[SethTro]

Article doesn't seem to mention price which is $4,000 which makes it comparable to a 5090 but with 128GB of unified LPDDR5x vs the 5090's 32GB DDR7.

[Tepix]

They're in a different ballback in memory bandwidth. The right comparison is the Ryzen AI Max 395 with 128GB DDR5-8000 which can be bought for around $1800 / 1750€.

[bilekas]

$4,000 is actually extremely competitive. Even for an at-home enthusiast setup this price is not our of reach. I was expecting something far higher, that said, nVidia's MSRP is something of a pipe dream recently so we'll see when it's actually released and the availability. Curious also to see how they may scale together.

[Xss3]

A warning to any home consumer throwing money at hardware for AI (fair enough if you have other use cases)...

Things are changing rapidly and there is a non insignificant chance that it'll seem like a big waste of money within 12 months.

[Tepix]

If you compare DGX Spark with Ryzen AI Max 395, do you still think that $4000 for the NVidia device is very competitive?

To me it seems like you're paying more than twice the price mostly for CUDA compatibility.

[EnPissant]

A 5090 is $2000.

[sandworm101]

Msrp, but try getting your hands on one without a bulk order and/or camping out in a tent all weekend. I have seen people in my area buying pre-biult machines as they often cost less than trying to buy an individual card.

[adrian_b]

But you put in a $1500 PC (with 128 GB DRAM).

Still, a PC with a 5090 will give in many cases a much better bang for the buck, except when limited by the slower speed of the main memory.

The greater bandwidth available when accessing the entire 128 GB memory is the only advantage of NVIDIA DGX, while a cheaper PC with discrete GPU has a faster GPU, a faster CPU and a faster local GPU memory.

[CamperBob2]

And about 1/4 the memory bandwidth, which is what matters for inference.

[threeducks]

More precisely, the RTX 5090 has a memory bandwidth of 1792 GB/s, while the DGX Spark only has 273 GB/s, which is about 1/6.5.

For inference, the DGX Spark does not look like a good choice, as there are cheaper alternatives with better performance.

[nialse]

Well, that’s disappointing since the Mac Studio 128GB is $3,499. If Apple happens to launch a Mac Mini with 128GB RAM it would eat Nvidia Sparks’ lunch every day.

[pjmlp]

Only if it runs CUDA, MLX / Metal isn't comparable as ecosystem.

People that keep pushing for Apple gear tend to forget Apple has decided what industry considers industry standards, proprietary or not, aren't made available on their hardware.

Even if Metal is actually a cool API to program for.

[newman314]

Agreed. I also wonder why they chose to test against a Mac Studio with only 64GB instead of 128GB.

[moondev]

Just don't try to run a NCCL