| RTX Pro 6000 | RTX 5090 Founder | RTX 4090 Founder | A100 80 GB PCIe | H100 PCIe | H100 SXM | RTX 6000 Ada | |
|---|---|---|---|---|---|---|---|
| CUDA Cores | 24,064 | 21,760 | 16,384 | 6,912 | 8,192 | 18,176 | |
| Tensor Cores | 680 | 512 | 432 | 512 | 568 | ||
| AI TOPS | 4000 | 3352 | 1321 | ||||
| FP16 Tensor operations | 330.42 TFLOPS Pure FP16 | 624 TFLOPS Sparse | 1513 TFLOPS Sparse | 1979 TFLOPS Sparse | 362 TFLOPS Sparse | ||
| Memory | 96 GB GDDR7 | 32 GB GDDR7 | 24 GB GDDR6X | 80 GB HBM2e | 80 GB HBM2e | 80 GB HBM3 | 48 GB GDDR6 ECC |
| TGP | 575W | 450W | 300W | 350W | 300W | ||
| Memory bandwidth | 1792 GB/s | 1792 GB/s | 1008 GB/s | 1935 GB/s | 2000 GB/s | 3350 GB/s | 960 GB/s |
| Price | $6,320 | $1,999 | $1,599 |
| Type | Item | Price |
|---|---|---|
| CPU | AMD Ryzen 9 9950X 4.3 GHz 16-Core Processor | $529.99 @ Amazon |
| CPU Cooler | Thermalright Phantom Spirit 120 SE 66.17 CFM CPU Cooler | $35.90 @ Amazon |
| Motherboard | MSI MPG X670E CARBON WIFI ATX AM5 Motherboard | $379.99 @ Amazon |
| Memory | Corsair Vengeance RGB 64 GB (2 x 32 GB) DDR5-6000 CL30 Memory | $244.99 @ Amazon |
| Storage | Samsung 990 Pro 4 TB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive | $269.99 @ Abt |
| Video Card | RTX Pro 6000 Blackwell *2 | |
| Case | Sliger CX4170a | |
| Power Supply | SeaSonic VERTEX PX-1200 1200 W 80+ Platinum Certified Fully Modular ATX Power Supply | $319.99 @ Amazon |
| Case Fan | Corsair RS120 72.8 CFM 120 mm Fans 3-Pack | $44.99 @ Amazon |
This build is a dual RTX Pro 6000 setup in a 4U rack-mount case. The two GPUs share 16 PCIe 5.0 lanes, which is fine as long as you don’t try to run a single model across both cards (i.e., a large model won’t fit on one GPU). Data-parallel workloads are not affected.
The GPUs are the Max-Q version with a 300W TDP, which allows them to fit on an ATX motherboard in this 4U chassis. The downside is a 10–20% performance reduction compared with full workstation cards.
This configuration represents a balanced, budget-friendly setup. Any upgrades — such as adding more GPUs or switching to full workstation cards — would quickly increase costs, making the system significantly more expensive than this optimized baseline.