It is the same CPU previously used in the overpriced DGX Spark.
It has 10 big Cortex-X925 cores, which are competitive with the Intel P-cores and with the AMD Zen cores, plus 10 small Cortex-A725 cores, which are similar in performance with the older Intel E-cores, from the Meteor Lake, Raptor Lake and Alder Lake generations. The current Intel E-cores are similar to Cortex-X4, i.e. they are much faster.
This Arm based CPU is more powerful that any Arm-based CPU previously used in a non-Apple PC, but in multi-threaded applications it is inferior to AMD Strix Halo CPUs.
The GPU of this is different from that of DGX, which was good only at ML/AI, but poor for graphics.
Here the GPU is likely to be good for graphics, and the top model will have up to 6144 FP32 execution units compared to 2560 of Strix Halo. But I assume that at least the top models will also be much more expensive than Strix Halo.
This NVIDIA CPU+GPU is limited to 128 GB of DRAM, while the successor of Strix Halo, which has been announced recently, offers up to 192 GB of DRAM, so NVIDIA continues its tradition of always providing less memory than its competitors, in order to have better profit margins.
From somewhere in the middle of Nvidia’s endless press waffle:
“The RTX Spark superchip features an NVIDIA Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores with FP4 precision, connected via the NVIDIA NVLink®-C2C chip-to-chip interconnect to a high-performance, 20-core NVIDIA Grace™ CPU.
MediaTek, a market leader in Arm-based system-on-a-chip designs, collaborated with NVIDIA on the custom CPU design, contributing to its best-in-class power efficiency, performance and connectivity.“
Sooooo they binned all their old DGX Spark crap to pawn on poor clueless consumers to try to be an ARM fighter and probably still fail miserably behind Apple Silicon and even likely older Qualcomm metal running Windows.
And Mediatek? Oof. I assume the SOC comes pre compromised out of the box.
As best as I can tell its something like the Apple M series SoC, but for Windows: CPU + GPU with unified memory.
It has 6,144 CUDA cores is similar to a RTX 4070 (5,888) but a lot less than a 4090 (16,384), but what it does have is support for FP4.
When they claim "1 Petaflop AI compute", thats what they mean. For comparison, a RTX 4090 has ~1.3 Petaflops of FP8 processing.
The second big deal is the NVLink-C2C interconnect, which provides up to 900 GB/s of bidirectional bandwidth between GPU and CPU. For comparison, the Apple M4 has 120 GB/s and the M3 Ultra has 819 GB/s. Notably, the Apple M series does not have FP4 support, so this could mean a significant performance improvement over Apple's offerings.
I don't understand why this isn't bigger news - this is a laptop SoC with actual gaming hardware running on ARM - unlike Apple's M series, which tend to have rather underwhelming perf in games compared to what the specs would suggest, finally we can have a thin-and-light with an efficient gaming GPU.
Considering how much Valve invested into ARM emulation, it's quite possible the next Steam Deck/handheld will use a variation of this (or at least there will be one using this as the SoC).
“Top agentic and AI developer workloads like GitHub Copilot, Claude Code, ComfyUI, Cursor and more now run across all modern PC silicon – making Windows the ideal platform for AI-assisted development”
None of these tools make use of local GPU whatsoever.
> None of these tools make use of local GPU whatsoever.
You can use local models with most of those, you’d just need to get something like Ollama, LM Studio or vLLM working - but that’s the boring and not flashy part, it’s probably easier to name the dev tools.
A Ryzen 9 laptop with 24GB of RAM and a 5070 is £1200. This thing better be dirt cheap. Basically add the price of the RAM, and deduct the price of having to use ARM and being forced into a Windows ecosystem. So what's that total? £1500 to £2000 depending on who you ask? Who is this for? Not gamers, not average Windows users, not professional developers or content creators.
Thing is, hardly any games are optimized for ARM. And no serious AI development occurs on Windows.
