Synoros Foundry R8-192
Maximum VRAM per dollar in an 8-GPU chassis
Budget-conscious buyer who needs raw VRAM capacity over speed / $4,000-$7,000
Total VRAM
192GB
GPUs
8x Tesla P40 24GB
Chassis
Used 8-GPU DDR4 Server (Supermicro/Dell class)
Form Factor
4U Rackmount
PCIe
PCIe 3.0
Starting at
$6,399
Model Compatibility
7B
13B
24B-35B
70B
122B
400B+
Sweet spot: 122B low-quant with 8-way parallelism
Stretch: 400B experiments with aggressive quantization + CPU offload
Notes
•P40 has no tensor cores and weak FP16. This is a capacity play, not a speed play
•8x250W passive cards = 2kW GPU draw alone. 208/240V mandatory
•PCIe 3.0 host limits load/offload speed across all 8 cards
•Older compute architecture. Inference speed per token is slower than Ampere/Volta equivalent VRAM
⚡240V required. 8x250W GPUs + dual Xeon/EPYC host = ~2.5-3kW steady state.
PCIe: Gen3-on-Gen3: capacity monster, not a speed build
Upgrade path: Synoros Foundry R4-192A or R4-320
Configure Your Build
Select upgrades to see live pricing.
Replace P40s with V100 32GB (better compute, same socket)included
Freight shipping, calculated at checkout
Powered by Stripe. Ships within the listed lead time.
Need something different? Contact sales for a custom build