# Reference builds and worked designs

> UpStation catalog · Home › Reference builds and worked designs · https://ai.upstation.io/designs

Every price on this site is on the [homepage](https://ai.upstation.io/). Each product type and each product has its own page with the full details (spec, options, Intel/AMD equivalents, power and colo): [GPU servers](https://ai.upstation.io/gpu-servers) · [GPU hosts without GPUs](https://ai.upstation.io/gpu-hosts) · [Switches, cables and optics](https://ai.upstation.io/switches) · [Compute nodes](https://ai.upstation.io/compute-nodes) · [Storage nodes](https://ai.upstation.io/storage-nodes) · [Parts rate card](https://ai.upstation.io/rate-card) · [Power and colo](https://ai.upstation.io/power) · [How solutions are sized](https://ai.upstation.io/sizing) · [Reference builds and worked designs](https://ai.upstation.io/designs). AI assistants: the map is [llms.txt](https://ai.upstation.io/llms.txt); the data is [catalog.json](https://ai.upstation.io/catalog.json); every page is also Markdown (append .md, or send Accept: text/markdown).

Every build below is computed from the catalog data and the [sizing rules](https://ai.upstation.io/sizing); use them to check arithmetic. Prices in USD as of 2026-09-09; final price confirmed on order. Continental US; Canada quoted case by case. Orders and quotes: [book a meeting](https://calendly.com/alecglover/intro-call). Every page: [llms.txt](https://ai.upstation.io/llms.txt) (map) · [homepage](https://ai.upstation.io/) (all tables) · [catalog.json](https://ai.upstation.io/catalog.json) (data).

### One 8× H200 node, 3-year colo

| Item | Qty | Unit price | Extended | Provisioned kW |
|---|---|---|---|---|
| 8x NVIDIA H200 (HGX H200), Intel — SYS-821GE-TNHR | 1 | $319,421.64 | $319,421.64 | 11.34 |
| Hardware total |  |  | $319,421.64 | 11.34 |
| Colo, 3-year term, $225/kW/month × 11.34 kW |  |  | $2,552/month |  |
| Colo over 36 months |  |  | $91,872 |  |
| Hardware + colo over 36 months |  |  | $411,293.64 |  |

### Four 8× B300 nodes, 5-year colo

| Item | Qty | Unit price | Extended | Provisioned kW |
|---|---|---|---|---|
| 8x NVIDIA B300 (HGX B300 NVL8), Intel — SYS-822GS-NB3RT-01-G2 | 4 | $572,364 | $2,289,456 | 67.84 |
| Hardware total |  |  | $2,289,456 | 67.84 |
| Colo, 5-year term, $175/kW/month × 67.84 kW |  |  | $11,872/month |  |
| Colo over 60 months |  |  | $712,320 |  |
| Hardware + colo over 60 months |  |  | $3,001,776 |  |

### Two 8× RTX PRO 6000 nodes, one 48× 25G switch, 1-year colo

| Item | Qty | Unit price | Extended | Provisioned kW |
|---|---|---|---|---|
| 8x NVIDIA RTX PRO 6000 Blackwell Server Edition, Intel — SYS-422GA-NRT | 2 | $191,888.10 | $383,776.20 | 16.94 |
| +1× 25G 2× SFP28 CX6 NIC (listed option, one per node) | 2 | $391 | $782 | — |
| 48× 25G + 8× 100G switch — S5470-48Y, PicOS | 1 | $6,049 | $6,049 | 0.19 |
| 25G 2 m passive DAC (2 per node) | 4 | $39 | $156 | — |
| Hardware total |  |  | $390,763.20 | 17.13 |
| Colo, 1-year term, $250/kW/month × 17.13 kW |  |  | $4,282/month |  |
| Colo over 12 months |  |  | $51,384 |  |
| Hardware + colo over 12 months |  |  | $442,147.20 |  |

### Worked design: 4× 8× B300 training cluster (32 GPUs), 800G fabric, 200G front-end, 5× NVMe storage nodes, 5-year colo

