Which build do you need?
- 1Use
- 2What
- 3Size
Compare every model, app and build
Every model, every build
Each model at its most common precision. A check means it runs fully on the GPU.
| Model | 16 GBRTX 5060 Tifrom $1,499 | 24 GBRTX PRO 4000from $7,999 | 32 GBRTX 5090from $11,499 | 96 GBRTX PRO 6000from $19,499 | 192 GB2× RTX PRO 6000from $92,998 |
|---|---|---|---|---|---|
| Llama 3.1 8B8-bit · ~10 GB | |||||
| Qwen3 8B8-bit · ~10 GB | |||||
| Gemma 3 12B4-bit · ~11 GB | |||||
| Qwen3 14B4-bit · ~11 GB | |||||
| gpt-oss-20bAs released · ~14 GB | |||||
| Mistral Small 3.24-bit · ~17 GB | – | ||||
| Gemma 3 27B4-bit · ~19 GB | – | ||||
| Qwen3 Coder 30B-A3B4-bit · ~20 GB | – | ||||
| Qwen3 32B4-bit · ~22 GB | – | ||||
| DeepSeek-R1 Distill Llama 70B4-bit · ~44 GB | – | – | – | ||
| Llama 3.3 70B4-bit · ~44 GB | – | – | – | ||
| gpt-oss-120bAs released · ~66 GB | – | – | – | ||
| Qwen3 235B-A22B4-bit · ~130 GB | – | – | – | – | |
| Llama 3.1 405B4-bit · ~230 GB | Bigger than one build. Ask about a custom multi-GPU build | ||||
| Stable Diffusion XLFull precision · ~12 GB | |||||
| FLUX.1 [dev]Full precision · ~28 GB | – | – | |||
| Whisper large-v3Full precision · ~5 GB | |||||
| Wan 2.2 TI2V-5BAs released · ~24 GB | – | ||||
| Wan 2.2 T2V-A14BAs released · ~80 GB | – | – | – | ||
Approximate GPU memory for the model plus a working context. Longer context and more users need more. Models that don't fit can run partly on the CPU and system memory, much more slowly. Columns show builds with an NVIDIA card, which run every model listed. Our Radeon AI PRO builds also have 32 GB, for language models.
Every app, every workstation
Each project size against each workstation. A check means it meets the size.
| Project | RTX 5060 TiRyzen 7$1,499 | RTX 5070 TiRyzen 7$2,899 | RTX 5070 TiUltra 7$3,999 | RTX 5080Ryzen 7$4,099 | RTX 5080Ultra 9$6,499 | RTX PRO 4000Ultra 9$7,999 | RTX 5090Ultra 9$11,499 | RTX PRO 6000Ryzen 9$19,499 | RTX PRO 6000Ultra 9$28,799 |
|---|---|---|---|---|---|---|---|---|---|
| HD and 4K editingDaVinci Resolve | |||||||||
| 4K with Fusion and AI toolsDaVinci Resolve | – | – | – | – | |||||
| 6K and 8KDaVinci Resolve | – | – | – | – | – | – | |||
| HD editingPremiere Pro | |||||||||
| 4K editingPremiere Pro | |||||||||
| 6K and 8KPremiere Pro | – | – | – | – | |||||
| HD motion graphicsAfter Effects | |||||||||
| 4K compositingAfter Effects | |||||||||
| 3D and multi-frame renderingAfter Effects | – | – | – | – | |||||
| Modeling and small scenesBlender | |||||||||
| Large scenesBlender | – | – | – | – | – | – | |||
| Film-scale scenesBlender | – | – | – | – | – | – | – | ||
| Modeling and motion graphicsCinema 4D | |||||||||
| Complex scenes in RedshiftCinema 4D | |||||||||
| Heavy, high-resolution rendersCinema 4D | – | – | – | – | – | – | |||
| Learning and small projectsUnreal Engine | |||||||||
| Full games with long lighting buildsUnreal Engine | – | – | – | – | |||||
| Archviz, VR and several 4K screensUnreal Engine | – | – | – | – | – | – | |||
| Parts and small assembliesSOLIDWORKS | – | – | – | – | – | – | |||
| Medium assembliesSOLIDWORKS | – | – | – | – | – | – | |||
| Large assembliesSOLIDWORKS | – | – | – | – | – | – | – | ||
| 2D draftingAutoCAD | |||||||||
| 3D modelingAutoCAD | – | – | – | – | – | – | |||
| Large models and point cloudsAutoCAD | – | – | – | – | – | – | |||
| Small projectsRevit | |||||||||
| Large, complex modelsRevit | – | – | – | – | |||||
| GPU renderingRevit | – | – | – | – | – | – | – | ||
| Photos and designsPhotoshop | |||||||||
| Files of 500 MB to 1 GBPhotoshop | |||||||||
| Files over 1 GBPhotoshop | |||||||||
| Editing and exportsLightroom Classic | |||||||||
| AI Denoise and Lens BlurLightroom Classic | |||||||||
| Large catalogs with Photoshop openLightroom Classic |
Project sizes are our recommendation, from each maker's requirements and Puget Systems' hardware guides. A workstation meets a size when its GPU memory, system memory, card type and GPU score all reach it.
Sizing questions
What do 4-bit and 8-bit mean?
They're compressed versions of a model. 4-bit needs about a quarter of the memory of full precision with a small drop in quality, and 8-bit is close to full quality at about half the memory. Most people run large language models at 4-bit or 8-bit.
What happens if a model doesn't fit?
Tools like Ollama and llama.cpp can run part of the model on the CPU and system memory, but it runs much more slowly. For everyday use, pick a build where the model fits fully in GPU memory.
Why leave room to spare?
Long prompts, big documents and several people at once all need GPU memory on top of the model itself. A few GB of headroom keeps responses fast.
Can two GPUs share one model?
Yes. On a multi-GPU server, inference software like vLLM splits one model across the cards, so it can use their combined memory.
Why do the CAD sizes ask for an RTX PRO card?
Dassault Systèmes advises a certified workstation card for SOLIDWORKS, and only certified cards get official support for features like RealView. In Puget Systems' tests, even a low-end professional card beat the fastest GeForce card in some views.
Where do the GPU scores come from?
For now, from Puget Systems' published tests, measured on their test bench rather than on our builds. As we record our own results they replace Puget's, and the finder, the bars on each build and the solution pages all update together.
Where do the Radeon AI PRO builds fit?
They are our lowest-priced builds with 32 GB of GPU memory, and the finder picks them for chat, reasoning and coding models. They run those through AMD's ROCm software. For image, video and speech models, and in the Software tab, the finder only picks builds with an NVIDIA card, because many of those tools need NVIDIA CUDA.
How do you size a streaming PC?
By what the stream asks of the graphics card and processor. Each setup lists the hardware encoders, graphics card, processor cores and memory it needs, and every GeForce build is checked against it. Encoder counts are NVIDIA's: three on the RTX 5090, two on the RTX 5080 and RTX 5070 Ti, one on the other cards.
