There is not enough verified evidence to say that an Axera AX8860 beats NVIDIA’s DGX Spark by 25% on LLMs. The AX8860 product identity, its claimed 128 GB memory and 273 GB/s bandwidth, the 40 W power figure, and the four-module PCIe card have not been substantiated by the sources available. NVIDIA does publish those memory figures for DGX Spark—but memory bandwidth alone cannot establish which system generates LLM output faster.
What is verified—and what is not?
NVIDIA’s published DGX Spark specifications provide a concrete baseline. Its official product page lists 128 GB of LPDDR5x coherent unified system memory, 273 GB/s memory bandwidth, up to 1 PFLOP of FP4 tensor performance, a 140 W GB10 system-on-chip (SoC) thermal design power, and a 240 W power supply. The product page qualifies the theoretical FP4 figure with a sparsity footnote. NVIDIA’s DGX Spark specifications
As an Amazon Associate I earn from qualifying purchases.
NVIDIA’s DGX Spark hardware guide also lists up to 1,000 TOPS of inference performance and up to 1 PFLOP FP4 with sparsity. It specifies 128 GB LPDDR5x unified memory, 273 GB/s bandwidth, a 140 W GB10 SoC TDP, and 100 W available for other system components. These are distinct measures: peak compute, memory capacity, bandwidth, chip thermal design power, and power available to other components. None is a direct measurement of LLM generation speed.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →By contrast, no AX8860 datasheet, product page, or benchmark methodology was located among the available sources. That does not prove an AX8860 does not exist; it means the specific product and performance claims cannot currently be treated as verified. The headline’s 25% advantage, 128 GB LP5X memory, 273 GB/s bandwidth, 40 W consumption, and four-module card delivering more than 1,000 TOPS remain unsubstantiated.
#1 Best Overall
Why 273 GB/s and 128 GB do not prove an LLM speed advantage
The figures attributed to AX8860 match two published DGX Spark specifications, but matching them would not establish that another accelerator is faster. Memory capacity indicates how much data can be held; bandwidth describes how quickly data can move under specified conditions. Neither tells a reader how many tokens per second a system produces on a particular model.
Likewise, TOPS or FLOPS are peak-compute measures whose meaning depends on precision and calculation assumptions. NVIDIA’s FP4 peak figure explicitly uses sparsity. A theoretical TOPS figure is not interchangeable with measured tokens per second, time to first token, or end-to-end response time.
A meaningful “25% faster” comparison would need the same model and model version, precision or quantization, prompt and output lengths, batch size, decoding settings, software and runtime versions, and a clearly stated metric. A power comparison would also need to identify whether it measures a chip, accelerator card, or complete system. No such AX8860 comparison details are established.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallRank #2
- 900-5G172-2260-000
Is the AX8860 claim a mix-up with other accelerator products?
One potential source of confusion is the distinction between Axera Semiconductor and Axelera AI. They are separate companies, and neither company’s cited material verifies the AX8860 claims.
Axera Semiconductor: documented AX8850 card
Axera Semiconductor’s October 17, 2025 announcement describes an AX8850-based M.2 2242/2280 computing card with power consumption under 8 W. It says the card is intended to expand edge and terminal AI deployments, supports model architectures including DeepSeek, Qwen, and Llama, and can scale from an M.2 card to a PCIe card. That is an AX8850 product, not evidence for an AX8860, a 40 W figure, or the claimed four-module PCIe configuration. Axera Semiconductor’s AX8850 announcement
A separate M5Stack specification identifies the LLM-8850 Card’s SoC as Axera AX8850 and lists an M.2 M-key 2242 form factor with PCIe 2.0 x2. It is corroboration for an adjacent AX8850 card only. M5Stack LLM-8850 Card specification
Rank #3
- Extreme AI Performance: Powered by NVIDIA GB10 Grace Blackwell Superchip delivering 1 petaFLOP of AI performance and 128GB memory for 200B model fine-tuning.
- Developer-Optimized Platform: Designed for AI developers building secure, long-running agentic workflows, with compatibility across frameworks such as OpenClaw and NemoClaw, supporting private on-device inference, sandboxed execution, and governed data access.
- Scalable Architecture: Featuring NVIDIA NVLink-C2C for ultra-fast CPU-GPU memory communication and NVIDIA ConnectX-7 networking to support dual GX10 system stacking, unlocking superior scalability and performance.
- Advanced Thermal Design: Engineered cooling ensures sustained high performance and reliability in an ultra-small form factor.
- Full Stack AI Solution: The GB10 and NVIDIA AI software stack provide a full stack solution for AI development and deployment.
Axelera AI: different company and products
Axelera AI’s Edge products page describes Edge 130p and Edge 232p PCIe cards. It advertises up to 214 TOPS for Edge 130p and up to 629 TOPS for Edge 232p, and reports 83.4 tokens per second for Qwen 30B on Edge 232p. The page labels its comparison data as internal results against publicly available competitor data and says competitor data were current as of August 2026. Those are vendor-reported figures for Axelera AI products, not an AX8860 benchmark. Axelera AI Edge products
Free tools Windows power users keep installed
One-click scans. No signup required.
Axelera AI’s corporate site separately advertises more than 2,500 TOPS for a Server 150p with four Europa AIPUs and reports 83.4 tokens per second for Qwen3-30B-A3B against a stated competitor result of 61 tokens per second. These are distinct, company-reported claims about Axelera AI hardware; they do not verify a four-module AX8860 PCIe card or a DGX Spark comparison. Axelera AI
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What evidence would settle the comparison?
Before treating the AX8860 as a faster or more efficient alternative, ask for an identifiable product and reproducible results. A useful evidence package would include:
- An AX8860 SKU and manufacturer datasheet identifying the chip or card.
- Memory type, organization, capacity, and measured or specified bandwidth.
- Power measurements that state whether they cover the chip, card, or full system and describe the measurement conditions.
- The exact four-module PCIe configuration, host requirements, and availability.
- A complete LLM benchmark report identifying the model and version, precision or quantization, software and runtime versions, batch size, prompt and output lengths, decoding settings, and measured metric.
- The DGX Spark configuration used for the comparison and confirmation that both systems were tested under equivalent conditions.
Until those details are available, the defensible conclusion is limited: NVIDIA’s DGX Spark specifications are published, while the AX8860 specifications and its claimed 25% LLM advantage are not verified by the available cited sources.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
