AI Hardware Catalog

Inference-ready GPU systems — sourced, integrated, cabled, imaged, and burn-tested. Every system ships rack-ready to your colocation facility.

24–48 hr

Quote turnaround

3-year

On-site warranty (NBD)

No minimum

Single-unit orders welcome

In Stock

Rack Bundles

Turnkey solutions — every bundle is sourced, integrated, burned-in, and shipped rack-ready. Pick your scale tier and request a quote.

Startup

Rivram Seed

Your first production GPU node

01

Use case: LLM inference up to ~13B params; API gateway node

Server Supermicro SYS-111E-FWTR (1U)
GPU 1× NVIDIA L40S 48 GB
CPU Intel Xeon W-2465 (16C / 32T)
RAM 128 GB DDR5 ECC
Storage 2× 3.84 TB NVMe U.2 (RAID-1)
Network 2× 25 GbE SFP28
Power 800 W 80+ Platinum
OS Ubuntu 22.04 LTS + CUDA 12.4 + vLLM

Chassis also supports (per Supermicro):

L4 · L40 · L40S · RTX 6000 Ada · RTX PRO 6000 Blackwell

Get a Quote for Rivram Seed →
Growth

Rivram Ranger

Dual-GPU workhorse for growing teams

02

Use case: Multi-model serving, RAG pipelines, image/video AI

Server Supermicro SYS-221GE-NR (2U)
GPU 2× NVIDIA L40S 48 GB
CPU Dual Intel Xeon Gold 6454S (32C × 2)
RAM 512 GB DDR5 ECC
Storage 4× 3.84 TB NVMe U.2 (RAID-10)
Network 2× 100 GbE QSFP28
Power 2× 2000 W 80+ Titanium (redundant)
OS Ubuntu 22.04 LTS + CUDA 12.4 + vLLM + TGI

Chassis also supports (per Supermicro):

L40S · L40 · A100 PCIe · RTX PRO 6000 Blackwell

Get a Quote for Rivram Ranger →
Production

Rivram Trail Boss

4-GPU cluster for serious production loads

03

Use case: 70B-class model inference, multi-tenant API serving

Server Supermicro SYS-421GE-TNRT (4U)
GPU 4× NVIDIA L40S 48 GB
CPU Dual Intel Xeon Platinum 8468 (48C × 2)
RAM 1 TB DDR5 ECC
Storage 8× 7.68 TB NVMe U.2 (RAID-6)
Network 2× 100 GbE QSFP28 + 1× OOB IPMI
Power 3× 2000 W 80+ Titanium (N+1 redundant)
OS Ubuntu 22.04 LTS + CUDA 12.4 + vLLM + Ray Serve

Chassis also supports (per Supermicro):

A10 · L4 · L40 · L40S · RTX 5000 Ada · RTX 6000 Ada · RTX PRO 6000 Blackwell

Get a Quote for Rivram Trail Boss →

Need 8× SXM, H100 / H200 / B200, or a custom config? See the Rivram Titan custom build below.

Rivram Titan — Custom HGX Build

8× SXM5 systems are configured per workload and aren't sold off-the-shelf — every Rivram Titan build is BOM-reviewed, sourced through our Supermicro channel, integrated, burn-tested, and shipped to your colo.

Custom · Flagship 8-GPU HGX SXM5 — flagship build, configured to spec

Use case: Frontier model inference, MoE serving, high-throughput batch, multi-tenant training

Server Supermicro SYS-821GE-TNHR (8U HGX) · or AS-8125GS-TNHR (AMD EPYC)
GPU 8× NVIDIA H100 SXM5 80 GB · or 8× H200 SXM5 141 GB · or 8× B200 SXM5 (via SYS-A21GE-NBRT 10U air / SYS-421GE-NBRT-LCC 4U liquid)
Fabric NVLink 4 + NVSwitch — 900 GB/s GPU↔GPU
CPU Dual Intel Xeon Platinum 8568Y+ (up to 64C × 2) or Dual AMD EPYC 9004
RAM 2–4 TB DDR5-5600 ECC (up to 8 TB / 32 DIMM)
Storage 16× 7.68 TB NVMe U.2 + 2× M.2 boot
Network 8× NVIDIA ConnectX-7 NDR 400 Gb/s InfiniBand (1:1 GPU:NIC, OSFP) + 2× ConnectX-7 / BlueField-3 north-south
Power 8× 3000 W 80+ Titanium (4+4 redundant)
OS Ubuntu 22.04 LTS + CUDA 12.4 + vLLM + Triton + Ray Serve

