Verdict: HotHardware’s AI, rendering and media tests show the MSI GeForce RTX 5090 Suprim Liquid SOC winning or tying several workloads, but usually by a narrow margin over NVIDIA’s RTX 5090 Founders Edition. Its more convincing advantage is the 360 mm cooler’s low temperatures and quiet operation, not a consistently large professional-performance uplift.
What Page 2 actually tests
This page of HotHardware’s review is not a gaming roundup. It covers local AI inference, 3D rendering and video decoding, using the following comparison cards: Radeon RX 7900 XT, GeForce RTX 4090, GeForce RTX 5090 Founders Edition and MSI’s RTX 5090 Suprim Liquid SOC.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
ASUS TUF Gaming GeForce RTX™ 5080 16GB GDDR7 OC Edition Graphics Card | PCIe 5.0, HDMI/DP 2.1,... | $1,899.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
The test system used an AMD Ryzen 7 9800X3D, MSI X870E Carbon WiFi, 32 GB of DDR5-6000 memory, a Samsung SSD 990 and Windows 11 Pro 24H2. Resizable BAR and EXPO were enabled. The NVIDIA driver was 571.86 and the AMD driver was 24.12.1, reflecting the review’s January 24, 2025 launch-period software rather than a guaranteed 2026 result. Full methodology and charts are on HotHardware’s Page 2 review.
Chart positions should also be read with their measurement context. Some tests report scores, some report throughput or render results, and small differences can be smaller than run-to-run variation. A first-place finish is not automatically a useful production advantage.
#1 Best Overall
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
| Workload | What it measures | MSI versus RTX 5090 Founders Edition |
|---|---|---|
| MLPerf Client | Local machine-learning inference | MSI led |
| Geekbench AI | Cross-platform AI workloads | Founders Edition slightly led; within margin of error |
| UL Procyon AI Text Generation | LLM inference across several models and prompts | MSI led most models; Phi-3.5 favored the Founders Edition |
| UL Procyon AI Image Generation | Stable Diffusion XL image generation | MSI led the tested comparison, with a major software caveat |
| Blender 4.3 | GPU rendering across three default scenes | Founders Edition slightly led; not meaningful |
| V-Ray GPU | CUDA/RTX-oriented rendering | Effectively tied |
| Blackmagic RAW Speed Test | GPU-assisted RAW decoding at 4K and 8K | MSI was at the top alongside the Founders Edition |
AI performance: mostly small wins, with an important software warning
MLPerf Client
The Suprim Liquid SOC finished ahead of the Founders Edition in MLPerf Client. HotHardware attributed the edge to MSI’s higher clocks, but MLPerf results depend on model versions, drivers, runtime paths and vendor tuning. This is evidence of a lead in this configuration, not proof that the card is faster in every local-AI application.
Geekbench AI
Geekbench AI placed MSI slightly behind NVIDIA’s card. HotHardware considered the difference small enough to be within the benchmark’s margin of error. That counterexample matters: a higher advertised boost clock does not guarantee a measurable advantage when two cards use the same RTX 5090 GPU and memory subsystem.
Procyon AI Text Generation
UL Procyon used seven prompts, including retrieval-augmented and non-RAG queries, with Phi-3.5-mini, Mistral-7B, Llama 3.1 8B and Llama 2 13B. MSI led every tested model except Phi-3.5, generally by only a few points.
The execution path is crucial. HotHardware ran ONNX with DirectML rather than NVIDIA’s preferred TensorRT route. These results therefore describe that DirectML configuration; they do not establish equivalent performance in CUDA, TensorRT-LLM, llama.cpp, Ollama, ComfyUI or a particular professional application.
Stable Diffusion XL
The Stable Diffusion XL test used FP16 through ONNX because the benchmark did not yet have updated TensorRT support for Blackwell. MSI beat the RTX 4090 and finished ahead of the other tested cards, but the reviewer warned that the RTX 5090 was not running an optimized Blackwell TensorRT path.
That makes this a useful launch-software snapshot, not a definitive measure of the architecture’s maximum image-generation performance. A mature TensorRT implementation, later drivers or a different application can change the ordering. The review was published on January 24, 2025, so current users should not treat its exact result as a 2026 guarantee.
