Hardware Procurement — €10k Investment
Bestellte Systeme
| Gerät | Spezifikation | GPU | Kosten |
|---|---|---|---|
| ASUS Ascent GX10 (Storage + Orchestration) |
GB10 Prozessor (20c), 128GB RAM, 1TB NVMe | — | €3.415,83 |
| Dell Pro Max Tower T2 (Master Training) |
Core Ultra 7-265 (20c, 5.2GHz), 32GB RAM, 1TB NVMe | RTX 2000 16GB | €3.192,50 |
| Dell Pro Max Tower T2 (Secondary Training) |
Core Ultra 7-265 (20c, 5.2GHz), 32GB RAM, 1TB NVMe | RTX A1000 8GB | €2.199,17 |
| Total System | €8.807,50 (€10,5k inkl. 20% MwSt) |
||
GPU Compute Capacity
| Parameter | RTX 2000 16GB | RTX A1000 8GB | Combined |
|---|---|---|---|
| CUDA Cores | 2560 | 1792 | 4352 |
| Memory | 16GB GDDR6 | 8GB GDDR6 | 24GB total |
| Memory Bandwidth | 288 GB/s | 192 GB/s | 480 GB/s |
| FP32 Perf | ~13 TFLOPS | ~7 TFLOPS | ~20 TFLOPS |
| Max Power | 70W | 50W | 120W |
| Best For | ESM-2, cVAE Training | NN Training, BO | Sequential pipeline |
Key Difference vs. A100: A100 = 312 TFLOPS; RTX 2000+A1000 = 20 TFLOPS (~15× slower). Compensated by sequential, not parallel training.
Work Package Alignment (DRG + Aorta)
WP1: Target Definition (M1–M3)
Hardware Requirement: ASUS + 1× Dell (RTX 2000)
Task: Sequence Library Assembly, Initial Screening, ESM-2 Embedding Pilot
| DMS Setup | RAW FASTQ/BAM import → HDF5 conversion (CPU-bound on ASUS 128GB) |
| ESM-2 Pilot | 100–500 sequences on RTX 2000 (test embedding quality) |
| GPU Hours | ~20–40 GPU-h (RTX 2000) |
| Duration | ~1 week (sequential, includes validation) |
WP2: DMS Library (M2–M8)
Hardware Requirement: ASUS (Cache) + Dell1 (RTX 2000, 24/7)
Task: Full ESM-2 650M Embedding (500k sequences)
| Input | 500k AAV sequences (DRG + Aorta combined) |
| ESM-2 Model | 650M parameters on RTX 2000 (BatchSize=8, FP32) |
| Output | ~600GB HDF5 (500k × 1280-dim embeddings) |
| GPU Hours | ~1200 GPU-h (RTX 2000) = 72h continuous |
| Timeline | M2–M8 (6 months for library; embedding = first 3 weeks) |
| Storage | ASUS 128GB RAM (embedding cache) + 1TB NVMe (output) |
✓ Fits within M2–M8 window
WP3: Oracle v1.0 (M4–M10)
Hardware Requirement: Dell1 (RTX 2000) + Dell2 (RTX A1000)
Task: Multi-Task Neural Network + cVAE Architecture Design
| Model Training | 4-Task NN (Tropismus, Yield, Thermostability, Immunogenicity) on RTX 2000 |
| cVAE Training | Conditional VAE on RTX A1000 (lower memory footprint) |
| GPU Hours | ~150 GPU-h (RTX 2000) + ~100 GPU-h (RTX A1000) = 250 GPU-h total |
| Timeline | M4–M10 (6 months design + validation; training = 4–6 weeks) |
| Parallel Work | Both Dells used simultaneously: DDP over Ethernet (limited by network) |
✓ Fits within M4–M10 window (slack for hyperparameter tuning)
WP4: BO Runden (M8–M18)
Hardware Requirement: Both Dells (RTX 2000 + RTX A1000) + ASUS (Orchestration)
Task: 3 Rounds of Bayesian Optimization · Library Screening · Ranking
| Round 1 (M8–M11) | Embedding: 72h (RTX 2000) → NN: 18h (RTX A1000) → BO: 48h (CPU Gaussian Process) = 8–9 days |
| Round 2 (M11–M14) | Same as R1 (validated 2nd candidates) |
| Round 3 (M14–M18) | Same as R1 (refined final ranking) |
| Total GPU Hours | ~2400 GPU-h (72h × 3) on RTX 2000 + 54h × 3 on RTX A1000 = 378 GPU-h A1000 |
| Continuous Runtime | ~30 calendar days for 3 rounds (with sequential GPU scheduling) |
| Storage Pressure | Cumulative: 3 × 600GB embeddings + 3 × checkpoints = ~2TB (need external NAS) |
✓ CRITICAL: Requires external 4TB NAS by M8 (€1.5–2k additional)
WP5: Lead Validation (M14–M24)
