What are the benefits of running GLM 5.3 vs Deepseek V4.1 Flash?
The lower hallucination rates? Less tokens on reasoning?
5
3
1,170
Ok fine I won’t sell my sparks @meinknee
Mac Studio 512GB ×1 vs ASUS GX10 128GB ×4
DeepSeek V4.1 Flash, original weights, same bench
Price: $25,799 vs $24,000
Concurrency 1: 53/84 vs 73/122 tok/s (text/code)
Concurrency 16: 137/148 vs 330/847 tok/s (text/code)
128K prefill: 3,325 vs 7,522 tok/s
DGX Spark recipes keep coming and keep getting faster. That's a big plus.
Mac's plus: one person, one session at a time? It's also your dev machine. Code and run the LLM on the same box, while the Sparks still need another computer to work from.
Mac: mlx-serve @ddalcu · GX10: spark-ds41f @christopherowen
1
2
245
Pretty devastated. Been speaking to people at Kubecon that are 20-year industry vets that knew Jensen in the early days. He was a genuine dude, probably still is. Being unfaithful in your marriage is so lame.
That being said a small part of me thinks there is a bit of pile-on going on. Nvidia is pro open weights (mostly due to the nature of them selling chips) but that helps the local crew so much. If he goes away that’s another pivotal voice we won’t have on the side of this closed/open battle.
Holy shit he knows ball
I use C/C++ for all my projects. Debt Badge, Clucks Capacitor, homemade power-wall, sculpting software for my haptic interface.
I suppose with AI assisting, the clunkiness of developing in C disappears and the efficiency shines. How soon until the language doesn’t matter?
4
635
Is it not obvious? Hahah Fable 5.2 dropped. Why wouldn’t they take the time to try to distill it?
It might not be possible. But entire cohorts are definitely trying.
1
2
283
My buddy Owen and I just got 2nd place in the second hackathon I’ve ever been to. 130 teams (800+ people signed up but only 300ish showed up). Small potato stuff but we both won $2,000.
I was a failed web developer. I didn’t know the difference between storage and ram. Basic concepts in CS were so hard for me.
But I’ve never stopped asking questions.
You can just. Do. Things.
If Mac Studio’s had a black/dark gray option, kind of like the Apple Watch Ultra.
That’d be cool.
1
7
355
Claude Code / Opus 5.5 orchestrating OMP / Deepseek V4.1 Flash is the most productive AI coding I’ve done yet.
4
9
658
🤯🤯
@christopherowen has done amazing work with his TP4 GB10 Deepseek V4.1 Flash recipe.
github.com/christopherowen/s…
4
18
1,795
Dflash
DSpark
Dflash2
DSpine
Speculative decoding just keeps getting better and better.
THIS IS HUGE !!
Now you can run Qwen3.8-27B even faster !!
@TencentHunyuan just released DSpine for Qwen3.8-27B and it beats DFlash2 really well.
I ported this to llama.cpp and have created ggufs for it,
find all my drafters in my hf link below.
Since it's not supported in upstream llama.cpp, you can clone my updated fork and give it a spin.
- Baseline: 28.2 tok/s
- DFlash2: 50.6 tok/s.
- DSpine: 94–132 tok/s
DSpine's lead over DFlash2 is about 1.75x either way.
DSpine predicts more tokens per verify step on math and code (7.3–7.6 vs 5.7–5.8), which is where its big wins come from.
github.com/Ajay9o9/llama.cpp…
All information is provided on model card.
huggingface.co/aj9o9/Qwen3.8…
9
687
Being a wife-guy pays off.
Jensen should’ve taken one of the only good plays out of Zuck’s book.
1
5
414