Deskside and tower workstations
NVIDIA DGX Station vs HP Z8 Fury G6i
Memory, bandwidth, software stack, price and the models each one can hold. Generated from the two sourced machine records; estimates are labelled.
NVIDIA GB300
NVIDIA DGX Station
496 GB CPU + 252 GB GPU · 748 GB coherent
Open the record
NVIDIA RTX PRO
HP Z8 Fury G6i
96 GB per GPU · up to 4 GPUs
Open the recordWhat actually differs
- Different software stacks: the NVIDIA DGX Station runs CUDA, the HP Z8 Fury G6i runs CUDA. Check that your runtime and model format are supported before comparing numbers.
- The NVIDIA DGX Station has 156 GB more model memory (252 GB vs 96 GB). That decides which models fit at all.
- Memory bandwidth is not in the record for the HP Z8 Fury G6i, so the Lab cannot compare generation speed ceilings.
- No public list price in the record for either machine. Ask the supplier for a configured quote.
- Form factor: Deskside workstation vs Expandable tower.
Which models fit
Estimate · 4-bit · 8K context · 1 request| Model | NVIDIA DGX Station | HP Z8 Fury G6i |
|---|---|---|
| Llama 3.1 8B | Fits 12 GB≤ 1251 tok/s | Fits 12 GBBandwidth not in record |
| Qwen3 14B | Fits 16 GB≤ 721 tok/s | Fits 16 GBBandwidth not in record |
| Qwen3 32B | Fits 28 GB≤ 338 tok/s | Fits 28 GBBandwidth not in record |
| Llama 3.3 70B | Fits 50 GB≤ 164 tok/s | Fits 50 GBBandwidth not in record |
| gpt-oss-20b (MoE) | Fits 19 GB≤ 2871 tok/s | Fits 19 GBBandwidth not in record |
| gpt-oss-120b (MoE) | Fits 76 GB≤ 2008 tok/s | Fits 76 GBBandwidth not in record |
| Qwen3 235B-A22B (MoE) | Fits 147 GB≤ 499 tok/s | Too large 147 GB |
| DeepSeek V3 / R1 671B (MoE, MLA) | Too large 404 GB | Too large 404 GB |
Specifications
Sourced and dated| NVIDIA DGX Station | HP Z8 Fury G6i | |
|---|---|---|
| Chip / accelerator | GB300 Grace Blackwell Ultra Desktop Superchip | Intel Xeon 600 · RTX PRO 6000 Blackwell Max-Q option |
| Platform | NVIDIA GB300 | NVIDIA RTX PRO |
| Model memory | 496 GB CPU + 252 GB GPU · 748 GB coherent | 96 GB per GPU · up to 4 GPUs |
| Memory bandwidth | GPU 7.1 TB/s · CPU 396 GB/s · NVLink-C2C 900 GB/s | Confirm selected configuration |
| Software stack | CUDA | CUDA |
| Operating system | NVIDIA DGX OS (Ubuntu) · Windows edition announced for Q4 2026 | Windows 11 Pro · confirm Linux option |
| Storage | 4 × M.2 PCIe Gen 5 slots · drives configured by the partner | Up to 104 TB · configured drives |
| Networking | ConnectX-8 SuperNIC · 2 × 400 GbE QSFP112 · 10 GbE · 1 GbE BMC | Confirm selected configuration |
| Form factor | Deskside workstation | Expandable tower |
| Dimensions | Set by each partner’s chassis | 439.1 × 219.4 × 558.8 mm |
| Weight | Set by each partner’s chassis | From 22.2 kg |
| Power rating (not consumption) | 1,600 W platform system-power specification · 20 A circuit specified by NVIDIA | 1,350 / 1,700 W single PSU by voltage · up to 2,700 W dual aggregate capacity |
| Supplier list price | Price from the manufacturer · No public list price · manufacturer page | Price from the manufacturer · No public list price · manufacturer page |
| Availability | Order through partner manufacturers · configuration and price set by each partner | Check regional availability |
| Model configurations in the Lab | 4 | 2 |
| Setups in the Lab | 3 | 1 |
| Record reviewed | 2026-09-17 | 2026-09-07 |
Before you choose
NVIDIA DGX Station
- 748 GB coherent memory is 252 GB HBM3e on the GPU plus 496 GB LPDDR5X on the CPU, joined by NVLink-C2C. The two tiers have very different bandwidths; large models spill from the fast tier into the slow one.
- NVIDIA does not sell the DGX Station directly: it is ordered through partner manufacturers, each with its own chassis, drives, warranty and price. Partner listings observed in 2026 ran from about $85,000 to $175,000 depending on configuration.
- NVIDIA specifies 1,600 W total system power and a 20 A circuit. Confirm OEM and regional installation requirements before choosing an office outlet. The GB300 can be paired with an RTX PRO 6000 Blackwell workstation GPU for graphics and additional compute.
HP Z8 Fury G6i
- Up to 2 TB DDR5 ECC system RAM is separate from GPU memory. The memory screen uses one 96 GB GPU.
- Multiple GPUs do not automatically form one memory pool. Framework support, sharding and interconnect overhead must be evaluated.
Own one of these? A measured run (model revision, runtime, context, tokens per second) is worth more than this whole page. Publish it.