NVIDIA DGX Spark Technical Comparison

The newly announced DGX Spark is NVIDIA’s latest appliance for developers looking for a desktop solution to develop and fine tune generative AI and reasoning AI models. These are particularly challenging workloads as they have a large memory footprint, so are too large to run on many existing GPUs.

For this reason the DGX Spark has a totally different architecture to a traditional AI workstation. In this article we’ll compare these two types of AI development systems alongside cloud alternatives, so you can pick the best solution for your AI project.

Traditional System Architecture

Traditional AI workstations are built around the industry-norm of an x86 CPU (either AMD or Intel) plus one or more NVIDIA RTX GPUs connected via the PCIe bus. The major benefit of this approach is that as a well-established system architecture over decades, such systems are extremely cost-effective and very easy to expand with more powerful processors, memory, storage and networking as your needs grow. The downsides are that the CPU and GPU have separate pools of memory and that NVIDIA RTX GPUs are currently limited to 96GB of memory, so require multiple cards to run generative AI and reasoning AI models effectively.

DGX Spark System Architecture

In contrast, the DGX Spark has been built from the ground-up to provide a single large memory pool. It achieves this by replacing the old discrete CPU and GPU paradigm with a SoC, combining CPU and GPU cores together in a single unit. Known as the GB10 Grace Blackwell Superchip, this SoC comprises a Blackwell GPU and 20 Arm CPU cores, sharing a unified 128GB of memory. Two DGX Sparks can also be clustered together using a special LinkX cable, effectively creating a combined 256GB memory pool for running even larger models.

The GPU element of the SoC is equipped with the latest 5th gen Tensor cores, which are optimised for performing FP4 calculations, the most commonly used precision level for generative AI and reasoning AI models, while the CPU element has ten Cortex X925 and ten Cortex A725 cores.

The SoC is supported by 4TB of NVMe SSD storage, alongside a 10GbE Ethernet port, ConnectX-7 SmartNIC and WiFi 7 for ingesting data. Just like a typical desktop PC, the DGX Spark has a HDMI port to connect a monitor, plus four USB Type C ports to connect peripherals.

A side benefit of the SoC approach is that the DGX Spark is much more energy efficient than a traditional AI workstation, so it has much lower power requirements and is far more compact, being a fraction of the size.

It’s also worth noting that the DGX Spark is designed for a single user. In contrast, AI workstations can be used by up to four users in parallel, with the GPU(s) being partitioned into separate instances by MIG.

DGX Spark Software Stack

The DGX Spark runs DGX OS, which is a NVIDIA-customised version of Ubuntu Linux. This enables developers to unleash the full potential of their DGX Spark with a proven software platform built around a complete AI software stack including access to NVIDIA NIM microservices, Blueprints and AI Enterprise.

This is of course the same software stack as other NVIDIA AI appliances, making it easy to transfer your newly-developed AI model onto more powerful DGX, HGX, EGX and MGX systems, whether they’re in your own datacentre or in the cloud.

Which system is right for me?

So, now that we’ve covered how the DGX Spark architecture differs from traditional AI workstations, let's take a look at which system is right for different AI projects. This is crucial, as one system is not necessarily superior to the other, each have their own strengths and weaknesses. We’ll be including cloud services in this comparison as they are also a great choice for AI model development.

In the first row of the table below you can see the AI performance, measured in TOPS, of the five different system configurations. What should be apparent here is that in terms of raw throughput the DGX Spark is outpaced by the various discrete GPUs in the other systems. However, this is only part of the story and doesn’t take into account AI model size. For this you need to look at how much GPU memory is available, as this will determine how large your AI model can be.

For instance, the popular GeForce RTX 5090 is roughly three times as powerful as the DGX Spark, but thanks to its extremely limited memory (32GB) it can only run small AI models of up to 51 billion parameters. In contrast the DGX Spark’s 128GB memory allows it to run AI models with up to 200 billion parameters. In short, a 3XS AI workstation with GeForce RTX 5090 GPU(s) will be faster than the DGX Spark with small models, but simply cannot run large models. Given that both systems have a similar cost it’s therefore crucial you pick the right configuration for your AI project and are aware that the once desirable RTX 5090 is no longer as attractive for AI model development.

Moving up the price stack the newly announced RTX PRO 6000 Blackwell strikes a better balance between raw AI performance and memory. It is four times faster than the DGX Spark and thanks to its 96GB of memory is able to run AI models with up to 153 billion parameters. It’s also worth keeping an eye out for the RTX PRO 6000 Blackwell Max-Q version, which runs slightly slower, but as it consumes less power we’re able to offer systems with up to four GPUs to run really large models with up to 614 billion parameters.

Finally, it shouldn’t be forgotten that buying a desktop system outright is not your only option for AI model development and that we also offer cloud workstations with no up-front costs or hidden fees on a flexible subscription model. Our cloud workstations can be configured with all the previously mentioned GPUs, plus a system with up to eight RTX PRO 6000 Blackwell Server GPUs which could run up to 1.22 trillion parameter model with very high performance.

Conclusion

Thanks to its innovative SoC architecture, that ditches the ageing x86 instruction set in favour of the far more efficient Arm instruction set, the DGX Spark is the only reasonably-priced desktop device that can develop AI models with a large number of parameters, such as generative AI and reasoning AI models.

Being a desktop device, the DGX Spark is also a practical choice for organisations that are not able to upload their data into a cloud service. Ultimately, as with all things in AI, you need to pick the right configuration for the job, or you’ll not be able to develop, train or inference your AI model efficiently.

Follow this link to pre-order your NVIDIA DGX Spark now or talk to our AI experts about your project requirements or any of the other system configurations discussed in this article.