sfide critiche nel raffreddamento per le infrastrutture AI di nuova generazione e supportano gli obiettivi di sostenibilità attraverso un miglioramento dell'efficienza energetica. L'azienda presenterà queste capacità al Data Center World a Washington, DC, e presenterà ricerche sulle piastre fredde per server verticali al Summit OCP EMEA a Dublino.
Accelsius ha anunciado dos importantes avances en las pruebas térmicas para soluciones de refrigeración de computación AI. Su tecnología patentada NeuCool, de dos fases y directa al chip, se probó con éxito a 4,500W por socket GPU, estableciendo un récord en la industria para la capacidad de refrigeración líquida directa al chip.
La empresa demostró la capacidad de su CDU de dos fases para enfriar un rack de servidores AI de 250kW manteniendo las temperaturas de la GPU por debajo de los límites térmicos, incluso con agua de instalación a 40°C. Este rendimiento permite un ahorro energético del 25% en refrigeración y un aumento en las horas de refrigeración gratuita. El sistema mantuvo las temperaturas de unión de la GPU por debajo del límite de estrangulación térmica de 87°C de NVIDIA bajo carga máxima.
Estos logros abordan desafíos críticos de refrigeración para la infraestructura AI de próxima generación y apoyan los objetivos de sostenibilidad a través de una mejor eficiencia energética. La empresa mostrará estas capacidades en el Data Center World en Washington, DC, y presentará investigaciones sobre placas frías para servidores verticales en la Cumbre OCP EMEA en Dublín.
Accelsius는 AI 컴퓨팅 냉각 솔루션을 위한 열 테스트에서 두 가지 주요 혁신을 발표했습니다. 그들의 독점 기술인 NeuCool의 2상 직결 칩 기술이 GPU 소켓당 4,500W에서 성공적으로 테스트되어, 직결 칩 액체 냉각 용량에 대한 업계 기록을 세웠습니다.
회사는 40°C의 시설수에서도 GPU 온도를 열 한계 이하로 유지하면서 250kW AI 서버 랙을 냉각할 수 있는 2상 CDU의 능력을 입증했습니다. 이 성능은 25%의 냉각 에너지 절약과 무료 냉각 시간을 증가시킵니다. 시스템은 최대 부하에서 NVIDIA의 87°C 열 제어 한계 이하로 GPU 접합 온도를 유지했습니다.
이러한 성과는 차세대 AI 인프라의 중요한 냉각 문제를 해결하고 에너지 효율성을 개선하여 지속 가능성 목표를 지원합니다. 이 회사는 워싱턴 DC의 데이터 센터 월드에서 이러한 기능을 선보이고, 더블린의 OCP EMEA 서밋에서 수직 서버를 위한 냉각판 연구를 발표할 예정입니다.
Accelsius a annoncé deux percées majeures dans les tests thermiques pour des solutions de refroidissement en informatique AI. Leur technologie propriétaire NeuCool, à deux phases et directe au chip, a été testée avec succès à 4 500W par socket GPU, établissant un record dans l'industrie pour la capacité de refroidissement liquide directe au chip.
L'entreprise a démontré la capacité de son CDU à deux phases à refroidir un rack de serveurs AI de 250kW tout en maintenant les températures des GPU en dessous des limites thermiques, même avec de l'eau d'installation à 40°C. Cette performance permet des économies d'énergie de 25% en refroidissement et augmente les heures de refroidissement gratuit. Le système a maintenu les températures de jonction des GPU en dessous de la limite de throttling thermique de 87°C de NVIDIA sous charge maximale.
Ces réalisations répondent à des défis critiques de refroidissement pour l'infrastructure AI de prochaine génération et soutiennent les objectifs de durabilité grâce à une amélioration de l'efficacité énergétique. L'entreprise présentera ces capacités au Data Center World à Washington, DC, et présentera des recherches sur les plaques froides pour serveurs verticaux au Sommet OCP EMEA à Dublin.
Accelsius hat zwei bedeutende Durchbrüche in der thermischen Prüfung von Kühlungslösungen für KI-Computing bekannt gegeben. Ihre proprietäre NeuCool Technologie mit zwei Phasen, die direkt auf den Chip wirkt, wurde erfolgreich mit 4.500W pro GPU-Sockel getestet und hat einen Branchenrekord für die direkte Chip-Flüssigkeitskühlkapazität aufgestellt.
Das Unternehmen demonstrierte die Fähigkeit seines zweiphasigen CDU, einen 250kW KI-Server-Rack zu kühlen, während die GPU-Temperaturen unter den thermischen Grenzen blieben, selbst bei 40°C Zuluft. Diese Leistung ermöglicht 25% Energieeinsparungen bei der Kühlung und erhöht die Stunden der kostenlosen Kühlung. Das System hielt die GPU-Junction-Temperaturen unter dem thermischen Drosselungsgrenzwert von 87°C von NVIDIA bei maximaler Last.
