It won't, the top tier RTX Spark has the same exact CPU and GPU as DGX Spark, so you can check DGX Spark CPU benchmarks to see how it fares. Spoiler: it's about M3 Max level. And they're only coming this fall.

Nah, still ~300GB/s memory bandwidth. That will be slower than the M5 max, by a wide margin for LLM inference.

M5 max is 3x stronger and 50% more power efficient. nice try though.

My DGX Sparks are the first and only devices I have with 200W USB-C PD. Low power by AI workstation standards, but intolerable in a laptop.

It's worth noting that Nvidia power management on Linux has been absymal. There also aren't any of the usual power management options to see how much power things are using, which is quite atypical for a modern system.

Nvidia really threw stuff over the wall with the DGX Spark release. They don't seem to really care. I sort of think they'll spend a little more time on Windows, where there's no pesky upstreaming to do and they can just do whatever, but man, it's such typical hubris from Nvidia to build such an expensive box with good chips but make it basically unsupportable and roasty hot all the time.

You also generally have to run an ever more stale two year old Ubuntu derived DGX OS to get anywhere, with bespoke kernel and drivers all. None of it is well supported, none of it just works like a comparable PC or even well behaved arm system would.

As for other ARM, there were rumors AMD Sound Wave is/was going to be a ~10W arm APU, but there hasn't been much said about it lately. Honestly given the ram crunch, it's maybe just not worth trying to build a system with a cheap core, if the rest of your costs are going to stay so stratospheric. https://www.techpowerup.com/341848/amd-sound-wave-arm-powere...

ARM cpu made by MediaTek.

2 comments in total there

yes, same chip

+ Windows

+ Screen

- ConnectX-7 Smart NIC

I heard leaked geekbench putting it behind the m3, which is couple years old now.

All I care about is if I can get one of these for significantly less than a dgx and get Linux on it for some cuda Blackwell kerneling.

Strix halo's 8060S gpu is very weak, and is roughly equivalent to a 4060 laptop GPU, whereas GB10's gpu is equivalent to a desktop 5070. For LLM throughput, tok/s is similar due to bottleneck by memory bandwidth, but the GB10 has 3x faster prefill. People have also been able to squeeze out much better performance on GB10 using NVFP4 and other improvements in the months after the DGX Spark launch, so don't be misled by early lackluster benchmarks. For the RTX Spark, which also targets gaming and creative applications, the 3x faster GPU is quite nice.

Or like a m4 max? This thing has <300GB/s vs the max with 550GB/s

All those CUDA cores in the sparks but they're starved for memory bandwidth.

I am still waiting for NVidia to release a system that legit beats 3090 maxxing for the home gamer...

  Spark:
  OS: Windows/Ubuntu
  Mbw: 300GB/s
  Cuda cores: 6000
  GPU accelerated containers: yes


  M5 max:
  OS: macOS
  Mbw: 600GB/s
  Cuda cores: 0
  GPU accelerated containers: no

Is CUDA really a lead for long? Aren’t all the latest competitive approaches avoiding all the standard software stacks and writing deeply customized software that is very directly tied to whatever hardware they use?

And is it really a way to lock in people? With AI coding tools, isn’t it trivial to write software on top of CUDA and rewrite it to target some other hardware?

The only Question is is it worth suffering hip and x86? I suspect a lot of folks might like a machine that mimics their GB300 But costs less than a dgx.

Also I heard the tensor core instructions on the dgx are gimped and you’re better off with a rtx pro x000. Is that the same with these machines?

Not at all. This is a more like what Apple has been doing the past few years. A bunch of decent arm cores paired with a beefy integrated GPU.

No.

Developers weren’t really “forced” to support ARM. They simply recognized that all future Macs would be ARM, whereas most new PCs would continue to run on x86. So the incentive to adopt ARM was much weaker on the PC side.

They didn’t though. Rosetta 2.

Microsoft can do the same for windows - they need to address the fat bundle solution that Apple came up with, but for Windows, though ..

"Unified" means that it's shared between CPU and GPU, I believe.

But yes, it tends to be soldered on.

It's clear gaming was not a major concern, it's just "good enough" for someone running AI models and occasionally wants to play some games, not made to primarily play games.

Yep. I noticed the press releases talk about all the partners they have. It seems like a desperate attempt to manufacture a consensus to invest in this new hardware instead of leaving it sort of abandoned like the other Windows ARM stuff. But the problem is that these attempts end up having a few very visible apps working on the architecture and others not actually doing anything substantial.

Sure the graphics capabilities are probably very good. But if you’re a game developer who has traditionally built on Windows on x86 chips, would you want to invest in this new chip or invest in making games for the Apple ecosystem? Aren’t there more new customers to reach in the Apple world than this new Nvidia world?

Microsoft sees windows purely as a platform to sell AI products these days.

That's what they're working on, in theory, with Windows K2.