Loading...
> >

Surface RTX Spark Dev Box

StarkdogStarkdog Posts: 178
in The Commons

I recently saw this announced by Microsoft as a more "portable" workstation for AI and other advanced uses.  What caught my eye was the processors and RAM:

Processor:

NVIDIA RTX Spark™ N1X with 6144-core GPU and 20-core CPU

Graphics: 

NVIDIA RTX Blackwell 6144-core GPU

Memory:

128GB LPDDR5x unified memory

 

At $6k, it is a bit out of my budget for now, but would this run D|S 4 or D|S 6?

I'm not up on all the NVIDIA updates, but is this more of a Tensor-core GPU, or does it also have CUDA cores needed for iRAY, dForce, etc?

I do like the 128GB of RAM though!

Thanks, -David

Comments

  • TimberWolfTimberWolf Posts: 408

    A few thoughts on this as I've been looking at the Spark box for my small games company since it was announced. First off, these are ARM64 boxes so everything Windows related that doesn't have an explicit ARM64 build will be run under the translation layer. Think Windows applications running under WINE in Linux - it's essentially the same thing. When ARM64 was first released the performance of non-native applications was appalling but it's much, much better now. DS does run under PRISM, the ARM translation layer, according to Reddit forums, with some occasional slowdowns and the odd visual bug. You won't be able to use iRay in DS4 but it will work in DS6 - those 6144 cores are CUDA cores. The main benefit here is the shared memory - subtract the memory used by the OS and any applications you have running and the rest is available to the GPU. Let's say for arguments sake that you have LibreOffice (ARM64 native) Writer and Calc open, a browser running a YouTube video with many other tabs loaded, you could still be looking at 100GB free for VRAM.

    Based on the specs (it supports OpenCL 3.0) dForce should work but at the moment there's no guarantee it will due to that translation layer. Also bear in mind that the whole setup has a 100W thermal limitation. It's an SoC - System on a Chip. In terms of rendering speed I don't see how it can compete with a conventional PC system - an RTX 4090 has 16384 cores for example - and the memory bandwidth on a standard PC is much higher. My read of this hardware is that it's primarily intended for local AI but it would be an excellent system for high VRAM renders as long as you don't mind waiting. If your scenes rarely exceed 24GB or even 48GB, you might be better off looking at more conventional hardware. I'm interested due to the shared memory and the fact that Blender has a native ARM64 distribution, but  I'll wait and see until the real-world performance figures are widely available.

Sign In or Register to comment.