Sangama → Results

Qwen3.5-397B across three US states

30 September 2026, round 13

Our earlier three-stage tests kept every machine in one datacenter. This one puts real distance between them, to measure two things we had built but not tested where they matter: connecting machines directly instead of through a relay, and QUIC as the transport instead of TCP.

Result:


Setup

ModelQwen3.5-397B-A17B, 4-bit, 3 stages of 20 layers
Stages96 GB GPUs in Maryland, Utah and Texas
Relay and clientOne machine in Colorado
Settings48 sessions per machine, shared steps, no guessing
CostAbout $4.80 an hour

Round trips between the machines: relay to Maryland 43 ms, relay to Utah and to Texas 26 ms each, Maryland to Utah 60 ms, Utah to Texas 34 ms.

Numbers

ModeOne requestNetwork per tripFirst token16 requests48 requests
TCP through the relay8.8-8.998.6 ms0.40-0.45 s113225
QUIC through the relay8.7-8.896-97 ms0.33-0.46 s110203
TCP direct10.3-10.482.6 ms0.55 s127247
QUIC direct10.4-10.581.4 ms0.28-0.31 s128-129243-245

Tokens per second. The 16 and 48 request columns count tokens over the whole run. GPU computing was 14 ms per trip in every mode. Single runs, except the two direct modes under load, which were run twice.

What the testing caught

Next

Full report and raw data: docs/test-results.


Sangama is open source under the MIT License. Contact: diljit@diljitpr.net · Discord