Is the DGX Spark Worth It? $4,699 Price vs Real Speed

Bottom Line

• Price: $4,699 NVIDIA list price. The cheapest Amazon US unit was $4,999.99 in September 2026, and the NVIDIA store was out of stock. NVIDIA added an official $700 in February, blaming memory supply. Correction (2026-09-13): the ASUS Ascent GX10 no longer undercuts it. OEM GB10 boxes rose to $5,400-8,200 by September 2026 (every GB10 price).
• Speed: stack-dependent to a degree nobody expects: gpt-oss 120B decode ran ~11.7 tok/s on Ollama, ~38.5 on tuned llama.cpp, ~50 on SGLang — same box, ~4x spread.
• The real product: CUDA parity with your deployment target, on a desk. Not tokens per second.
• Verdict: worth it for NVIDIA-stack developers. Not worth it as a pure inference box — a ~$3,500 Ryzen AI Max+ 395 machine does that job for about $1,200 less.

DGX Spark price: what $4,699 actually buys

GB10 Grace Blackwell, 128GB unified memory, ConnectX networking for pairing two units, and the full NVIDIA software stack in a quiet desktop box. The price history matters: it launched at $3,999, and in February 2026 NVIDIA raised it to $4,699 in an official notice citing memory supply constraints — the same shortage that reset RAM and GPU prices everywhere.

DGX Spark price history

The list price did not change after February. The street price did. In September 2026 you pay about $300 over list, or 25% over the launch price. For the OEM boxes on the same chip, see every GB10 box price.

The spec that defines it is 273 GB/s of memory bandwidth. An RTX 5090 moves ~1,792 GB/s — about 6.5x more. Token generation is bandwidth-bound, so no amount of Blackwell compute makes single-stream chat fast on this box. That is not a flaw discovered by reviewers; it is the design tradeoff: capacity and ecosystem over bandwidth.

The benchmark chaos, explained

The Spark’s launch reviews disagreed with each other by 4x, and both sides were measuring honestly:

Two lessons. First, on this machine the software stack is worth more than 3x the hardware delta between it and its rivals — if you buy one, do not run Ollama defaults and conclude you were robbed. Second, all three numbers obey the same ceiling: gpt-oss 120B is MoE with a small active set, which is why it moves at all. A dense 70B reading ~40GB per token caps in single digits — if dense 70B is your goal, that money buys a much better rig.

The three buyers it is right for

1. The deploy-to-NVIDIA developer. You prototype locally, ship to A100/H100/Blackwell in production, and need identical CUDA/TensorRT behavior. The Spark is the only 128GB desk box that gives you that. This is its honest, load-bearing use case.
2. The DGX/NIM ecosystem team. Desk-side node, same containers, same tooling as the fleet.
3. The two-Spark buyer. NVIDIA’s 200B-class and 1M-context configurations pair two units over ConnectX. Niche, real, and nothing else at this price does it.

If you are none of these, you are buying a badge.

What to buy instead

Same chip, different price: the ASUS Ascent GX10 — the same GB10 with 128GB — listed at $3,999 (1TB) in August 2026, but the ASUS US store listed it at $6,999 on 2026-09-13. Check every GB10 box price before you pick one. On that date the GX10 cost $2,300 more than the Spark’s $4,699 list price.

Same job, less money: a 128GB Ryzen AI Max+ 395 box ($3,499.99 GMKtec EVO-X2 or $3,799 Minisforum MS-S1 MAX on the vendor stores, 2026-09-17; it was ~$2,000-2,400 in August, sold out) is bandwidth-limited to the same order and runs the same MoE models at broadly comparable speeds — our model picks for it show gpt-oss 120B at 31-55 tok/s. No CUDA. If your stack is llama.cpp/Ollama anyway, that is $900-1,200 saved.

More speed, less capacity: if your models fit in 24-48GB, a used-GPU build demolishes every unified-memory box on generation speed.

Sources

• TechPowerUp / NVIDIA developer forums — official DGX Spark price change notice, February 2026
• pi3g — NVIDIA DGX Spark price in September 2026 (list price, Amazon US price, NVIDIA store stock)
• llama.cpp discussion #16578 — performance of llama.cpp on NVIDIA DGX Spark
• BigGo — DGX Spark performance and price debate
• NVIDIA developer forums — best inference results on Spark: SGLang
• ASUS eShop — Ascent GX10 listings, August 2026

See Also

• Best Models for the DGX Spark — what to run if you buy one
• The Cheapest Way to Run a 70B Locally in 2026 — where the Spark ranks on pure inference value
• Best Models for the Ryzen AI Max+ 395 — the half-price rival’s real numbers
• Dual RTX 3090 vs RTX 5090 — the discrete-GPU path at similar money
• Should You Buy RAM Now, or Wait Out the 2026 Shortage? — the shortage behind the Spark’s $700 hike
• Jetson Thor vs DGX Spark — NVIDIA’s other 128GB box, with identical 273 GB/s bandwidth
• DGX Spark Price: Every GB10 Box Compared — September 2026 prices for the Spark and all seven OEM boxes