Top
Best
New

Posted by ricardbejarano 10 hours ago

Can I run AI locally?(www.canirun.ai)
619 points | 170 commentspage 3
freediddy 5 hours ago|
i think the perplexity is more important than tokens per second. tokens per second is relatively useless in my opinion. there is nothing worse than getting bad results returned to you very quickly and confidently.

ive been working with quite a few open weight models for the last year and especially for things like images, models from 6 months would return garbage data quickly, but these days qwen 3.5 is incredible, even the 9b model.

sroussey 5 hours ago|
No, getting bad results slowly is much worse. Bad results quickly and you can make adjustments.

But yes, if there is a choice I want quality over speed. At same quality, I definitely want speed.

cafed00d 4 hours ago||
Open with multiple browsers (safari vs chrome) to get more "accurate + glanceable" rankings.

Its using WebGPU as a proxy to estimate system resource. Chrome tends to leverage as much resources (Compute + Memory) as the OS makes available. Safari tends to be more efficient.

Maybe this was obvious to everyone else. But its worth re-iterating for those of us skimmers of HN :)

GrayShade 6 hours ago||
This feels a bit pessimistic. Qwen 3.5 35B-A3B runs at 38 t/s tg with llama.cpp (mmap enabled) on my Radeon 6800 XT.
Aurornis 5 hours ago|
At what quantization and with what size context window?
GrayShade 4 hours ago||
Looks like it's a bit slower today. Running llama.cpp b8192 Vulkan.

$ ./llama-cli unsloth_Qwen3.5-35B-A3B-GGUF_Qwen3.5-35B-A3B-UD-Q4_K_XL.gguf -c 65536 -p "Hello"

[snip 73 lines]

[ Prompt: 86,6 t/s | Generation: 34,8 t/s ]

$ ./llama-cli unsloth_Qwen3.5-35B-A3B-GGUF_Qwen3.5-35B-A3B-UD-Q4_K_XL.gguf -c 262144 -p "Hello"

[snip 128 lines]

[ Prompt: 78,3 t/s | Generation: 30,9 t/s ]

I suspect the ROCm build will be faster, but it doesn't work out of the box for me.

azmenak 3 hours ago||
From my personal testing, running various agentic tasks with a bunch of tool calls on an M4 Max 128GB, I've found that running quantized versions of larger models to produce the best results which this site completely ignores.

Currently, Nemotron 3 Super using Unsloth's UD Q4_K_XL quant is running nearly everything I do locally (replacing Qwen3.5 122b)

rcarmo 4 hours ago||
This is kind of bogus since some of the S and A tier models are pretty useless for reasoning or tool calls and can’t run with any sizable system prompt… it seems to be solely based on tokens per second?
nicklo 2 hours ago||
the animation of the model name text when opening the detail view is so smooth and delightful
am17an 5 hours ago||
You can still run larger MoE models using expert weight off-loading to the CPU for token generation. They are by and large useable, I get ~50 toks/second on a kimi linear 48B (3B active) model on a potato PC + a 3090
vednig 2 hours ago||
Our work at DoShare is a lot of this stuff we've been on it for 2 years
orthoxerox 5 hours ago||
For some reason it doesn't react to changing the RAM amount in the combo box at the top. If I open this on my Ryzen AI Max 395+ with 32 GB of unified memory, it thinks nothing will fit because I've set it up to reserve 512MB of RAM for the GPU.
bityard 5 hours ago|
Yeah, this site is iffy at best. I didn't even see Strix Halo on the list, but I selected 128GB and bumped up the memory bandwidth. It says gpt-oss-120b "barely runs" at ~2 t/s.

In reality, gpt-oss-120b fits great on the machine with plenty of room to spare and easily runs inference north of 50 t/s depending on context.

kuon 3 hours ago|
I have amd 9700 and it is not listed while it is great llm hardware because it has 32Gb for a reasonable price. I tried doing "custom" but it didn't seem to work.

The tool is very nice though.

More comments...