Workstation build · full tower · New parts
a16z Personal AI Workstation
Cutting-edge 4x RTX Pro 6000 Blackwell workstation
Built by a16z · Source: a16z: Building a16z's Personal AI Workstation · Updated
- Total cost
- $43,568
- VRAM
- 384GB
- GPUs
- 4
- System RAM
- 256GB
What can this build run?
With 384GB of VRAM across 4 GPUs, this build comfortably runs models up to GLM-5.3-Flash (Q6_K, ~36 tokens/sec).
Parts list
Every part in the a16z Personal AI Workstation
We may earn a commission when you buy through links on this page. It never changes what we recommend. How we make money
| Part | Price | Buy |
|---|---|---|
GPU 4×NVIDIA RTX Pro 6000 Blackwell Max-Q | $35,960$8,990 each | Amazon |
CPU AMD Ryzen Threadripper PRO 7975WX | $4,253 | Amazon |
Memory 256GB ECC DDR5 RAM | $1,395 | Amazon |
Storage 2TB PCIe 5.0 NVMe SSD | $960 | Amazon |
Power Supply Thermaltake Toughpower GF3 1650W 80 PLUS Gold | $330 | Amazon |
Motherboard Gigabyte MH53-G40 | $1,250 | eBay |
Case Professional workstation case | — | |
| Total | $43,568 |
Prices at time of build (2025); check current prices. Individual parts add up to $44,148.
Capability
Models this build runs well
Estimated for a single user with 8K context, with the model split across all GPUs (for example with llama.cpp). Best quantization that fits, ranked by our capability score.
Best overall · also coding, reasoning, vision, largest that fits
GLM-5.3-FlashQ6_K · ~36 tokens/sec
Runner-up
DeepSeek V4 FlashFP4 · ~51 tokens/sec
Fastest good model
Qwen3.5 9BQ8_0 · ~80 tokens/sec
| Model | Quant | Speed | Fit |
|---|---|---|---|
| GLM-5.3-Flash | Q6_K | ~36 tok/s | Runs well |
| DeepSeek V4 Flash | FP4 | ~51 tok/s | Runs well |
| Qwen3.8 27B | Q8_0 | ~39 tok/s | Runs well |
| Qwen3.5 122B-A10B | Q8_0 | ~43 tok/s | Runs well |
| Qwen3.6 35B-A3B | Q8_0 | ~68 tok/s | Runs well |
| Gemma 4 31B | Q8_0 | ~36 tok/s | Runs well |
| gpt-oss-120b | MXFP4 | ~70 tok/s | Runs well |
| Gemma 4 26B A4B | Q8_0 | ~64 tok/s | Runs well |
| GLM-4.5-Air | Q8_0 | ~39 tok/s | Runs well |
| gpt-oss-20b | MXFP4 | ~73 tok/s | Runs well |
The build
About this build
a16z's under-desk workstation packs four RTX Pro 6000 Blackwell Max-Q GPUs for 384GB of VRAM on a Threadripper PRO platform with 256GB of ECC memory. It trades cloud scalability for full control, lower latency, and the privacy of running everything locally.
Why this build
In the era of foundation models, multimodal AI, LLMs, and ever-larger datasets, access to raw compute is still one of the biggest bottlenecks for researchers, founders, developers, and engineers. The cloud offers scalability, but a personal AI workstation delivers complete control over your environment, lower latency, custom configurations, and the privacy of running every workload locally.
This build pushes the limits of desktop AI computing with 384GB of VRAM (96GB per GPU), all in a shell that fits under your desk.
Source
Read the full write-up from a16z: Building a16z’s Personal AI Workstation with Four NVIDIA RTX 6000 Pro Blackwell Max-Q GPUs.
Specifications
Full specs
- GPU
- 4× RTX PRO 6000
- CPU
- AMD Ryzen Threadripper PRO 7975WX
- Motherboard
- Gigabyte MH53-G40
- Memory
- 256GB ECC DDR5
- Storage
- 2TB PCIe 5.0 NVMe SSD
- Power supply
- Thermaltake Toughpower GF3 1650W 80 Plus Gold
- Case
- Custom workstation case
- Form factor
- full tower
- GPU power
- 2,400W rated
Keep browsing