I understand that you want multiple models running concurrently, but then 128GB is starting to look cramped.
It's a bold move that goes one to one against Apple and AMD Strix Halo.
I'm looking at it and thinking, if it can run Linux at a fair price it could be great.
I'm curious. I am thinking, what does a non developer buy this thing, take it home, and do with it.
What does the unboxing and first 24 hours look like?
A mixture of emotions somewhere between thinking you are living in the future, and frustration at not actually being able to do much.
The target audience seems like wealthy early adopters, but that is about it.
Indeed. This NVIDIA CPU will have a lifetime overlapping with the successor of Strix Halo, which has been announced recently and which increases the maximum amount of DRAM to 192 GB.
The Strix Halo CPU has better multi-threaded performance than this, so the only advantage of NVIDIA is that the top variants will have a bigger GPU than Strix Halo and its successor, but I assume that the variants with a bigger GPU will also be much more expensive.
This reads as lip service for investors with no real life value. "Remember guys, we are still AI frontier material!"
ai frontier -- on your PC!
which is terrifying
MS announced Surface Laptop Ultra with this SoC. 15", 128GB RAM, no pricing yet.
1: https://www.theverge.com/tech/940584/microsoft-surface-lapto...
Nvidia will have held back the thing you are really hoping for, whatever it is.
Skimmed through the page no mention of what spark is. Is it a new ISA? SoC with CPU, GPU and NPU? Or just GPU+AI?
It is the same CPU previously used in the overpriced DGX Spark.
It has 10 big Cortex-X925 cores, which are competitive with the Intel P-cores and with the AMD Zen cores, plus 10 small Cortex-A725 cores, which are similar in performance with the older Intel E-cores, from the Meteor Lake, Raptor Lake and Alder Lake generations. The current Intel E-cores are similar to Cortex-X4, i.e. they are much faster.
This Arm based CPU is more powerful that any Arm-based CPU previously used in a non-Apple PC, but in multi-threaded applications it is inferior to AMD Strix Halo CPUs.
The GPU of this is different from that of DGX, which was good only at ML/AI, but poor for graphics.
Here the GPU is likely to be good for graphics, and the top model will have up to 6144 FP32 execution units compared to 2560 of Strix Halo. But I assume that at least the top models will also be much more expensive than Strix Halo.
This NVIDIA CPU+GPU is limited to 128 GB of DRAM, while the successor of Strix Halo, which has been announced recently, offers up to 192 GB of DRAM, so NVIDIA continues its tradition of always providing less memory than its competitors, in order to have better profit margins.
From somewhere in the middle of Nvidia’s endless press waffle:
“The RTX Spark superchip features an NVIDIA Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores with FP4 precision, connected via the NVIDIA NVLink®-C2C chip-to-chip interconnect to a high-performance, 20-core NVIDIA Grace™ CPU.
MediaTek, a market leader in Arm-based system-on-a-chip designs, collaborated with NVIDIA on the custom CPU design, contributing to its best-in-class power efficiency, performance and connectivity.“
https://nvidianews.nvidia.com/news/nvidia-microsoft-windows-...
Nvidia Grace is an ARM core.
Sooooo they binned all their old DGX Spark crap to pawn on poor clueless consumers to try to be an ARM fighter and probably still fail miserably behind Apple Silicon and even likely older Qualcomm metal running Windows.
And Mediatek? Oof. I assume the SOC comes pre compromised out of the box.
9 replies →
It's an ARM64 CPU with a nVidia GPU rather than garbage Adreno, not bad
As best as I can tell its something like the Apple M series SoC, but for Windows: CPU + GPU with unified memory.
It has 6,144 CUDA cores is similar to a RTX 4070 (5,888) but a lot less than a 4090 (16,384), but what it does have is support for FP4.
When they claim "1 Petaflop AI compute", thats what they mean. For comparison, a RTX 4090 has ~1.3 Petaflops of FP8 processing.