Inputs: 32 GPUs for multi-node training; compressed image/video data (0.5 GB/s per GPU aggregate); 70B-parameter model, synchronous checkpoints with a 60 s pause; 200 TB dataset, 4 checkpoints retained; buyer-supplied Ceph, replica 3; 40 kW racks; 5-year colo

| Item | Qty | Unit price | Extended | Provisioned kW |
|---|---|---|---|---|
| 8x NVIDIA B300 (HGX B300 NVL8), Intel — SYS-822GS-NB3RT-01-G2 | 4 | $572,364 | $2,289,456 | 67.84 |
| +2× 200G CX7 north-south (listed option, one per node: front-end ports) | 4 | $3,848 | $15,392 | — |
| 32× 800G OSFP switch — N8650-32OD, PicOS (GPU fabric, 1:1) | 1 | $37,499 | $37,499 | 1.25 |
| 800G link, 2× SR8 optic + MTP-12 OM4 10 m (1 per GPU port; every fabric link estimated optical) | 32 | $1,908 | $61,056 | — |
| 24× 200G + 8× 400G switch — N8510-24CD8D, PicOS (front-end pair, MLAG) | 2 | $12,400 | $24,800 | 1.5 |
| 200G AOC 10 m (GPU nodes to the front-end pair, 2 per node) | 8 | $1,034 | $8,272 | — |
| 200G→2×100G breakout 2 m (storage nodes to the front-end pair, 2 nodes per cable) | 6 | $119 | $714 | — |
| 400G 2 m passive DAC (peer link between the pair) | 2 | $174 | $348 | — |
| 2U all-NVMe storage node, dual socket, 24x 7.68TB NVMe (184TB raw) — SYS-222C-TN | 5 | $182,994.59 | $914,972.95 | 8.25 |
| 48× 1G RJ45 + 4× 10G switch — S3410-48TS, PicOS (OOB; Cat6 patches not priced) | 1 | $1,089 | $1,089 | 0.07 |
| Hardware total |  |  | $3,353,598.95 | 78.91 |
| Colo, 5-year term, $175/kW/month × 78.91 kW |  |  | $13,809/month |  |
| Colo over 60 months |  |  | $828,540 |  |
| Hardware + colo over 60 months |  |  | $4,182,138.95 |  |

- GPUs → nodes: 32 ÷ 8 = 4 nodes. Multi-node training means the GPUs act as one, so HGX class (B300, 288 GB per GPU, 2.3 TB per node); a PCIe server would not serve it. Host memory stays at the listed 2TB (no offload stated).
- GPU fabric: 1 port per GPU = 32 × 800G. 32 ≤ 32, so one 32× 800G OSFP switch is non-blocking at 1:1 (single point of failure, accepted; growth past 4 nodes needs the 64-port switch or a fat-tree: at 32 nodes, 8 leaves + 4 spines on 64-port switches). Links: all 32 GPU ports priced optical (2× SR8 + MTP-12 fibre, $1,908 per link) — at 2 nodes per 40 kW rack the cluster spans 2 racks, so most links leave the switch's rack and 800G passive DAC (3 m reach) cannot be assumed; swapping the 16 genuinely in-rack links to DAC is a final-engineering saving, not part of the estimate. No storage on this fabric.
- Front-end (converged with storage): each node takes the listed +2× 200G option and connects one port to each switch of an MLAG pair of 24× 200G + 8× 400G switches, 3:1 or better; the storage nodes' 2× 100G ports land on the same pair through 200G→2×100G breakouts (5 nodes × 2 ports = 10 ports, 3 breakout cables per switch); peer link 2× 400G DAC. Uplinks to the buyer's core are not priced. Storage stays converged because 16 GB/s across 4 nodes is far below the 40 GB/s-per-node dedicated-fabric threshold.
- Storage demand: read = 32 GPUs × 0.5 GB/s = 16 GB/s (aggregate, not the 4–5 GB/s single-node peak); checkpoint burst = 70 × 16 GB ÷ 60 s = 18.7 GB/s; write = max(½ × 16, 18.7) = 18.7 GB/s; capacity = 200 TB + 4 × 1.12 TB = 204.5 TB.
- Storage supply: 2U all-NVMe storage node, dual socket, 24x 7.68TB NVMe (184TB raw) = 184 TB raw, 20 GB/s planning per node. Replica 3 on N nodes plans to 184 × N × (N−1)/N × 0.333 × 0.85: N = 4 gives 156.2 TB (short), N = 5 gives 208.3 TB ≥ 204.5 TB. Throughput 5 × 20 = 100 GB/s ≥ 16. Erasure coding 4+2 would need 6 hosts and plan to 521.6 TB, 2.6× the need, so replica 3 on 5 nodes is the answer. Software (Ceph) is buyer-supplied and not priced.
- OOB: 4 GPU nodes + 5 storage nodes + 3 switches = 12 × 1G ports on one 48× 1G RJ45 + 4× 10G switch.
- Racks at 40 kW: B300 = 8 U and 16.96 kW provisioned, so 2 per rack (power binds). Rack A: 2× B300 + GPU switch + OOB switch = 35.24 kW; rack B: 2× B300 = 33.92 kW; rack C: 5× storage + front-end pair = 9.75 kW, 12 U. Racks, PDUs and cabling labour are quoted separately.
- Power and colo: 78.91 kW provisioned in total; 5-year colo $175/kW/month × 78.91 kW = $13,809/month. Management nodes (3× the listed 1U compute node) are recommended at this size and not included here.

