Performance
Proof over promises.
Independently measured performance, published methodology,and results stacked against hyperscalers.
Storage
#1 in Storage Performance
Our storage infrastructure ranked #1 across leading industry standard benchmarks
Solidigm PCIe 5.0 D7-PS1010 NVMe SSDs
Fastest single node / client for ML Perf 2.0 Storage Training Workloads
#1 in Clustermax 2.0 Storage, Torch Import benchmark
BlueField 3 DPU accelerated data path
ClusterMax
Semianalysis ClusterMax™ to the max
Every dimension ClusterMax v2 evaluates — Security, Lifecycle, Orchestration, Storage, Networking, Reliability, Monitoring, Pricing, Partnerships, Availability — is baked into how we designed the platform.
Evaluation Criteria
Our results
ClusterMax
The standard
Security
Confidential
Enterprise-grade
Lifecycle
Self-serve
Automated
Orchestration
SLURM + K8s
Multi-stack
Storage
High-IOPS NVMe
Performance tier
Networking
800G RoCE
Non-blocking
Reliability
Auto-healing
Active remediation
Monitoring
Integrated
End-to-end
Pricing
Per-second
Transparent
Partnerships
NVIDIA Preferred
Tier-1 vendors
Availability
99.9%
Production SLA
FIRST TRAINING STEP
First training step, faster.
From instance launch to import torch with CUDA verified.
Evaluation Criteria
Our results
Typical Provider
The usual experience
Cold start VMs
< 60 SEC
5-10 MINUTES
GPU Drivers, CUDA, PyTorch, DOCA
Included
MANUAL SETUP
Local NVMe Throughput
100-160GB/S
7GB/s
Torch Import
<1 second
tens of seconds
PRICING COMPARISON
Up to 70% lower cost per GPU hour.
H100 SXM on-demand pricing, per GPU hour.
Pricing
On-demand pricing
Hyperscalers
AWS, GCP, Azure
H100 SXM Per GPU/hr
~$2.99
~$7–13
Billing Model
PER-SECOND
PER-HOUR
Egress Fees
NONE
VARIABLE
Lock-in Required
NONE
1–3 YEARS
NETWORK FABRIC
Network Performance.
800G Ethernet fabric with RoCE, measured across real multi-node training jobs.
Metric
Our fabric
Hyperscalers
AWS, GCP, Azure
Inter-Node Bandwidth
400 GB/S
VARIES
Backend Fabric
800 GBPS
VARIES
RDMA Support
RoCE V2
CONFIG DEPENDENT
Packet Loss
< 0.1%
VARIES
Spin up a cluster, run the tests yourself, and compare against your current infrastructure.
Get started in minutes.
Deploy your first cluster today with transparent pricing and zero lock-in commitments.