Competitor hardware and benchmark signals

Nvidia is using MiniMax M2.7 to show how open artificial intelligence models can run on its graphics chips, even as rival chip and device benchmarks send a mixed signal. (developer.nvidia.com) MiniMax M2.7 is a mixture-of-experts language model with 230 billion total parameters, 10 billion active per token, 256 experts, and a 200,000-token context window, according to Nvidia’s April 2026 developer post. Nvidia said the model can be deployed through NemoClaw, an open-source stack that installs the OpenShell runtime for “always-on” assistants and other autonomous agents. (developer.nvidia.com) MiniMax’s public repository describes M2.7 as an open model built for “complex agent harnesses,” “Agent Teams,” and dynamic tool search rather than a basic chatbot. That framing puts the model in the part of the market where companies are trying to turn large language models into software workers that can call tools, keep state, and complete longer tasks. (github.com) A benchmark is a standardized test, like a lap time for chips, and the latest numbers point in different directions depending on the hardware category. Nvidia’s post focused on running open models on graphics processing units, while separate April 2026 reports centered on Intel’s workstation graphics card performance and Qualcomm’s next mobile system-on-chip scores. (developer.nvidia.com, videocardz.com, browser.geekbench.com) Intel’s Arc Pro B70 is a workstation card, not a gaming flagship, and Intel’s own materials pitch it for artificial intelligence inference, rendering, and certified professional software. Intel’s March 2026 quick reference guide and datasheet both describe the Arc Pro B-series as workstation-grade hardware built for scalable multi-graphics-processing-unit setups and creative workloads. (intel.com, intel.com) Even so, early game tests became a proxy for how Intel’s silicon stacks up against Nvidia in consumer-style loads. VideoCardz, citing Level1Techs results published in April 2026, said the Arc Pro B70 was about 45% faster than the lower-tier Arc Pro B60 across six gaming results, but those reports did not show it overtaking comparable Nvidia gaming cards. (videocardz.com) The Intel story is more complicated in artificial intelligence and workstation use. VideoCardz reported on April 9 that a four-card “Battlematrix” Arc Pro B70 setup could draw up to 720 watts and that several benchmark suites did not scale cleanly across multiple cards, with useful multi-card scaling mainly shown in LM Studio under Ubuntu. (videocardz.com) On the mobile side, Geekbench entries are user-submitted test results, not launch announcements, but they are often used to spot unreleased chips before products ship. Geekbench’s own benchmark pages say higher scores are better and that its charts are built from user-submitted Geekbench 6 results. (browser.geekbench.com, browser.geekbench.com) A recent Geekbench AI listing for the Honor Magic8 Pro identified the device’s graphics processor as “Qualcomm Snapdragon 8 Elite Gen 5,” with a quantized score of 4,011 and an upload date of April 10, 2026. A separate Geekbench 6 result for a Snapdragon 8 Elite device from December 2024 showed a 3,142 single-core score and 9,997 multi-core score, offering a public reference point for the prior generation rather than the leaked Oppo tablet itself. (browser.geekbench.com, browser.geekbench.com) Oppo’s tablet line adds another wrinkle because the shipping Oppo Pad 5 is already in the market with MediaTek’s Dimensity 9400+ chip, not Qualcomm silicon. GSMArena reported in October 2025 that Oppo announced the Pad 5 with the Dimensity 9400+, after earlier leak coverage had tied an Oppo Pad 5-family device to higher-end display specs and a possible flagship launch window. (gsmarena.com, gsmarena.com) Put together, the April 2026 signals show three separate races moving at once: Nvidia pushing open-model software on its graphics processors, Intel trying to prove workstation graphics can stretch into broader use, and Android device makers testing the next round of flagship mobile silicon in public databases. (developer.nvidia.com, intel.com, browser.geekbench.com)