Builds / 2x Nvidia GB10 (unified, DGX Spark)

2x DGX Spark (AI TOP ATOM + ThinkStation PGX): GLM-5.3 Flash EXL3 4 bpw on Mia's 1104 TensorFold recipe, independently measured at 40.6 tok/s (thinking off), prefill 1,804 tok/s at 32K

Last updated: 12 Nov 2026

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

GPU(s): 2x Nvidia GB10 (unified, DGX Spark)
VRAM / unified memory (GB): 256
CPU: Nvidia GB10 Grace (Arm) per node
Amazon · eBay (used) · Newegg
Motherboard: Gigabyte AI TOP ATOM + Lenovo ThinkStation PGX
Amazon · eBay (used) · Newegg
RAM: 128 GB LPDDR5X unified per box (256 GB total)
PSU: not stated
Other: 2 GB10 boxes (1x Gigabyte AI TOP ATOM + 1x Lenovo ThinkStation PGX) linked by CX7 RDMA (~109 Gb/s), driver 580.178.04, kernel 7.0.0-1019-nvidia, GPU clock capped at 2,200 MHz; Mia's AI Lab GLM-5.3-Flash-DGX-Sparks-1104 recipe (commit 3f44ec2, 4 Nov 2026), TensorFold native Zig engine f350cd3e + 105 recipe patches, TP=2, DFlash2 drafts (4-bit, incoai, CC BY-NC-ND), FP8 KV, 8 requests at once, 1,048,576 context
Reported total cost: not stated
Power: not stated

Measured speed (generation tokens/sec)

Model	Quant	Engine	tok/s	Prompt processing (tok/s)	Details
GLM-5.3 Flash	Mia EXL3 4 bpw (rev 078455ff) + DFlash2 drafts	TensorFold Zig, 2 nodes TP=2, 1 request	40.6	1,804 @ 32,768-token prompt est. TTFT @8k ≈ 4.5 s (calculated from measured prefill) measured TTFT: 4.9 s @ 8,192-token prompt	⚡ spec-decode (DFlash2)context: 1,048,576 ctx, 512 generated tokens, thinking off📏 Mean of code (42.4) and prose (38.7) medians, 2 runs each after 1 warm-up, temperature 1.0, seed 42, no prompt cache; draft acceptance 58%; 36.7-42.5 tok/s with 16K-128K prompts; measured 2026-11-04

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.

Independent re-run of Mia's AI Lab 1104 recipe by niklaslenz.com (Results v2), measured 4-5 Nov 2026, raw files published. Its own prompts, temperature 1.0 and thinking off, so not comparable with Mia's README figures (47.8 tok/s at temperature 0.7, thinking on). Mia's own sparkDash tool, run unchanged by the same tester (temperature 0, structured prompt): 94.3 tok/s decode for one request, prefill 1,822 at 32K. Several requests at once (total): 1 = 38.6, 2 = 51.9, 4 = 65.3, 8 = 80.0 tok/s. Peak memory rise 102.0 GiB per box.

Source: https://niklaslenz.com/v2/glm53-flash-2sparks-mia-tensorfold-1104/

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/2x-dgx-spark-glm-53-flash-mia-1104-recipe-independent-niklaslenz.svg)](https://getrigbase.com/builds/2x-dgx-spark-glm-53-flash-mia-1104-recipe-independent-niklaslenz/)

HTML

<a href="https://getrigbase.com/builds/2x-dgx-spark-glm-53-flash-mia-1104-recipe-independent-niklaslenz/" rel="noopener"><img src="https://getrigbase.com/badge/2x-dgx-spark-glm-53-flash-mia-1104-recipe-independent-niklaslenz.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 2x 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).