AI Compute RadarNVIDIA DGX Spark · hardware profile
Own the middle.
DGX Spark is not a faster graphics card. It is a different shape of machine: one pool of memory big enough for models a consumer GPU cannot hold at all.
What DGX Spark costs
NVIDIA's official price for the Founders Edition with 4 TB of storage is $4,699. It is the only DGX Spark figure this site quotes: partner systems from other makers carry their own prices, and street prices are not tracked here on principle.
| Machine | Launch list price | Source |
|---|---|---|
| DGX Spark | $4,699 | NVIDIA official price, Founders Edition, 4 TB |
| Mac Studio M5 Ultra 96GB | $5,499 | Apple launch list price, Sep 2026 (M5 Ultra base, 64-core GPU, 96 GB); watts = Apple-measured max wall draw of the 80-core 512 GB config, an upper bound for this one |
| Mac Studio M5 Max 36GB | $2,499 | Apple launch list price, Sep 2026 (M5 Max base, 32-core GPU, 36 GB); watts = Apple-measured max wall draw of this config |
| RTX 5090 | $1,999 | NVIDIA launch MSRP, Jan 2025 |
| RTX 4090 | $1,599 | NVIDIA launch MSRP, Oct 2022 |
Each figure is the vendor's launch list price with its date, taken from that machine's own profile — never a street price.
Why unified memory matters
A consumer GPU keeps model weights in its own VRAM. When a model does not fit, it spills into system RAM over PCIe and slows to a crawl. A unified-memory system gives CPU and GPU one large pool — the model simply fits. Capacity, not raw speed, is what it buys you.
| Metric | DGX Spark | RTX 4090 |
|---|---|---|
| Memory | 128 GB | 24 GB |
| Bandwidth | 273 GB/s | 1008 GB/s |
Roughly five times the memory, roughly a quarter of the bandwidth. Spark runs models a discrete GPU cannot hold at all — but on a model both can hold, the GPU generates tokens faster.
NVIDIA's published claims
- Inference, one unit
- Models up to 200B parameters
- Fine-tuning, one unit
- Models up to 70B parameters
- Two units linked
- Models up to 405B parameters
Our own arithmetic
Not an NVIDIA statement: 200 billion parameters only fit in 128 GB at roughly 4-bit precision (200B × 0.5 bytes ≈ 100 GB of weights) before any context memory. NVIDIA does not publish the precision behind that number, so read “200B-class” as heavily quantized.
Honest caveats
- Capacity is not speed: memory bandwidth is well below a high-end discrete GPU, so tokens per second on a model that fits both will favour the GPU.
- It is an ARM64 machine. Most mainstream AI tooling supports it, but niche x86-only packages may need work.
- Vendor model-size claims assume aggressive quantization. A '200B-class' model means heavily quantized, not full precision.
- If your heavy jobs are occasional, renting GPU hours is almost always cheaper than buying.
Why we publish no tokens-per-second figure
Tokens per second depend on the model, its quantization, the context length and the runtime — and every runtime update moves the number. A single figure would be wrong for most readers and stale within weeks, so this site publishes none. It measures what it can measure on a schedule: memory, prices, sizes.
What can be said comes from physics, not benchmarks. Generating a token means reading most of the model's active weights out of memory, so memory bandwidth is the hard ceiling on speed. DGX Spark's 273 GB/s against roughly 1,000 GB/s on a top consumer graphics card means that on a model which fits both, the card's ceiling is about four times higher — real runtimes reach only a fraction of either ceiling, but the ratio between them holds.
The consequence: Spark buys capacity — models that would not fit a consumer card at all — not speed. If you want fast output on a model that already fits a GPU, the GPU wins; if you want a model that needs 100 GB of memory to exist on your desk at all, nothing consumer-priced competes.
Bandwidth figures are vendor specifications (NVIDIA: 273 GB/s for DGX Spark; 1,008 GB/s for the GeForce RTX 4090). The ratio is arithmetic on those two numbers, not a measurement of any model.
Two Sparks, or four?
NVIDIA's own product page states that its ConnectX networking lets up to four DGX Spark systems be connected to work with models of up to 700 billion parameters. That is a capacity claim, and it is real: four pooled machines hold four times the weights.
It is not a speed claim. The link between boxes is a network cable, orders of magnitude slower than the memory inside each box, so a model split across units generates at the pace of that link, not of the chips. And the parameter figure carries the same fine print as the single-box claim: it assumes heavy quantization, and the precision behind the number is not stated.
Honest reading: linking Sparks is for people who need a very large model to fit at all and can accept slow output. If the goal is speed on a model that fits one box, the second box buys nothing — and for an occasional big job, an hour of a rented data-center GPU is cheaper than a second machine.
Source: NVIDIA DGX Spark product page ("up to four NVIDIA DGX Spark systems … AI models of up to 700 billion parameters").
DGX Spark or RTX Spark? Two different Sparks
Searches for “RTX Spark price” land on this page, so to be clear: DGX Spark and RTX Spark are different products. DGX Spark is the Linux desktop AI system described here, built on the GB10 Grace Blackwell superchip and sold by NVIDIA at $4,699. RTX Spark is a superchip for Windows 11 laptops and compact desktops that NVIDIA and Microsoft announced on 31 May 2026; systems from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI are, in NVIDIA's words, available this fall.
What NVIDIA has published for RTX Spark: up to 128 GB of unified LPDDR5X memory, a 20-core Grace CPU with a 6,144-core Blackwell RTX GPU in the top configuration (an 18-core, 5,120-core tier with up to 64 GB sits below it), up to 1 petaflop of FP4 AI compute, a 45 to 80 W TDP, Windows 11. What it has not published: a price and the memory bandwidth. Without the bandwidth this site cannot say how fast it reads a model, and without a price it cannot do the break-even, so RTX Spark gets a profile here once both exist.
| Metric | DGX Spark | RTX Spark |
|---|---|---|
| Memory | 128 GB LPDDR5x unified | up to 128 GB LPDDR5X unified |
| Memory bandwidth | 273 GB/s | not published |
| CPU | 20-core Arm (GB10) | up to 20-core NVIDIA Grace |
| GPU | Blackwell, GB10 superchip | up to 6,144-core Blackwell RTX |
| Power | 140 W chip TDP | 45–80 W TDP |
| Operating system | DGX OS (Ubuntu 24.04, ARM64) | Windows 11 |
| Price | $4,699 (Founders Edition, 4 TB) | not published |
| Availability | available now | systems “available this fall” (NVIDIA) |
Sources: NVIDIA press release of 31 May 2026 (“NVIDIA and Microsoft Reinvent Windows PCs for the Age of Personal AI”) and the NVIDIA RTX Spark product page, both read on 25 September 2026. NVIDIA has published neither a price nor a memory-bandwidth figure for RTX Spark.
What fits in 128 GB
Context length: 8K tokens · f16 (default)
All memory figures are estimates: measured quantized file size + computed context memory + runtime overhead, with a 12% safety margin on your hardware. Real usage varies with runtime version and settings.
Radar Card
DGX Spark as a Radar Card
The same machine as a collectible spec card: a hologram drawn from its dimensions and parts list, sourced facts with their status, and a number issued in claim order that stays in your browser.
Open the card →Spark or rent?
Buying makes sense when large local models are your daily routine and your data must stay in the building. If the heavy work happens a few times a month, renting is cheaper — the break-even maths is on the rental page.
Break-even →Who it is for
- + You run 70B-class or larger models locally, most days.
- + Your data cannot leave your desk.
- + Capacity matters more to you than tokens per second.
- + You are comfortable on an ARM64 Linux machine.