Specifications
Every component, what it is for, and how it extends
The full Mixpeek Appliance build sheet by phase. Each row says what the part does, where it sits, and what it grows into, so nothing is bought twice.
These are planning ranges, not quotes. Hardware pricing, availability and qualification status all move. Validate the seller, the warranty, the qualification status and the interface before buying anything, and treat totals as a way to size a budget rather than as an offer.
| Phase | Component | Purpose | Extends to | Planning range |
|---|---|---|---|---|
| NOW | Seagate Exos X24 24 TB SATA HDDEnterprise 3.5-inch SATA disk; SATA preserves Mac-enclosure portability | Start building the corpus now; disks can later move into server storage | 4 drives = 96 TB raw; 8 = 192 TB; later migrate into 24-bay storage | $350-$500 each |
| NOW | OWC ThunderBay 8 or equivalent JBOD8-bay Thunderbolt SATA enclosure | Immediate bulk storage without throwaway disks | Fill to 8; later reuse for migration, backup or a cold copy | $850-$1,300 |
| NOW | Independent backupSeparate NAS, disks or cloud copy | Avoid a single-copy migration event | Grows with the corpus | Workload dependent |
| V1 | 27-32U enclosed vertical rackFull-depth, lockable, wheeled, high-airflow cabinet | A clean movable appliance | Add pods until full, then duplicate the rack | $1,000-$2,500 |
| V1 | Compute Pod 01: 4U chassisFull-depth GPU chassis with server airflow and serviceability | A replaceable AI compute unit | GPU #2 vertically; Compute-02/N horizontally | $700-$2,000 |
| V1 | RTX PRO 6000 Blackwell 96 GB96 GB ECC NVIDIA accelerator | Large coding, VLM and video inference, Mixpeek extraction, training and fine-tuning | A second 96 GB GPU; future RTX PRO, MGX, HGX or DGX nodes | $13,000-$15,000 |
| V1 | Threadripper PRO 9975WX32 cores, 64 threads, high PCIe capacity | FFmpeg, Ray CPU workers, preprocessing, tokenization, databases, feeding the GPU | Add CPU or GPU nodes; a higher-core node on measured need | $4,000-$5,000 |
| V1 | ASUS Pro WS WRX90E-SAGE SEECC RDIMM, PCIe 5.0, dual 10GbE, AST2600 BMC/IPMI | Expansion, RAM bandwidth and remote recovery | Reserve an x16 slot for GPU #2 and a ConnectX NIC | $1,200-$1,500 |
| V1 | 512 GB ECC DDR5 RDIMMRegistered ECC memory | Dataset staging, Ray, dataloaders, CPU offload, training | Scale toward ~2 TB; each node adds RAM | $1,500-$3,000 |
| V1 | 2 × 2 TB mirrored boot NVMeRedundant system volume | OS, configs and critical service state | Larger mirror or dedicated management storage later | $300-$600 |
| V1 | 8 TB model/dataset NVMeFast reusable local tier | Weights, model cache, tokenized and hot datasets | Add or enlarge enterprise NVMe per node | $700-$1,500 |
| V1 | 4-8 TB high-endurance scratch NVMeSeparate high-write tier | Frames and audio, checkpoints, optimizer state, temp tensors | Add NVMe; shared hot data moves to Lustre at V2 | $500-$1,500 |
| V1 | 1600 W+ PSU or OEM redundant PSUGPU-capable power | Stable compute power with headroom | GPU #2 where the power envelope permits | $500-$1,000 |
| V1 | Cooling and fansTR5/server cooling, front-to-back airflow | Sustained 24/7 operation | OEM or liquid cooling for denser future accelerators | $250-$700 |
| V1 | Storage Pod 01: 4U 24-bayHot-swap server with ECC, BMC, HBA/backplane, redundant PSU | Makes durable storage independent from compute | Populate 24 bays; Storage-02/N; dense JBOD later | $4,000-$8,000 before disks |
| V1 | 12 × 20-24 TB enterprise SATA HDDsThe initial object pool | Canonical media, Mixpeek objects, datasets and artifacts | ~240-288 TB raw; fill 24 bays; then PB shelves | $4,000-$7,000 total |
