NVIDIA's $4,999 DGX Spark 64GB is a better buy for local agents than for fine-tuning
NVIDIA's 64GB DGX Spark goes on sale from six PC makers on October 23 at $4,999, about $2,000 to $4,000 below in-stock 128GB units today. Two clustered units cost more than one 128GB machine, so the purchase makes most sense for agents on models that fit in 64GB.
The Engineer · Build desk
What happened
- The 64GB model keeps the GB10 Grace Blackwell chip, DGX OS and software stack of the 128GB version, and NVIDIA says it runs models of up to 100 billion parameters.
- Two units joined by a direct QSFP cable pool their memory to 128GB, enough for models of up to 200 billion parameters by NVIDIA's count.
- Cluster Assistant in the NVIDIA Sync app detects connected units, validates their configuration and sets up the ConnectX-7 network between them.
- A Model Launcher due at the end of the month will install Qwen 3.8 27B on one unit or a pair and set up the OpenCode coding agent to use it.
Why it matters
- What it costs cost Getting to 128GB by pairing two 64GB units costs $998 to $2,998 more than one 128GB machine at today's street prices, so the low entry price only pays off while the work fits in 64GB.
- What it costs cost On NVIDIA's own scaling result, adding a second unit raises the price of each unit of throughput by about 18 percent.
- What doesn't add up contradiction NVIDIA pitches the platform for fine-tuning while Tom's Hardware assigns fine-tuning to the 128GB model, so a fine-tuning budget built on the 64GB unit depends on NVIDIA's word alone.
- What's now possible capability A team can buy one unit for agent work now and add a second later without reworking software, because both nodes run the same stack and Sync configures the link.
Next story · Build · 18h ago
One early test puts a 120B model at 63 tokens per second on an RTX Spark laptopConfirmed3 publishers
NVIDIA measured its 1.7x clustering gain on its own test bench, using Qwen 3.8 27B [6]. Tom's Hardware reports that the same model fits in 32GB of memory with limited context [14]. That means the speedup was measured on a model that already fits on one 64GB unit [6][14]. Two units gave 1.7 times the throughput of one, which is 85 percent of linear scaling [17]. For that result to carry over, a buyer's model would have to split across the link the same way, at a similar size and context length. The 200-billion-parameter ceiling for a pair comes from pooling memory. The only clustering speed figure NVIDIA gives is that same 1.7x [13][6].
The clustering software is the strongest engineering in the launch. Earlier this year, Tom's Hardware joined two Dell GB10 systems using utilities written by the community [8]. NVIDIA now does that job in its Sync app. Every node runs the same software stack, so adding a second unit needs no reconfiguration, according to NVIDIA [7]. The 64GB unit keeps the same ConnectX-7 RDMA NIC as the 128GB model [2]. NVIDIA's "without any additional setup" covers the software [12]. Someone still has to plug in the QSFP cable [13].
Now the cost. One 64GB unit starts at $4,999 [4]. Tom's Hardware reported that in-stock 128GB GB10 systems sell for roughly $7,000 to $9,000, well above NVIDIA's adjusted list price [20]. On those figures the small unit is about $2,000 to $4,000 cheaper today [21], though that sets a launch price against street prices. Tom's Hardware wrote of the launch price: "Given the ever-shifting prices of memory and storage right now, they might not stay there for long." [5] Clustering two units to reach 128GB costs $9,998 [22]. That is $998 to $2,998 more than one 128GB box at current street prices [23]. For the extra money the pair gets two memory pools at 273 GB/s each [2], 546 GB/s in total [19]. On NVIDIA's 1.7x figure, each unit of throughput costs about 18 percent more in the pair than in a single unit [18].
The sources disagree on fine-tuning. NVIDIA lists it among the platform's uses, alongside agents and inference [9]. Tom's Hardware says the 128GB Spark remains the machine for memory-hungry work such as fine-tuning [16]. Both models have the same chip and software [1], so the disagreement comes down to memory. NVIDIA's own workflow examples lead with agents: a coding or research agent left running around the clock, with a second unit added for longer context or several agents at once [10].
The sources do not include a cloud instance price. Any buy-versus-rent comparison has to come from a team's own bill. In my view the 64GB unit is a sound purchase for one job: an always-on agent running a model in the 27B class on the device, with no cloud dependency, as NVIDIA describes it [11][14]. For fine-tuning, the evidence here points to the 128GB machine [16].
What to watch
- Whether 64GB units actually sell at $4,999 after the October 23 launch, given Tom's Hardware's warning about memory prices.
- Independent clustering throughput on a model too large for one 64GB unit, somewhere between 100 and 200 billion parameters.
- Published fine-tuning results on a single 64GB Spark that would settle the disagreement between NVIDIA and Tom's Hardware.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence62
- Adoption
- Not scored
- Hype gap+30
- Incentives68
- Confidence60
Perspective Coverage
13 publishers- Builder
- Builder 49%
- Operator
- Operator 35%
- Investor
- Investor 16%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The 64GB configuration retains the GB10 Grace Blackwell Superchip, DGX OS and full NVIDIA AI software stack, same as the 128GB model, and supports up to 100-billion-parameter models fully on device.
