Model your own fleet.
Estimate what a shared KV cache tier returns on your GPUs, your workload and your latency target.
1 · WHAT'S YOUR WORKLOAD?
2 · YOUR CLUSTER
KV cache tier
Converged tier first; flash-tier nodes take the rest.
Average steady-state requests/sec. Sets rack count and batch size.
Latency objective. Tighter targets cap the batch, lowering GPU utilization and raising cost per million tokens.
Interactivity objective. Tighter targets cap decode batch size and may require more racks.
Typical 70–85% for long-context coding agents
MODELED
3-YEAR INFRASTRUCTURE SAVINGS, AFTER ACCELOVA COSTS
$0
on GPU CapEx, net of Accelova cost
———
Pick your cluster setup above to see the estimate.
How the model works
- Software on hardware you source. Accelova is modeled as a shared KV cache tier on the local SSD already inside your GPU servers, adding dedicated flash-tier nodes (commodity NVMe and RDMA) only for what those cannot hold or serve. vLLM / SGLang / TensorRT-LLM connect through a per-engine connector; no model or application changes.
- The baseline is node-local cache. The "without Accelova" column assumes each GPU reuses only the cache it built itself, at the baseline hit rate shown in section 2. If your workload already gets more local reuse, your savings will be smaller than shown. It also pays for its own SSD wear: every KV block its prefill servers cache is written to their local drives.
- The hit rate is solved, not set. The slider is what you ask for. The calculator uses the highest rate, never above it, that the KV capacity installed for the resulting fleet can hold, and the slider shows that rate when it is lower.
- Speedup is derived, not declared. TTFT and throughput come from a physical model of hit rate × model × rack: attention FLOPs, MFU curve, HBM bandwidth, PCIe ceiling.
- Net of everything Accelova costs. Savings are 3-year amortized GPU CapEx avoided, minus the Accelova subscription on every prefill and decode GPU, the cost of replacing the SSD its KV writes wear out (less what the baseline replaces on its own servers), and any flash-tier nodes' amortized cost, power and cooling.
YOUR FLEET
Walk through your scenario with us.
Copy the share link for your scenario and send it with the form. We will go through the assumptions against your workload and tell you plainly whether there is a case for your fleet.