Rendering results do not justify the premium by themselves
Blender 4.3
HotHardware used Blender 4.3’s GPU benchmark and its three default scenes. The Founders Edition was slightly faster, but the difference was judged insignificant. The liquid cooler and factory overclock did not create a clear Blender lead.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesProduction work can behave differently depending on OptiX or CUDA settings, scene complexity, asset loading and sustained temperature. A buyer should seek render times from their own scenes before paying extra for this model solely for Blender.
V-Ray GPU
V-Ray’s GPU benchmark likewise showed no material benefit from MSI’s higher clocks; the cards were effectively tied in this configuration. That does not mean every V-Ray scene will produce identical times, but it does mean the Suprim Liquid SOC’s premium is not supported by a demonstrated V-Ray advantage on this page.
Blackmagic RAW is the clearest creator win
Blackmagic RAW Speed Test measures GPU-assisted decoding across four 4K and 8K combinations of resolution, bitrate and compression. MSI finished at the top alongside, or effectively level with, the Founders Edition and was described as the fastest overall RAW performer in the comparison.
This result is most relevant to editors whose projects actually use Blackmagic RAW. It should not be generalized to ProRes, H.264, H.265 or AV1 encoding, Adobe exports, DaVinci Resolve timeline playback or an entire post-production workflow. It is a decode test, not a complete editing-system benchmark.
Why the liquid cooler may matter more than the charts
The card combines NVIDIA’s RTX 5090 with a 360 mm all-in-one cooler. The card itself is approximately 280 × 148 × 51 mm and about 2,969 g; the review lists a 394 × 121 × 55 mm radiator assembly and roughly 280 mm of tubing. It occupies about 2.7 slots, while the radiator uses three preinstalled 120 mm fans. An additional card fan and VRM heatsinks handle board-side cooling.
HotHardware found substantially lower temperatures and quiet operation compared with the Founders Edition, including fan stop at idle or light load. Gaming and Silent BIOS modes mainly changed the fan curve in practice. The card’s 2,580 MHz Gaming-mode clock (2,565 MHz listed boost) and 600 W power target provide thermal and power headroom, while Silent mode is listed at 2,527 MHz in MSI Center (2,512 MHz listed boost) and 575 W.
Those characteristics are valuable for long AI, rendering or compute sessions, where sustained acoustics and temperatures matter. They do not automatically turn short workloads into faster workloads, as Blender and V-Ray demonstrate.
Installation and power requirements
Check the complete radiator path
- Confirm clearance for the radiator and fans, not merely a case’s “360 mm support” label.
- Check motherboard VRM heatsinks, memory, front drive cages and top-panel clearance.
- Verify whether front mounting changes GPU intake temperature or blocks other hardware.
- Measure tube reach and routing; the approximately 280 mm tube length is a planning figure, not universal proof of fit.
- Allow for the card’s 2.7-slot width and the radiator’s 55 mm assembly depth.
Plan the 16-pin power connection
MSI specifies a 1,000 W recommended PSU, one 16-pin GPU connector and a supplied four-8-pin adapter. A modern ATX 3.x/PCIe 5.x supply with a native high-power cable is preferable. Fully seat the connector, avoid a sharp bend close to the plug and account for total system demand and transient loads rather than sizing only to the GPU’s nominal figure.
Recommended Free Tools
Who should buy it?
It makes sense when
- Your case has a genuinely compatible 360 mm radiator position and sufficient tube clearance.
- You run sustained GPU workloads and prioritize lower temperatures and noise.
- You value premium construction, lighting and thermal headroom more than price efficiency.
- Your editing work uses Blackmagic RAW and benefits from the tested decode result.
Choose a Founders Edition or air-cooled partner card when
- You need the simplest installation or have no spare 360 mm mount.
- You mainly run Blender or V-Ray and expect a large speed increase; this page does not show one.
- Your workloads are short bursts where a large radiator offers little practical benefit.
- The MSI premium is substantial relative to a readily available RTX 5090 alternative.
What this review can—and cannot—prove in 2026
The review establishes that MSI’s sample was generally the fastest or tied in several launch-period tests, with small wins in MLPerf, most Procyon text models and Blackmagic RAW. It also establishes that Geekbench AI, Blender and V-Ray did not show a meaningful consistent advantage.
It does not establish current retail pricing, stock, firmware, driver behavior, mature Blackwell TensorRT performance, DaVinci Resolve results, Adobe export times, AV1 encoding, VRAM capacity limits in specific models or multi-hour production stability. Those factors should be checked against current software and your own workload before purchase. MSI’s official specifications are listed on its product page.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