The second big deal is the NVLink-C2C interconnect, which provides up to 900 GB/s of bidirectional bandwidth between GPU and CPU. For comparison, the Apple M4 has 120 GB/s and the M3 Ultra has 819 GB/s. Notably, the Apple M series does not have FP4 support, so this could mean a significant performance improvement over Apple's offerings.
I don't understand why this isn't bigger news - this is a laptop SoC with actual gaming hardware running on ARM - unlike Apple's M series, which tend to have rather underwhelming perf in games compared to what the specs would suggest, finally we can have a thin-and-light with an efficient gaming GPU.
Considering how much Valve invested into ARM emulation, it's quite possible the next Steam Deck/handheld will use a variation of this (or at least there will be one using this as the SoC).
4 replies →
I just want a new NVIDIA Shield Pro; can't believe that 7 year old device is still the best media player on the market.
Copilot++ PC?
Just a bunch of nonsense text strung together.
“Top agentic and AI developer workloads like GitHub Copilot, Claude Code, ComfyUI, Cursor and more now run across all modern PC silicon – making Windows the ideal platform for AI-assisted development”
None of these tools make use of local GPU whatsoever.
> None of these tools make use of local GPU whatsoever.
You can use local models with most of those, you’d just need to get something like Ollama, LM Studio or vLLM working - but that’s the boring and not flashy part, it’s probably easier to name the dev tools.
Copilot: https://docs.github.com/en/copilot/how-tos/copilot-cli/custo... (also worked in Visual Studio Code)
Claude Code: https://code.claude.com/docs/en/model-config (here’s an example of a custom config in the wild with DeepSeek, same principle for local connection https://api-docs.deepseek.com/guides/agent_integrations/clau...)
Cursor: couldn’t find official docs but here’s z.AI docs https://docs.z.ai/devpack/tool/cursor (same principle for local API)
And you better be sure that I'm setting up linux on this. Windows is not the ideal platform for anything anyway.
> ComfyUI
Nitpick, but I remember running it specifically because there was a way to run LLMs on local hardware.
[dead]
[dead]
A Ryzen 9 laptop with 24GB of RAM and a 5070 is £1200. This thing better be dirt cheap. Basically add the price of the RAM, and deduct the price of having to use ARM and being forced into a Windows ecosystem. So what's that total? £1500 to £2000 depending on who you ask? Who is this for? Not gamers, not average Windows users, not professional developers or content creators.
Thing is, hardly any games are optimized for ARM. And no serious AI development occurs on Windows.
I understand that you want multiple models running concurrently, but then 128GB is starting to look cramped.
It's a bold move that goes one to one against Apple and AMD Strix Halo.
I'm looking at it and thinking, if it can run Linux at a fair price it could be great.
I'm curious. I am thinking, what does a non developer buy this thing, take it home, and do with it.
What does the unboxing and first 24 hours look like?
A mixture of emotions somewhere between thinking you are living in the future, and frustration at not actually being able to do much.
The target audience seems like wealthy early adopters, but that is about it.
I guess we shall see.
> but then 128GB is starting to look cramped.
Indeed. This NVIDIA CPU will have a lifetime overlapping with the successor of Strix Halo, which has been announced recently and which increases the maximum amount of DRAM to 192 GB.
The Strix Halo CPU has better multi-threaded performance than this, so the only advantage of NVIDIA is that the top variants will have a bigger GPU than Strix Halo and its successor, but I assume that the variants with a bigger GPU will also be much more expensive.
An M5 pro max with 128gb of RAM is $5200 so I’d expect some kind of price in that range.
I imagine it’s for AI researchers, professionals who work with models. Software engineers who want local models instead of cloud models.
To be honest, I don’t think it’ll be very popular with those demographics. But I think a company like Microsoft investing in local AI is a good thing.
Look to be disappointed Microsoft is all in on Copilot AI after having spent billions of dollars. It is boom or bust for Satya Nadella.