Hardware Requirement: Both Dells (parallel wet lab + simulation)
Task: Experimental Validation, Refinement, Final Lead Selection
| Simulation | Molecular dynamics, structure prediction (CPU on ASUS or Dell CPUs) |
| ML Support | Transfer learning from Oracle (retrain final NN on 50–100 new candidates per month) |
| GPU Hours | ~300 GPU-h (spread over 10 months, ~30 GPU-h/month) |
| Storage | Experimental results, wet lab data, logs (~100GB/month) |
| Availability | Both Dells in intermittent use (not 24/7 like WP4) |
✓ Sufficient capacity; Dells can be repurposed for other tasks in parallel
M1–M24 Integration Timeline
Monthly Hardware Utilization
| Month | WP Phase | RTX 2000 (Dell1) | RTX A1000 (Dell2) | ASUS (Storage) | GPU-h / Month |
|---|---|---|---|---|---|
| M1–M3 | WP1: Target Def | 30% (Pilot ESM-2) | — | 100% (Data Prep) | ~30 |
| M2–M8 | WP2: DMS Library | 95% (500k embedding) | — | 100% (Cache + Staging) | ~900 |
| M4–M10 | WP3: Oracle v1.0 | 50% (NN training) | 40% (cVAE training) | 100% (Coord.) | ~350 |
| M8–M11 | WP4.1: BO Round 1 | 99% (ESM-2 + NN) | 80% (BO inference) | 100% (Orchest.) | ~450 |
| M11–M14 | WP4.2: BO Round 2 | 99% (Sequential) | 80% | 100% | ~450 |
| M14–M18 | WP4.3: BO Round 3 | 99% | 80% | 100% | ~450 |
| M14–M24 | WP5: Validation | 20% (Transfer Learn.) | 10% (Inference) | 100% (Results Storage) | ~50 |
| Total M1–M24 | — | ~2730 GPU-h | |||
Status: ✓ Feasible with continuous operation M2–M18 (8 months heavy, 16 months light/idle)
Critical Path
- M2–M8: ESM-2 embedding (bottleneck) must complete before M8 BO start
- M8–M18: 3 BO rounds sequential; no parallelism due to GPU memory constraints
- M18+: Dells available for other projects (Aorta continuation, new features)
Kostenanalyse & Procurement (€13.5k Active Workbase)
Hauptinvestition: Hardware (€10.5k brutto)
| System | Komponenten | Netto | Brutto (+20% MwSt) |
|---|---|---|---|
| ASUS Ascent GX10 | GB10 20c, 128GB RAM, 1TB NVMe | €3.416 | €4.100 |
| Dell Pro Max T2 (RTX 2000) | Core Ultra 7-265, 32GB RAM, RTX 2000 16GB | €3.193 | €3.832 |
| Dell Pro Max T2 (RTX A1000) | Core Ultra 7-265, 32GB RAM, RTX A1000 8GB | €2.199 | €2.639 |
| HARDWARE SUBTOTAL | €8.808 | €10.571 | |
Zusätzliche Investitionen (Workstation-Ready)
| Komponente | Spezifikation | Kosten (brutto) | Timeline | Kritikalität |
|---|---|---|---|---|
| 4TB NAS | Synology DS420+ oder QNAP TS-464 (RAID-1) | €1.800–2.000 | Vor M8 | 🔴 KRITISCH |
| 10GbE Netzwerk | 2× 10GbE Card + 8-port Switch + Kabel | €2.500–3.500 | Vor M2 | 🟡 HOCH |
| UPS + PDU | APC Smart-UPS 6kVA + Eaton PDU | €3.000–3.500 | Vor M1 | 🟡 HOCH |
| Software Licenses | Weights & Biases + Optuna Pro (optional) | €200–300/Monat | Ongoing | 🟠 MEDIUM |
| INFRASTRUKTUR SUBTOTAL | €7.500–9.500 | |||
📊 TOTAL "ACTIVE WORKBASE": €18.0–20.0k (Hardware €10.5k + Infrastruktur €7.5–9.5k)
RTX 2000+A1000 vs. Enterprise A100 — Kostenvergleich
| Metrik | RTX 2000+A1000 (€8.8k) | 4× A100 80GB (€320k) | Speedup Faktor |
|---|---|---|---|
| GPU FLOPS (FP32) | 20 TFLOPS | 312 TFLOPS | 15.6× |
| GPU Memory | 24GB total | 320GB total | 13.3× |
| Memory Bandwidth | 480 GB/s | 2.4 TB/s | 5× |
| ESM-2 Embedding Time (500k seq) | 72 hours | 4 hours | 18× |
| Per-Round Total (3 batches) | 8–11 days | 3–4 days | 2.5–2.75× |
| Cost per GPU-Hour | €0.36 | €3.20 | 8.9× cheaper |
| Power Consumption | 360W peak | 1.6kW | 4.4× less |
✓ Fazit: €10.5k Setup ist 2.5–3× langsamer, aber 36× billiger und fits within 24-month project window.
