ycui1986

exactly. we need branch predictor for expert weight.

There are a lot of SSD streaming engines these days. But few to actually try some hard features.

There is one that could really improve the speed. Given almost all major models come with MTP head for speculative decoding. The same MTP head could also be used to speculative prefetch the expert weight residing on the SSD. If the expert weight can be preloaded before the GPU actually need them, the speed penalty from VRAM cache miss will be quite reduced.

If the technology demonstrates successful token rate improvement. future models could also come with pretraining heads to preload expert weights, and even make the training be aware of it.

i always thought Ryzen AI Halo, together with DGS Spark, has mismatched compute capacity with memory size. Given 128GB VRAM, people would want to run large models, but the GPU compute is constraint in these types of use case. If the box runs models that don't need high compute, then there is no need of 128GB VRAM.

On the high end side, it is too slow. On the low end size, it waste money on VRAM.

So, dual RTX PRO 6000

I really like the pro version. The pelican is so cute.

32GB RAM on mac also need to host OS, software, and other stuff. There may not even be 24GB VRAM left for the model.

just because a bunch of rockets went up without blowing up, does not mean they are profitable. it cost money to shot rocket, and it is very expensive, reusable or not. most launches are internal launch without external paying customers.

another 60 billion to save a failed AI endeavor.

outputting docx files does not have much to do with model capability. it is about whether tool calling has be configured .

There are also many Chines AI-target GPU/NPU producers. You can get a hold of some boards on taobao.com. They are usable in some way.

No, nVidia and AMD are not the only ones benefiting.

i give it in real ubuntu, no vm, no docker. so long I don't ask it to organize files, it will behave. it has not screw me so far.

qwen3.5 and qwen3.6 are both good at tool calling.

For many LLM load, it seems ROCm is slower than vulkan. What’s the point?

he won't. if anything, openai is falling behind recently. the trend won't change easily. it is like the old time Netscape.

only works if the users are evenly distributed around the globe (which is likely more of less the case). if the user concentrates in on century, the token rate will be terrible.

i hope someone do a 100b 1-bit parameter model. that should fit into most 16GB graphics cards. local AI democratized.

i am guessing, without any proof, that, when one breaker fails the server lose it all, or loose two GPUs, depending on whether one connected to the cpu side failed.

9070XT provide roughly same inference performance at double the power, half the cost, as RTX PRO 4500. So this one is optimized for total BOM cost.

they could had gone with the Max-Q version RTX PRO 6000 and only require 120V circuit. 10% performance hit, but half the power.

fundamentally, looks like they are shipping consumer off-the-shelf hardwares in a custom box.

for all past years, I have been told wayland is the future. but the decade long dragged out rolling out did not made much sense to me. neither did I investigate why. until today, I found out how difficult to force a 1920x1200 resolution over remote desktop. it is plain feature degradation.

people ask why do you need it. I have a 3440x1440 physical monitor on the server, I need to remove login with a 1920x1200 laptop. I want full screen at laptop's native resolution. Windows can do this decade ago.

the reality is no where to get the fuel. hydrogen stations are shutting down not building up.

China tests crewed spacecraft abort and rocket recovery in major lunar milestone

it is bizarre that a notepad app can have remote code execution. how much unnecessary function did MS add to get to this point?

If what Waymo wrote is true, this sounds more like kids fault or guardian’s.

everyone uses cellphone that transmit on the same frequency. they don't seem to cause interference. once enough lidar enters real word use. there will be regulation to make them work with each other.

used NI-GPIB on USB cost $100 on ebay. You don’t need $1000.

very impressive. better than anything on the market either NI or Keysight.

from the picture, the compressor and generator located inside the dome. the dome is filled with CO2. maintenance people have to carry oxygen tank, or they die.

i think it had something to do with CO2 can be made into supercritical state relatively easily, not for nitrogen or other common gases.