| V1 | ZFS + S3-compatible object serviceDurable implementation behind a stable object API | A canonical s3:// namespace independent of disk topology | The backend can become distributed object storage | $0+ software |
| V1 | 100 GbE ConnectX-class NICsHigh-speed node adapters | Keep storage traffic from starving the GPUs | 100 → 200/400 GbE RDMA | $500-$2,000 per node |
| V1 | 100 GbE managed data switchInternal high-speed fabric | Compute to storage, and the future shared tier | Upgrade to 200/400G; split fabrics later | $1,500-$5,000 |
| V1 | Management switch / VLANSeparate out-of-band network | BMC/IPMI, UPS, switches, controllers, sensors | Add all future nodes and racks | $200-$750 |
| V1 | Kubernetes control nodeSmall dedicated x86 host | Keeps orchestration independent of the GPU workers | 1 → 3 HA controllers | $300-$800 |
| V1 | Local consoleSmall monitor, keyboard and shelf | Break-glass install, BIOS and network recovery | Rack KVM later | $150-$500 |
| V1 | Rack PDU + UPSMetered power and short ride-through | Clean power and graceful shutdown | Larger or redundant UPS as load grows | $1,500-$4,000 |
| V1 | Rails, optics, DAC, SAS, power, sparesIntegration hardware | A repeatable, serviceable installation | Standardize for future pods | $500-$1,500 |
| V1 | Tailscale access planeSecure overlay for clients and admin | Engineers reach services without the rack depending on any one machine | Add users and sites; substitute an approved on-prem solution for true air-gap | Service dependent |
| V2 | Lustre shared hot tierParallel POSIX filesystem over a fast RDMA fabric | A shared training and video working set across nodes | Add MDS/OSS/OST capacity | $5,000-$25,000+ |
| V2 | Compute Pod 02Second GPU worker node | Horizontal throughput and distributed training | Compute-03/N | Future node pricing |
| V2 | Kubernetes + GPU Operator + KubeRayCluster scheduling and distributed execution | One logical resource pool | Add workers horizontally | $0+ software |
| FUTURE | 200/400 GbE or InfiniBand compute fabricTightly coupled inter-node network | Distributed training and model parallelism | Future HGX, DGX or MGX nodes | $10Ks+ |
| FUTURE | Dense JBOD or storage shelf60-106-drive expansion | Hundreds of TB to PB-scale object storage | Add shelves independently of compute | $50K-$150K+ |
| OFF-GRID | Starlink + secure accessRemote WAN | Connectivity at a remote site | Backup WAN | Hardware + service |
| OFF-GRID | 48 V LiFePO4 bankLarge-format battery storage | Overnight and low-solar operation | Add modules from measured kWh/day | ~$20K-$35K+ at V1 scale |
| OFF-GRID | Solar arrayGround array; planning target ~25-30 kW for year-round V1 resilience in the Pacific Northwest | Generate rack energy and recharge batteries | Expand from measured load and site yield | ~$15K-$40K+ installed |
| OFF-GRID | Hybrid inverter + generatorAC conversion and charging, plus long-dark-period backup | Resilient off-grid operation | Parallel or larger equipment as compute grows | $8K-$25K+ |
Showing 34 of 34 components.
Software stack
All of it runs in the rack. Nothing in this list depends on a hosted service to function.
- Mixpeek local stack
- Ray / KubeRay
- Kubernetes
- NVIDIA GPU Operator
- vLLM
- SGLang
- TensorRT-LLM
- NeMo
- Megatron
- PyTorch
- Hugging Face
- Qdrant
- S3-compatible object service
- ZFS
- Lustre (V2)
- Prometheus / Grafana-class observability
The one rule behind the whole sheet
Every NOW-phase line is chosen so it is not thrown away at V1. The disks are SATA specifically so they move from a Thunderbolt enclosure into the 24-bay pod. The enclosure becomes migration or backup capacity rather than surplus. That constraint is why the build order starts with storage: it is the only layer where buying early does not mean buying twice.