AI workstation builder
AI workstations, configured around the GPUs.
1× RTX 5090 through 4× RTX PRO Blackwell. Quote first; payment by ACH.Select a build
GPU count
Loading the AI workstation builder…
The AI workstation configurator is not yet available.
GPUs, memory, and storage are being added to live inventory. To specify a system now, email sales@jncs.com or call (585) 388-8780, or see AI workstations & servers for larger systems.
GPUs, memory, and storage are being added to live inventory. To specify a system now, email sales@jncs.com or call (585) 388-8780, or see AI workstations & servers for larger systems.
Build comparison
All three builds share the same assembly, burn-in, and quote process. They differ in total VRAM, which sets the largest model that can be loaded, and in the platform: PCIe lanes, memory channels, and power delivery sized to the GPU count.
Sizing reference: a 30B-parameter model at 4-bit quantization fits in 32GB of VRAM. A 70B model requires approximately 48GB at 4-bit and fits in 96GB at 8-bit. Context length and batch size add to these figures.
- Select a buildThen set the GPU count.
- Configure componentsPhysical fit, memory, and power are validated on every selection.
- Request a quoteNo payment at this stage. We confirm components and pricing and send a quote by email.
- Approve and pay by ACHThe system is assembled, burn-in tested under load, and shipped with Windows installed and updated.
Select a component
Three AI workstation builds
Each build is assembled by hand in Fairport, NY, stress-tested under sustained full load, and shipped with Windows 11 Pro installed, a 3-year warranty, and lifetime technical support. They differ in total VRAM, which sets the largest model that can be loaded, and in the platform sized to the GPU count.
JNCS AI 5090RTX 5090 AI workstationOne GeForce RTX 5090, 32GB VRAM, Ryzen 9000 and X870E, up to 128GB DDR5. Standard 15A circuit.
JNCS AI PRORTX PRO 6000 Blackwell workstationOne or two RTX PRO Blackwell GPUs, up to 192GB VRAM, Threadripper 9000, up to 512GB registered ECC DDR5.
JNCS AI MAXMulti-GPU AI workstationUp to four RTX PRO Blackwell GPUs, up to 384GB VRAM, Threadripper PRO 9000, up to 1TB registered ECC DDR5.
Sizing guideLocal LLM workstationHow much VRAM a model needs by parameter count and quantization, and which build provides it.
Guides by workload and buyer
Fine-tuningLLM fine-tuning workstationVRAM for full fine-tuning, LoRA, and QLoRA, by model size, and the build that fits.
Generative mediaImage and video generationVRAM for diffusion models, and one generation queue per GPU on multi-GPU builds.
Deep learningDeep learning workstationChoosing GPUs, CPU cores, memory, and storage for training.
ComparisonRTX 5090 vs RTX PRO 600032GB vs 96GB side by side, plus the RTX PRO 4500 and 5000.
2× RTX PRO 6000Dual RTX PRO 6000 workstation192GB of VRAM on two 96GB GPUs, and the circuit it needs.
Research groupsUniversities and labsSpec sheets and RFQs quoted as written, NDAs, matched sets.
For businessPrivate on-premises AIRun LLMs on your own hardware, sized for your team.
Larger systemsAI serversMore than four GPUs, rackmount, and large storage arrays.
GPU options and total VRAM
| GPU | Per card | 1 GPU | 2 GPUs | 4 GPUs |
|---|---|---|---|---|
| GeForce RTX 5090 (AI 5090) | 32GB GDDR7, 575W | 32GB | n/a | n/a |
| RTX PRO 4500 Blackwell | 32GB GDDR7, 200W | 32GB | 64GB | 128GB |
| RTX PRO 5000 Blackwell | 48GB GDDR7, 300W | 48GB | 96GB | 192GB |
| RTX PRO 5000 Blackwell | 72GB GDDR7, 300W | 72GB | 144GB | 288GB |
| RTX PRO 6000 Blackwell Workstation Edition | 96GB GDDR7, 600W | 96GB | 192GB | 384GB |
Two-GPU configurations are offered on the AI PRO and AI MAX; four-GPU configurations on the AI MAX. Every GPU in a multi-GPU build runs on a dedicated PCIe 5.0 x16 slot.
Power and circuit requirements
| Configuration | Estimated draw at full load | Power supply | Circuit |
|---|---|---|---|
| 1× RTX 5090 | 795W to 845W | 1200W | Standard 15A, 120V |
| 1× RTX PRO card, or 2× RTX PRO 4500 or 5000 | 650W to 1,050W | 1600W | Standard 15A, 120V |
| 2× RTX PRO 6000, or 4× RTX PRO 4500 or 5000 | 1,250W to 1,650W | 3000W | Dedicated 20A, 110–120V required; 30A, 220–240V recommended |
| 4× RTX PRO 6000 | approximately 2,850W | 3000W | Dedicated 30A, 220–240V required |
AI workstation questions
What is the difference between the three AI workstation builds?
Total VRAM and the platform sized to the GPU count. The JNCS AI 5090 runs one GeForce RTX 5090 (32GB) on Ryzen 9000. The JNCS AI PRO runs one or two RTX PRO Blackwell GPUs, up to 192GB, on Threadripper 9000 with quad-channel registered ECC memory. The JNCS AI MAX runs up to four, up to 384GB, on Threadripper PRO 9000 with eight memory channels and 128 PCIe 5.0 lanes.
How much VRAM do I need for a local LLM?
Model weights take roughly parameters times bits per weight divided by 8, plus overhead for context. A 30-billion-parameter model at 4-bit quantization fits in 32GB of VRAM. A 70B model requires approximately 48GB at 4-bit and fits in 96GB at 8-bit. See the local LLM workstation guide.
What electrical circuit does an AI workstation need?
Single-GPU builds, and two RTX PRO 4500 or 5000 cards, run on a standard 15A, 120V circuit. Two RTX PRO 6000 cards, or four RTX PRO 4500 or 5000 cards, require a dedicated 20A, 110-120V circuit, with a 30A, 220-240V circuit recommended. Four RTX PRO 6000 cards draw approximately 2,850W and require a dedicated 30A, 220-240V circuit.
Are prices shown in the configurator?
Yes. Every part shows its price, and the build summary shows an estimated total that includes assembly, burn-in, and testing. GPU and memory pricing changes frequently, so a technician confirms final pricing and lead time by email, typically within 1-2 business days. There is no cost and no obligation.
How do I pay for an AI workstation?
Payment is by ACH bank transfer. Credit cards are not accepted for AI workstations. An invoice with bank details is issued once the quote is approved.
Do you build systems with more than four GPUs?
The configurator covers towers with up to four GPUs. Systems with more than four GPUs and rackmount servers are specified directly with a technician. See AI servers and larger systems.