Agent PD Simulator

Agent Session

Concurrency Total=10xConc Turns Z Pausems ISL OSL Range

Prefill (PD)

Nodes TP DP PF TPStok/s/rank ChunkSz PFMaxReq TxCost%

Decode

Nodes TP DP MaxReq/Rank TPOTms MinTPOTms

Prefetch / Offload

BW UMBPGiB/s BW PCIeGB/s T Fixedms Thresholdtok UMBP Hit% Offload% BW OffloadGB/s

Model Config

KV B/tokB FLOPs/tokF GPU TFTFLOPS

Simulation

Speed8x

Sim Time0:00

Sessions Pool0

In-Flight0

Sessions Done0

Turns Done0

Prefetch Chosen0

Recompute Chosen0

Prefetch Win Rate-

Avg PF Tokens-

PF Queue0

KV Xfer Queue0

Tool Wait0

PF Utilization0%

DC Slot Util0%

Avg TTFT-

Avg E2E/Session-

Throughput-

HBM Tokens0

DRAM Tokens0

Offloads0

Break-Even Analysis