Builds / 4x Nvidia GB10 (unified)

4x DGX Spark (GB10, 2,200 MHz cap): GLM-5.3 Flash NVFP4 at 104.1 tok/s code / 3,159 prefill (vLLM TP4, DFlash2)

Last updated: 8 Nov 2026

Source: measuredlocally.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source

GPU(s): 4x Nvidia GB10 (unified)
VRAM / unified memory (GB): 512
CPU: NVIDIA GB10 (20-core Arm) ×4
Amazon · eBay (used) · Newegg
Motherboard: 4× NVIDIA DGX Spark (ConnectX-7 200Gb/s)
Amazon · eBay (used) · Newegg
RAM: 128 GB unified per box
PSU: not stated
Other: Four DGX Sparks joined through a MikroTik CRS812 switch; NVIDIA DGX OS; vLLM 0.1.dev20051 tensor parallel 4; graphics chips power-capped at 2,200 MHz; timed from another machine on the same network
Reported total cost: not stated
Power: each GB10 capped at 2,200 MHz

Measured speed (generation tokens/sec)

Model	Quant	Engine	tok/s	Prompt processing (tok/s)	Details
GLM-5.3 Flash	NVIDIA NVFP4 (4-bit, 204.5 GB on disk), dense layers re-encoded at 8 bits (recipe: lossless8)	vLLM 0.1.dev20051 TP4, DFlash2 (FP8 draft), 1,048,576-token window	104.1	3,159 @ 16,000-token prompt est. TTFT @8k ≈ 2.6 s (calculated from measured prefill)	⚡ spec-decode (DFlash2)context: code prompt, thinking off; 1,048,576-token window configured📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); graphics chips capped at 2,200 MHz (about 2,400 uncapped) to cut power use — author's August check: writing within 1%, reading 2–4% slower; middle of three code runs 100.0–104.2 tok/s; prose 62.7 (62.6–63.4); first word 0.15 s; 4 users at once: 112.3 tok/s in total

Prompt processing = prefill speed (how fast the prompt is read) as reported by the source, with the prompt length it was measured at; blank = the source does not report it. est. TTFT @8k is calculated from the measured prefill speed (8,192 ÷ prefill tok/s), shown only when prefill was measured on a prompt of 512+ tokens. It is not a measurement and ignores load time and caching. measured TTFT is the source's own time-to-first-token. Why prefill matters for coding assistants.

measuredlocally.com hardware leaderboard, GLM-5.3 Flash row for four DGX Sparks, tested 11 Oct 2026. Power-capped; a small draft model (DFlash2 converted to FP8) guesses several tokens ahead. Same author's method as the one-Spark and two-Spark Qwen3.8 27B rows added 2026-11-07.

Source: https://measuredlocally.com/hardware-leaderboard.html

Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.

Embed this result

Copy into a README or blog. Links back to this build. See badge docs (CC BY 4.0 attribution). The SVG does not invent speeds — numbers stay on the build page.

Markdown

[![Measured on RigBase](https://getrigbase.com/badge/4x-dgx-spark-glm-53-flash-vllm-dflash2-measuredlocally.svg)](https://getrigbase.com/builds/4x-dgx-spark-glm-53-flash-vllm-dflash2-measuredlocally/)

HTML

<a href="https://getrigbase.com/builds/4x-dgx-spark-glm-53-flash-vllm-dflash2-measuredlocally/" rel="noopener"><img src="https://getrigbase.com/badge/4x-dgx-spark-glm-53-flash-vllm-dflash2-measuredlocally.svg" alt="Measured on RigBase" height="20"/></a>

Preview: · How to use

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Quick answers: Can 4x DGX Spark (GB10) run GLM-5.3 Flash?

Speculative-decoding rows on this page are not like-for-like with plain decode. See how much MTP and DFlash really speed things up (measured pairs).