HP OMEN 45L Review (2026): The RTX 5090 PC for AI Video Generation

This HP OMEN 45L review covers the machine to buy when your AI work is visual. The OMEN 45L, model GT22, is a large gaming tower built around an NVIDIA GeForce RTX 5090 with 32GB of GDDR7 memory. When your work is images or video, raw graphics power wins outright.

That 32GB card chews through image batches. More importantly, it makes local video generation practical rather than theoretical. It also runs 30B to 70B language models in quantised form. And because it is a proper tower with full CUDA support, everything in the AI ecosystem simply works.

It is loud and power hungry next to the Apple and mini-PC options. Still, nothing else here touches it for creative generation.

Expect to pay about $3,959 for the RTX 5090 configuration with a 2TB drive. Pricing runs across a wide band, though, roughly $3,600 to $4,800, depending on storage, memory and the deals of the week. So it is one of the pricier picks here. Still, it is also the only one built for high-end visual generation.

Worth knowing before you buy: much of that cost is the RTX 5090 itself. That card is expensive and often hard to find on its own. So getting it inside a complete, warrantied, liquid-cooled system takes the sourcing headache away. For many buyers, that is a real part of the value.

This is the reason the OMEN 45L exists. The RTX 5090’s 32GB of memory is one of the few consumer setups that can run modern local video pipelines without heavy compromise. For example, models like Wan 2.1 14B run natively. And HunyuanVideo fits comfortably in a quantised form. So you can generate short AI video clips on your own machine, instead of renting a cloud GPU.

That is the practical line between this and every unified-memory rival in the lineup. The Apple and AMD machines hold bigger language models. But local video generation is a CUDA game, and CUDA belongs to NVIDIA. So if moving pictures are on your to-do list, this is the machine.

On the language side, the 32GB card handles a lot, though its memory is the ceiling. At common 4-bit quantisation:

• 8B models fly at around 145 tokens per second, far faster than you can read.
• 32B models run near 45 tokens per second, comfortable for real work.
• 70B models land around 15 to 23 tokens per second, but only at tight quantisation, since a 70B barely fits 32GB. They run, but with little headroom.

So the honest summary is simple. For language models, 32B and below is the comfortable sweet spot here, and 70B is possible rather than pleasant. If your priority were the very biggest language models, a 128GB unified-memory machine would suit you better. But this tower earns its price on speed and on visual work.

This is where the OMEN 45L pulls clear. In image generation, the RTX 5090 produces a 1024px SDXL image in roughly 2 seconds. A more demanding Flux image takes around 9 seconds. Meanwhile, the full CUDA toolchain of ComfyUI and every popular front end is available without workarounds.

In video, as above, its 32GB runs Wan 14B and HunyuanVideo locally, where lesser cards have to offload and stall. Combined with the Blackwell architecture’s native FP4 support, this is comfortably the fastest mainstream desktop in the lineup for anyone whose output is pictures and clips rather than text.

Key HP OMEN 45L specs for the RTX 5090 configuration:

The 45-litre tower is a serious piece of hardware. It has tempered glass, tool-less side access and HP’s Cryo Chamber liquid cooler, which seals the radiator in its own chamber to feed it cooler air. The payoff is real. Reviewers measured some of the lowest CPU temperatures among current RTX 5090 systems.

What it is not, is quiet or frugal. The RTX 5090 alone can pull up to 575W. And the whole system draws far more than an Apple desktop’s 75W or so. So under a real load, you hear it and you feel it in the room. This is an office or studio machine, not a bedroom one. Its appetite for power is the price of that speed.

Two things are worth knowing before you buy. First, independent reviewers, notably GamersNexus, found HP running the Core Ultra 9 a little below its rated clocks in this prebuilt. So the processor underperforms the same chip in a well-tuned system. For GPU-bound AI work, which is most of what this machine is for, that matters little. But for heavy CPU tasks, it is a genuine mark against it.

Second, value. Because it is built around a card you can technically buy yourself, a carefully sourced do-it-yourself 5090 rig can cost less and clock the CPU correctly. So you are paying a premium for assembly, cooling, warranty and not having to hunt down a 5090. Whether that premium is worth it depends on how much you value a turnkey machine.

Both are serious AI machines, but they solve different problems. The OMEN 45L is the CUDA workhorse for local image and video generation, while the Mac Studio M4 Max is the quieter, more efficient choice once your models grow past what 32GB can hold. Here is how the two compare side by side.

In short, pick the OMEN 45L for creative generation and raw speed, and the Mac Studio M4 Max for silent, power-sipping large language model work.

What we liked

• The fastest mainstream machine here for local image and video generation
• A 32GB RTX 5090 runs Wan 14B and HunyuanVideo locally where lesser cards stall
• Full CUDA, so every diffusion, video and PyTorch pipeline works without workarounds
• Very quick image generation, roughly a 2-second SDXL image and a 9-second Flux image
• Genuinely well cooled: the Cryo Chamber posts some of the lowest CPU temps in its class
• A big, upgradeable tower with tool-less access and a 1200W ATX 3.0 power supply
• Turnkey and warrantied, which removes the pain of sourcing a scarce RTX 5090

What could be better

• Loud and power hungry, with the GPU alone drawing up to 575W, unlike the Apple and mini-PC picks
• Reviewers found HP running the Core Ultra 9 CPU below its rated clocks in this prebuilt
• 32GB of memory caps the biggest language models; a 70B only just fits, heavily quantised
• The non-modular power supply and some proprietary wiring complicate the deepest upgrades
• Weak value against a carefully sourced do-it-yourself 5090 build

Buy it if your work is image or video heavy, and you want the fastest mainstream way to generate locally. It also suits you if you want the flexibility to open the case and upgrade later. So it fits creators and researchers who live in ComfyUI and diffusion tools, plus anyone who wants a real gaming PC alongside their AI work.

Skip it if you need the very largest language models. There, a 128GB GMKtec EVO-X2 goes further. Skip it too if silence and low power are non-negotiable. There, the Mac Studio M4 Max is the calmer, more efficient choice.

Overall, the HP OMEN 45L is the creative generation champion of this lineup. Its RTX 5090 makes local video and image work practical in a way no unified-memory machine can match. Meanwhile, CUDA means nothing is off limits, and the tower is well cooled and genuinely upgradeable. Still, the price of that power is honest and loud. It draws real wattage, it is audible under load, and HP’s CPU tuning leaves performance on the table.

Judged as the machine for AI video generation in our best desktop for AI lineup, it does its job better than anything else here. So we rate the HP OMEN 45L 4.2 out of 5. It is the fastest creative-AI desktop for the money, marked down for noise, power and a throttled CPU.