News:

Willkommen im Notebookcheck.com Forum! Hier können Sie über alle unsere Artikel und allgemein über notebookrelevante Dinge diskutieren. Viel Spass!

Main Menu

Post reply

Other options
Verification:
Please leave this box empty:
Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by Thing is
 - Today at 14:05:39
And here is another Spark- and Strix Halo-fit MoE model request/begging (would be perfect for this 273 GB/s (256-bit, LPDDR5X) device):
Quote from: reddit.com/r/LocalLLaMA/comments/1u90awk/glm52_flash_when_jokeGLM-5.2 Flash when? (joke)

I'm very happy with Z.ai's decision to open source GLM 5.2...

With that being said, a successor to GLM-4.7-flash would be AMAZING. Literally anything in the 27-120B range (MoE or dense) 🤤
The from the comments:
Quote from: huggingface.co/zai-org/GLM-5.2/discussions/3We need some Air or at least some Flash
(99 hearts voted on the first post, as of this writing).

GLM-4.5-Air and GLM-4.7-Flash are among my favorite models, please consider bringing back models for local users.
Posted by Errors were made
 - Today at 13:33:00
Quote from: Thing is on Today at 10:16:22AMD Strix Halo: 256 GB/s = 256-bit * 8000 MT/s / 1000 / 8.
NVIDIA Spark: 300 GB/s = 256-bit * 9400 MT/s / 1000 / 8.

Quote from: Thing is on Today at 10:16:22that it would be really nice, if these devices had twice the memory (= 256 GB RAM / unified memory).

Well it exists, you just have to pay for it:

videocardz.com/newz/asus-expertcenter-pro-et900n-g3-launches-with-nvidia-gb300-748gb-memory-and-future-windows-support-costs-117600

wccftech.com/apples-tim-cook-cries-out-that-memory-prices-have-become-unsustainable-warns-that-unavoidable-price-hikes-are-coming/

Looks like apple are about to rise prices.

Quote from: Thing is on Today at 10:16:22techpowerup.com/gpu-specs/nvidia-gb20b.g1114 states that the chip is built on a 5nm node, not 3nm.
And why do both of the following state that DirectX is not supported?

This is from an official Nvidia slide over a year ago:

cdn.wccftech.com/wp-content/uploads/2025/08/2025-08-26_19-59-53-scaled.png

Better to refer to official documentation from the company that makes the chip than 3rd party sites which do make mistakes sometimes.

The graphics api support depends on the OS running on the chip. DGX Spark runs an distro based on Ubuntu, so it's obviously not going to support it. RTX Spark runs windows, so it will.
Posted by Thing is
 - Today at 10:31:22
PS: The Spark/GB10 chip in a Yoga Pro 9n and not in a Legion doesn't exactly point to the rumors, that the Spark's performance is going to be between a 5070 Laptop and 5070 Ti Laptop. Also, techpowerup.com/gpu-specs/nvidia-gb20b.g1114 states that the chip is built on a 5nm node, not 3nm. Another not exactly reassuring performance rumor expectation, more like a rumor buster. But we will see.

And why do both of the following state that DirectX is not supported?
Quote from: techpowerup.com/gpu-specs/gb10.c4342Built on the 5 nm process, and based on the GB20B graphics processor, the card does not support DirectX. Since GB10 does not support DirectX 11 or DirectX 12, it might not be able to run all the latest games.
Quote from: techpowerup.com/gpu-specs/nvidia-gb20b.g1114NVIDIA's GB20B GPU uses the Blackwell 2.0 architecture and is made using a 5 nm production process at TSMC. GB20B does not support DirectX.
Posted by Thing is
 - Today at 10:16:22
QuoteThe company claims it can run 120 billion parameter models locally
Thing is:
Quote from: reddit.com/r/LocalLLaMA/comments/1u8kr2o/we_need_a_80160b_model_urgently_the_unified/We need a 80-160B model urgently. The unified memory device market needs more Models.

(read full text there)

For anyone who doesn't know, an alternative from AMD is called Strix Halo:
AMD Strix Halo: 256 GB/s = 256-bit * 8533 MT/s / 1000 / 8.
NVIDIA Spark: 273 GB/s = 256-bit * 8533 MT/s / 1000 / 8.

There are so many recently released 200B to 300B parameters MoE models, that it would be really nice, if these devices had twice the memory (= 256 GB RAM / unified memory). Then a good quant of those models (see below) could be run.

An alternative in the meantime could be to run a Qwen3.6-27B quant on a used 24 GB VRAM GPU (typically RTX 3090) (24 GB is the absolute bare minimum), a model that performs much better for its parameters size, because it's a dense model, see:
artificialanalysis.ai/?models=gpt-oss-120b%2Cgpt-oss-120b-low%2Cgemma-4-31b%2Cgemma-4-31b-non-reasoning%2Cmistral-medium-3-5%2Cdeepseek-v4-flash-non-reasoning%2Cdeepseek-v4-flash-high%2Cdeepseek-v4-flash%2Cminimax-m2-7%2Cstep-3-7-flash%2Cmimo-v2-5-0424%2Cqwen3-6-35b-a3b%2Cqwen3-5-122b-a10b%2Cqwen3-6-27b%2Cqwen3-6-27b-non-reasoning%2Cqwen3-6-35b-a3b-non-reasoning%2Cqwen3-5-122b-a10b-non-reasoning&intelligence=artificial-analysis-intelligence-index
Posted by Sergey
 - Yesterday at 20:50:15
Quote"15" here to align to with the rest of the article

Good point. Thanks!
Posted by Typo
 - June 03, 2026, 14:42:36
Quote from: Redaktion on June 03, 2026, 14:24:18Lenovo Yoga Pro 9n: A 16-inch creator-focused laptop

Should state 15" here to align to with the rest of the article.

Curse Lenovo and their non centred unsymmetrical trackpads! >:(
Posted by Redaktion
 - June 03, 2026, 14:24:18
Lenovo has shown off its new 15-inch Nvidia RTX Spark-powered laptop, the Yoga Pro 9n. It has a large trackpad, top-firing speakers, all-day battery life, and all the ports pro creators might need. Official pricing and launch date for the Lenovo Yoga Pro 9n havne't been shared yet.

https://www.notebookcheck.net/Lenovo-reveals-first-Nvidia-RTX-Spark-laptop-15-inch-screen-large-trackpad-SD-card-slot.1313464.0.html