GPU options on this chassis (per Supermicro):

8× H100 SXM5 · 8× H200 SXM5 · 8× B200 SXM5 (Blackwell path via separate chassis)

Lead Time

8–12 weeks

Pricing (indicative)

From ~$280K (H100) · ~$400K (H200) · ~$500K (B200)

Wall Power

~10–12 kW per node

Request a Rivram Titan Consultation →

We'll review workload, fabric, and power requirements before quoting.

GPU Cards

All GPUs are available standalone or pre-integrated into Supermicro server platforms. Contact us for current availability and lead times.

🚀

Startup / SMB

Accessible Performance for Dev & Inference

GPU Model VRAM FP32 (TFLOPS) INT8 (TOPS) TDP
NVIDIA RTX 4090 24 GB GDDR6X 82.6 450 W
NVIDIA RTX 6000 Ada 48 GB GDDR6 91.1 300 W
NVIDIA L4 24 GB GDDR6 30.3 242 72 W
NVIDIA A10G 24 GB GDDR6 31.2 250 150 W

Mid-Scale / Growth

Production Inference at Reasonable Cost

GPU Model VRAM FP32 (TFLOPS) INT8 (TOPS) TDP
NVIDIA L40S 48 GB GDDR6 91.6 733 350 W
NVIDIA RTX PRO 6000 Blackwell 96 GB GDDR7 125 4,000 600 W
NVIDIA RTX 4000 Ada 20 GB GDDR6 26.7 212 130 W
🏢

Enterprise — Custom Builds Only

Maximum Throughput & Scale — sold as part of Rivram Titan custom builds

GPU Model VRAM FP32 (TFLOPS) INT8 (TOPS) TDP
NVIDIA A100 PCIe 80 GB 80 GB HBM2e 77.6 624 400 W
NVIDIA H100 PCIe 80 GB 80 GB HBM3 51.2* 3,958 350 W
NVIDIA H100 SXM5 80 GB 80 GB HBM3 66.9* 3,958 700 W
NVIDIA H200 SXM5 141 GB 141 GB HBM3e 66.9* 3,958 700 W
NVIDIA B200 SXM5 192 GB HBM3e 80* 9,000 1000 W

* H100 and H200 FP32 dense = 66.9 TFLOPS; FP8 sparse = 3,958 TOPS. SXM variants require NVLink-compatible HGX platforms.

Server Platforms

All systems ship with Ubuntu 22.04 LTS, NVIDIA drivers, and CUDA pre-installed. Custom OS images available on request.

🚀

Entry / Startup

1–2 GPU Systems

Model Form GPU Slots
Supermicro SYS-111E-FWTR 1U 1× PCIe Gen5
Supermicro SYS-221GE-NR 2U MGX up to 4× PCIe Gen5

Mid-Scale / Growth

4–8 GPU Systems

Model Form GPU Slots
Supermicro SYS-421GE-TNRT 4U up to 10× PCIe Gen5
Supermicro AS-4125GS-TNRT 4U 8× PCIe Gen5
Supermicro SYS-521GE-TNRT 5U 10× PCIe Gen5
🏢

Enterprise — Custom Builds

8× SXM HGX Systems · Build-to-Order

Model Form GPU Slots
Supermicro SYS-821GE-TNHR 8U HGX 8× SXM5
Supermicro AS-8125GS-TNHR 8U HGX 8× SXM5
Supermicro SYS-A21GE-NBRT 10U HGX 8× SXM5 (Blackwell)

How It Works

Request a quote, we confirm lead times and availability, you approve the BOM, and we handle the rest — sourcing, integration, burn-in, and rack delivery.

Quote Turnaround

24–48 hours

Same-day for bundles listed above

Lead Time — Startup / Growth

2–4 weeks

Standard PCIe configurations

Lead Time — Enterprise SXM

6–10 weeks

HGX / NVLink platforms

Minimum Order

Single unit

No volume minimum required

Warranty

3-year on-site (NBD)

Extended 5-year options available

Financing

12–36 month terms

Available via partner lenders

Request a Quote

Tell us which system you're interested in — or describe your workload and we'll recommend the right configuration.