### Worked design: 2 PB usable archive on 5× HDD storage nodes (RAID 6), 100G front-end pair, 5-year colo

Inputs: 2,000 TB usable, archive / cold data, RAID 6 protection, no stated throughput; buyer-supplied file software (ZFS/NFS or similar); 40 kW racks; 5-year colo

| Item | Qty | Unit price | Extended | Provisioned kW |
|---|---|---|---|---|
| 2U HDD storage node, single socket, 24x 24TB SAS (576TB raw) — ASG-2015S-E1CR24H | 5 | $51,743.45 | $258,717.25 | 3.9 |
| 32× 100G QSFP28 switch — N8560-32C, PicOS (front-end pair, MLAG) | 2 | $7,919 | $15,838 | 0.62 |
| 100G 2 m passive DAC (storage nodes to the pair, 2 per node) | 10 | $54 | $540 | — |
| 100G 2 m passive DAC (peer link between the pair) | 2 | $54 | $108 | — |
| 48× 1G RJ45 + 4× 10G switch — S3410-48TS, PicOS (OOB; Cat6 patches not priced) | 1 | $1,089 | $1,089 | 0.07 |
| Hardware total |  |  | $276,292.25 | 4.59 |
| Colo, 5-year term, $175/kW/month × 4.59 kW |  |  | $803/month |  |
| Colo over 60 months |  |  | $48,180 |  |
| Hardware + colo over 60 months |  |  | $324,472.25 |  |

- Per node: 2U HDD storage node, single socket, 24x 24TB SAS (576TB raw) = 576 TB raw. RAID 6 as two 12-drive groups (10+2) = 0.833 efficiency; plannable = 576 × 0.833 × 0.85 (fill ceiling) = 407.8 TB per node.
- Nodes: ceil(2,000 ÷ 407.8) = 5 nodes = 2,039.0 TB plannable (1.02× the need, inside the 2× overshoot check). Throughput 5 × 5 GB/s ≈ 25 GB/s streaming (estimate); no stated demand, so capacity binds. Each node is one failure domain: a node outage takes its 407.8 TB offline until repaired.
- Alternative, distributed across nodes with Ceph erasure coding 4+2 (0.667 efficiency, 6-host floor): plannable = 576 × N × (N−1)/N × 0.667 × 0.85; N = 7 gives 1,959.4 TB (short), N = 8 gives 2,285.9 TB, so 8 nodes ($413,947.60 for the nodes) plus NVMe DB/WAL drives at 1–4% of HDD capacity via the rate card. Survives a whole node.
- Network: each node has 2× 100G; one port to each switch of an MLAG pair of 32× 100G QSFP28 switches (10 ports of 32 per switch), peer link 2× 100G DAC, uplinks to the buyer's core not priced; the listed −$874 option swaps to 2× 25G if the core is 25G. OOB: 5 nodes + 2 switches on one 48× 1G RJ45 + 4× 10G switch.
- Rack and power: 5 × 2 U + 3 U of switches = 13 U, 4.59 kW provisioned: one rack, one 208V 1-ph 30A circuit pair. 5-year colo $803/month.

Generated 2026-09-22 from UpStation's catalog data; prices as of 2026-09-09; re-checked monthly. Book a meeting: https://calendly.com/alecglover/intro-call. Map of every page: https://ai.upstation.io/llms.txt