Research by Anamika Rawat and Tulsi Patel, full-time co-op student researchers at Conestoga College, supervised by Dushyant Puri. This work is part of SimpleDirect®’s collaboration with Conestoga’s Centre for Commercialization; the benchmarking comes through the SMART Centre.
Before recommending local AI, I want to know what someone can get done on the hardware they have. How long will they wait? Will the answer be complete? Does it actually help with the task?
Anamika and Tulsi have been testing those questions with Vinci Piccolo 1.0 on two Macs, compact computers and an Android phone. Their early results include faster generation, responses marked truncated and a calculation that never arrived. Each tells us something useful about the next test.
They performed the benchmarking. I’m writing about the results and how I think we should use them. SimpleDirect® hasn’t independently rerun the hardware tests.
What the first results tell us
In the Mac comparison, the Mac mini generated text faster on all 24 matched prompts. Some responses were also marked truncated. Before using that speed difference to justify a purchase, I’d want to inspect the full answers and check whether the tasks were completed. Matched prompts help us compare these configurations; they don’t establish repeat-run reliability.
The Jetson and Raspberry Pi tests covered four prompts, run once on each device. Jetson generated text faster, using CUDA GPU acceleration while Pi used its CPU. That software setup matters to the comparison.
One prompt asked for the price of a $250 item after a 20 percent discount and 13 percent tax. The answer is $226. Neither setup supplied it: Jetson returned something resembling a tool call, and Pi declined the calculation. Both requests were marked completed. I’d want that failure investigated before treating either setup as useful for this task.
On the 6 GB Samsung S20 FE, the app reported generation durations of 201.8 and 96.7 seconds in two sequential observations. A memory warning appeared during setup. The records don’t establish answer quality or a minimum RAM requirement, and the two observations cannot isolate the effect of thinking mode. [5]
Why work with Conestoga
SimpleDirect® works with Conestoga’s Centre for Commercialization, which supports entrepreneurs and smaller businesses with intellectual property and commercialization. The SMART Centre provides applied research, technical services and training. They are distinct parts of the college that also collaborate on IP services. [1, 2]
Dushyant is a senior researcher at SMART, with work spanning software development, IoT, data processing and business integration. That experience is relevant when a result depends on the model, its software and its hardware working together. [3]
Our collaboration continues through at least the end of November 2026. It gives the researchers a practical company problem to investigate. Conestoga’s wider research program offers students paid co-op and part-time opportunities. [4]
Follow-on conversation records shared on October 8 ask another question: can Piccolo preserve the details of a plan as the requests change? This is part of the same continuing collaboration, with a separate test scope from the hardware observations above. We examine the records and their limits in Piccolo’s multi-turn planning tests.
Use these findings to choose your next test
Piccolo is our small model for conversation and drafting, released on June 29, 2026 and available under Apache-2.0.
If you’re considering it, start with a task whose answer you can check. Write down an acceptable wait, save the complete output and record what needed correcting. Repeat the task before deciding to depend on the setup or spend money on another device.
That is the standard I want these results to help us work toward: enough evidence for a person to make a useful decision. Our practical resources can help you plan that check.
Sources
- Conestoga College — Centre for Commercialization
- Conestoga College — SMART Centre
- Conestoga College — Dushyant Puri, senior researcher
- Conestoga College — Students in research
SimpleDirect Vinci-Piccolo Benchmarking.pptx, slides 2–37. Phone, edge-board and Mac observations credited to Conestoga SMART Centre. The accompanying Mac CSV exports and prompt bank underpin the Mac article, which provides the downloads. The supplied deck is cited as a source record and is not offered as a download.
Research by Anamika Rawat and Tulsi Patel, supervised by Dushyant Puri, through Conestoga College’s SMART Centre. SimpleDirect®’s collaboration is with Conestoga’s Centre for Commercialization. Article and interpretation by George Pu.
George Pu est le fondateur et chef de la direction de SimpleDirect®, l’entreprise canadienne indépendante d’IA qui bâtit Vinci pour une vie au-delà de l’écran.