Procurement & Shopping Links
- Deutschland (Primär):
- Spezial-Hardware:
- Distributor (B2B):
Procurement Timeline
| M0 (Jetzt) | Planung, RFQ an 3 Anbieter, CIO-Genehmigung |
| M1 | Bestellung Dells + ASUS, 10GbE-Netzwerk, OS-Planung |
| M2 | Inbetriebnahme: BIOS, Ubuntu 24.04 LTS, NVIDIA Driver 555+ |
| M3 | Pilot-Testing: ESM-2 Embedding (100 seq), NN Training |
| M8 | ⚠️ NAS-Deployment Deadline (vor BO Round 1) |
Kritische Entscheidungen
| Entscheidung | Option A | Option B | Empfehlung |
|---|---|---|---|
| GPU Pair | RTX 2000+A1000 | 2× RTX 2000 | A (Spezialisierung) |
| NAS RAID | RAID-1 (2TB usable) | RAID-6 (3TB usable) | RAID-1 (Budget) |
| Netzwerk | 10GbE dedicated | Shared 1GbE | 10GbE (M8 deadline) |
| OS | Ubuntu 24.04 LTS | Windows 11 Pro | Ubuntu (CUDA stability) |
| Backup Strategy | iCloud + S3 Glacier | Local HDD | iCloud + NAS |
Storage Architecture (3TB → 4TB Required)
Tier 1: Local NVMe (Installed)
| ASUS 1TB | OS + Python env + cache layer | ~150GB headroom |
| Dell1 1TB (RTX 2000) | Embeddings cache + Models + Checkpoints | ~250GB headroom |
| Dell2 1TB (RTX A1000) | Training data staging + BO results | ~300GB headroom |
| Total Installed | 3TB | ✓ Sufficient for M1–M10 (single round) |
⚠️ Bottleneck for WP4 (3 rounds): 3× 600GB embeddings = 1.8TB. Need external storage by M8.
Tier 2: External NAS (Required by M8)
Recommendation: Synology DS420+ or QNAP TS-464 (4TB RAID-1)
Cost: €1.5–2k (budget separately from €10k hardware)
| Purpose | Warm archive (Embeddings Rounds 1–3, Model Checkpoints, Logs) |
| Connection | 10GbE (or Gigabit Ethernet, acceptable for sequential access) |
| Capacity | 4TB RAID-1 (2TB usable, 2TB redundancy) |
| Retention | Historical rounds (keep all for reproducibility) |
Tier 3: Cloud S3 (Optional)
AWS S3 Glacier or Azure Blob (cold archive after M24) for long-term regulatory compliance (~€0.03/GB/month)
ML Software Stack (Adjusted for RTX 2000/A1000)
Core Stack
| Component | Version | Notes for RTX Hardware |
|---|---|---|
Python | 3.12 | via uv venv (locked deps) |
PyTorch | 2.4.0 + cu125 | Torch.compile() disabled (overhead). FP32 default (RTX A1000 stability). |
transformers | 4.46.0 | ESM-2 650M (not 3B). gradient_checkpointing=True for 16GB VRAM |
biotite | 0.45.0 | Sequence I/O, validation |
scikit-optimize | 0.9.0 | BO, Gaussian Processes (CPU-bound OK) |
botorch | 1.14.0 | Multi-objective BO (simplified GP kernels for CPU-bound execution) |
optuna | 3.6.0 | Hyperparameter search (not critical; manual tuning OK on this budget) |
pandas / numpy | 2.2.0 / 2.0.0 | Data manipulation (not GPU-dependent) |
Development & Monitoring
- Jupyter Lab 4.2.0: Prototyping on ASUS or Dell (remote via SSH)
- Weights & Biases: Experiment tracking (cloud logging, budget-friendly)
- pytest 8.0.0: Unit tests
- black / isort / pylint: Code quality
OS & Drivers
Recommended: Ubuntu 24.04 LTS (Dual-boot from Windows 11 Pro)
| NVIDIA Driver | Latest (555+) for RTX 2000 / A1000 support |
| CUDA Toolkit | 12.5 (matches PyTorch cu125) |
| cuDNN | 9.0 (RTX 2000: SM 7.0; RTX A1000: SM 8.0 compatible) |