OnpremBench.aiby Understand.tech

Same chip, different box

NVIDIA DGX Station vs Exxact Valence DGX Station

Memory, bandwidth, software stack, price and the models each one can hold. Generated from the two sourced machine records; estimates are labelled.

Add a third machine
NVIDIA DGX Station
NVIDIA GB300

NVIDIA DGX Station

496 GB CPU + 252 GB GPU · 748 GB coherent

Open the record
Exxact Valence DGX Station
NVIDIA GB300

Exxact Valence DGX Station

496 GB CPU + 252 GB GPU

Open the record

What actually differs

Which models fit

Estimate · 4-bit · 8K context · 1 request
ModelNVIDIA DGX StationExxact Valence DGX Station
Llama 3.1 8BFits 12 GB≤ 1251 tok/sFits 12 GB≤ 1251 tok/s
Qwen3 14BFits 16 GB≤ 721 tok/sFits 16 GB≤ 721 tok/s
Qwen3 32BFits 28 GB≤ 338 tok/sFits 28 GB≤ 338 tok/s
Llama 3.3 70BFits 50 GB≤ 164 tok/sFits 50 GB≤ 164 tok/s
gpt-oss-20b (MoE)Fits 19 GB≤ 2871 tok/sFits 19 GB≤ 2871 tok/s
gpt-oss-120b (MoE)Fits 76 GB≤ 2008 tok/sFits 76 GB≤ 2008 tok/s
Qwen3 235B-A22B (MoE)Fits 147 GB≤ 499 tok/sFits 147 GB≤ 499 tok/s
DeepSeek V3 / R1 671B (MoE, MLA)Too large 404 GBToo large 404 GB

Speed is a memory-bandwidth ceiling for one request, not a measured result. Change context, users or precision in the finder

Specifications

Sourced and dated
NVIDIA DGX StationExxact Valence DGX Station
Chip / acceleratorGB300 Grace Blackwell Ultra Desktop SuperchipGB300 Grace Blackwell Ultra Desktop Superchip
PlatformNVIDIA GB300NVIDIA GB300
Model memory496 GB CPU + 252 GB GPU · 748 GB coherent496 GB CPU + 252 GB GPU
Memory bandwidthGPU 7.1 TB/s · CPU 396 GB/s · NVLink-C2C 900 GB/sGPU 7.1 TB/s · CPU 396 GB/s
Software stackCUDACUDA
Operating systemNVIDIA DGX OS (Ubuntu) · Windows edition announced for Q4 2026NVIDIA DGX OS (Ubuntu)
Storage4 × M.2 PCIe Gen 5 slots · drives configured by the partner2 × PCIe 5.0 M.2 + 2 × PCIe 6.0 M.2 slots · drives configured at order
NetworkingConnectX-8 SuperNIC · 2 × 400 GbE QSFP112 · 10 GbE · 1 GbE BMCConnectX-8 · 2 × 400 GbE QSFP112 · 10 GbE · 1 GbE BMC
Form factorDeskside workstationDeskside workstation
DimensionsSet by each partner’s chassisNot published in source extract
WeightSet by each partner’s chassisNot published in source extract
Power rating (not consumption)1,600 W platform system-power specification · 20 A circuit specified by NVIDIA1,600 W NVIDIA platform system-power specification · installed PSU needs confirmation
Supplier list pricePrice from the manufacturer · No public list price · manufacturer page≈ $95k · Supplier list price · United States · Sept 2026
AvailabilityOrder through partner manufacturers · configuration and price set by each partnerConfigure and quote online · lead time set by the seller
Model configurations in the Lab44
Setups in the Lab33
Record reviewed2026-09-172026-09-17

Sources: NVIDIA · Exxact. Prices are supplier snapshots, not a live feed.

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.

Exxact Valence DGX Station

  • One of the few GB300 workstations with a public starting price. The starting configuration excludes drives, the optional display GPU and support options; the configurator sets the final price.
  • Same GB300 superchip, memory and networking as every DGX Station; the chassis, cooling and warranty are Exxact’s.
  • Supports one double-wide graphics card for display alongside the GB300.
Own one of these? A measured run (model revision, runtime, context, tokens per second) is worth more than this whole page. Publish it.