- [2]
64GB Sparks retain the same ConnectX 7 RDMA NIC as the 128GB models; the same 20-core Arm CPU complex and 273 GB/s of shared CPU-GPU memory bandwidth carry over unchanged.
- [3]
NVIDIA DGX Spark will be available this month with 64GB of unified memory from manufacturer partners Acer, ASUS, Dell, Gigabyte, HP and MSI.
- [4]
64GB GB10 systems from Acer, Asus, Dell, Gigabyte, HP and MSI are slated to start at $4999 when they launch on October 23.
- [5]
Given the ever-shifting prices of memory and storage right now, they might not stay there for long.
- [6]
In NVIDIA's Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7x performance compared with a single system.
- [7]
The cluster assistant feature of the NVIDIA Sync app detects connected units, validates device configuration and configures the ConnectX-7 network; every node runs the same NVIDIA software stack, so nothing needs to be reconfigured when scaling from one unit to two.
- [8]
Community-developed utilities have sprung up over the past year to make clustering more seamless; Tom's Hardware used them to cluster two Dell GB10 systems earlier this year.
- [9]
NVIDIA describes DGX Spark as a complete local AI platform for agents, inference, fine-tuning, data science and edge development.
- [10]
NVIDIA's workflow example: keep a coding or research agent running around the clock on DGX Spark; a cluster provides capacity for larger models, longer context windows or multiple agents at once.
- [11]
NVIDIA says the new SKU runs capable local agents on device, privately, without cloud dependency.
- [12]
Two units can cluster together via NVIDIA Sync Cluster Assistant without any additional setup.
- [13]
Every DGX Spark ships with a built-in ConnectX-7 NIC; two units can connect directly with a QSFP cable, pooling memory to 128GB and expanding model support to up to 200 billion parameters, with twice the memory bandwidth and up to 1.7x the performance.
- [14]
Dense models like Qwen 3.8 27B can now fit within 32GB of RAM, albeit with limited context.
- [15]
NVIDIA Sync Model Launcher, coming at the end of the month, downloads and launches Qwen3.8 27B on a single DGX Spark or a cluster, makes it accessible from users' laptops, and sets up OpenCode to use the model.
- [16]
The 128GB DGX Spark and Spark-alikes will remain available for those who need space for larger models, for scaling across clusters, or for more memory-hungry local AI work like model fine-tuning.
- [17]
A 1.7x gain from two units is 85 percent of linear scaling.
- [18]
On NVIDIA's 1.7x figure, cost per unit of throughput is about 18 percent higher for a two-unit cluster than for one unit.
- [19]
Two clustered units have 546 GB/s of combined memory bandwidth.
- [20]
A 128GB GB10 system, if in stock, costs roughly $7000 to $9000 right now, far above Nvidia's adjusted MSRP.
- [21]
At launch price, a 64GB unit is about $2,000 to $4,000 below current street prices for in-stock 128GB GB10 systems.
- [22]
Two 64GB units at the starting price cost $9,998.
- [23]
Two clustered 64GB units cost $998 to $2,998 more than one 128GB GB10 system at current street prices.
Sources
12 independent publishers whose own reporting we read for this story.
- NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI | NVIDIA Blog
blogs.nvidia.com
1 article · October 2, 2026
- cryptobriefing.comNvidia launches DGX Spark 64GB configuration with hardware partners
1 article · October 2, 2026
- dev.toNVIDIA's 64GB DGX Spark makes local AI a working-set decision
2 articles · October 6, 2026
- digitaltrends.comNVIDIA DGX Spark 64GB is for people who want to build with AI, not just use it - Digital Trends
1 article · October 2, 2026
- hothardware.comNVIDIA DGX Spark 64 GB Reduces Memory to Stem RAM Price Increases
1 article · October 2, 2026
- nvidianews.nvidia.comNVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI
1 article · October 2, 2026
- pcmag.comNvidia announces a version of DGX Spark with 64 GB of unified memory for $4,999, or $1,000 more than the 128 GB version at launch
1 article · October 2, 2026
- prnewswire.comGIGABYTE AI TOP ATOM 64GB Unified Memory Version Expands Possibilities for Desktop AI Development
1 article · October 2, 2026
- servethehome.comNVIDIA DGX Spark 64GB Launched and Big 128GB GB10 Price Increases - ServeTheHome
1 article · October 3, 2026
- superpowerdaily.comNVIDIA Adds a $4,999 DGX Spark With 64GB of Memory for Local AI | Superpower Daily
1 article · October 2, 2026
- tomshardware.comNvidia introduces 64GB DGX Spark to throw local AI fans a lifeline amid the RAMpocalypse
1 article · October 2, 2026
- videocardz.comNVIDIA DGX Spark drops to 64GB memory but costs more than the original 128GB version - VideoCardz.com
1 article · October 2, 2026
- xda-developers.comNvidia's DGX Spark is getting a 64GB RAM variant to make running local models more affordable
1 article · October 2, 2026
Topics and entities
Topics
- Local AI Inference
- Memory Chip Pricing
- Desktop AI workstations
Entities
- Qwen 3.8 27B
- Nvidia
- Micron
- DGX Spark
- GIGABYTE AI TOP ATOM
- NVIDIA GB10 Grace Blackwell Superchip