Diese Erfolge adressieren kritische Kühlherausforderungen für die Infrastruktur der nächsten Generation von KI und unterstützen Nachhaltigkeitsziele durch verbesserte Energieeffizienz. Das Unternehmen wird diese Fähigkeiten auf der Data Center World in Washington, DC, präsentieren und Forschung zu Kaltplatten für vertikale Server auf dem OCP EMEA Summit in Dublin vorstellen.
- Achieved industry-leading cooling capacity of 4,500W per GPU socket
- Demonstrated ability to cool 250kW AI server rack
- Delivers 25% cooling energy savings
- Maintains performance even with 40°C facility water
- Shows scalability for future 600kW racks
- Testing by infrastructure power limit, not cooling system capacity
Insights
Accelsius has achieved a significant technical breakthrough in liquid cooling for AI computing with their NeuCool technology demonstrating 4,500W per GPU socket cooling capacity and a full 250kW AI server rack cooling capability. These aren't incremental improvements but rather step-changes in thermal management that address critical bottlenecks in AI infrastructure scaling.
The thermal headroom demonstrated has profound implications for data center economics. By operating with facility water temperatures
What truly differentiates this announcement is the forward compatibility demonstrated. While current-gen NVIDIA H100 GPUs typically draw 700-800W, next-gen AI accelerators are expected to push beyond 1000W per chip. Accelsius is clearly positioning for the upcoming 600kW rack architectures and NVIDIA's Kyber racks with vertically oriented servers.
The technology's ability to maintain GPU temperatures below thermal limits even with 40°C facility water represents a competitive moat. Traditional cooling solutions struggle with warm-water environments, often requiring energy-intensive chilling. This breakthrough potentially unlocks deployment in regions with cooling infrastructure or enables significant operational cost reductions in established data centers.
- Successfully tested NeuCool technology at 4,500W per GPU socket—the highest in the industry for direct-to-chip liquid cooling
- Demonstrated the highest cooling capacity two-phase CDU on the market, supporting a full 250kW AI server rack while maintaining GPU temperatures below thermal limits even with 40°C facility water
-
Accelsius will showcase these capabilities at Data Center World in
Washington, DC (Booth #524), and present innovative cold-plate research for vertically oriented servers at the OCP EMEA Summit inDublin
In preparation for the next wave of ultra-high-power GPUs, Accelsius tested a thermal test vehicle representative of a GPU socket. In the experiment, the cold plate was heated to 4,500W before reaching the power limit of the test infrastructure, not the cooling system. Cold-plate temperatures remained stable and within operational limits even at these unprecedented loads. This represents the highest cooling capacity achieved in the industry for direct-to-chip liquid cooling, proving NeuCool has the thermal headroom to support multiple upcoming generations of processors.
In a second test, Accelsius successfully demonstrated the ability of its in-row two-phase CDU to cool a full 250kW rack of AI servers. A four-way H100 server was retrofitted with two-phase cold plates directly contacting the switches, CPUs and GPUs, and then installed in a densely configured 250kW rack.
The CDU was subjected to facility water (PG25) at 20°C, 30°C and 40°C, and 375 liters per minute of flow. Even under maximum rack load and using 40°C inlet temperatures, the junction temperature of the hottest GPU remained below NVIDIA’s thermal throttle limit (~87°C), highlighting the flexibility and efficiency of two-phase cooling even in warm-water environments.
The ability to operate with facility water 6-8°C higher than competing solutions can translate to over
“We’re showing customers that we can easily meet current performance requirements and scale our performance to meet the needs of the recently announced 600kW racks,” said Dr. Richard Bonner, chief technical officer at Accelsius. “Our R&D team has also prepared us for rapidly evolving chip and server architectures, such as 4,500W TDP sockets and vertically oriented blade servers.”
Accelsius will be on the show floor at Data Center World in
Accelsius will also present innovative cold-plate research at the OCP EMEA Summit’s Future Technologies Symposium in
To learn more about Accelsius’ NeuCool two-phase direct-to-chip in-rack liquid cooling solution, email info@accelsius.com or visit accelsius.com.
About Accelsius
Founded by Innventure, Inc. (NASDAQ:INV), Accelsius empowers data center and edge operators to achieve their business, financial and sustainability goals through advanced cooling solutions. The proprietary NeuCool platform provides best-in-class thermal efficiencies through a safe, two-phase liquid cooling system that scales from single racks to entire data centers. For more information, visit accelsius.com or follow us on LinkedIn.
View source version on businesswire.com: https://www.businesswire.com/news/home/20250410459051/en/
Media Contact
Treble
McKenzie Covell
accelsius@treblepr.com
Source: Accelsius
FAQ
What cooling capacity did Accelsius's NeuCool technology achieve in recent tests?
How much power can Accelsius's two-phase CDU cool in a single AI server rack?
What energy savings can data centers achieve with Accelsius's cooling solution?
What is the maximum facility water temperature at which Accelsius's cooling system remains effective?
How does Accelsius's cooling solution perform with NVIDIA GPU thermal limits?
