正在加载视频...

视频加载失败

返回首页

TWO BOXES THE SIZE OF A MAC MINI JUST RAN A 235 BILLION PARAMETER MODEL ON A DESK It is two NVIDIA DGX Spark units linked by a single cable. A year ago a model this size meant renting a GPU cluster by the hour. Now it sits next...

slash1sol's profile picture

slash1s

11,537 subscribers

101,317 次观看 • 2 个月前 •via X (Twitter)

科学技术
LIVE

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

This Chinese developer linked two $2,999 NVIDIA DGX Sparks into one box and runs the full Qwen3-235B at home, after dropping his $1,999-a-month cloud bill to zero. He wired 2 small boxes into a single computer, split a giant 235-billion-parameter model in half between them, and serves it across his own network at about 10 tokens a second, with no internet, no cloud, right there on the desk. No data center, no thousand-dollar graphics cards, no monthly cloud bill. Just him, 2 gold boxes the size of a sandwich, one cable between them, and 1 power strip. And here is the whole payoff. He used to pay the cloud $1,999 a month for the same model, and the meter ticked on every request. Now he paid $5,998 once for 2 boxes, they covered their cost in 3 months, and after that he sends as many requests as he wants for free, only electricity. The two Sparks talk over one fast cable, each holds 128GB of memory, and together they carry the whole model, about 73GB loaded per box, with the chip inside pinned near the limit at 96%. Both boxes work as one and keep trading data over the cable, with no cloud in the loop and no single word leaking out. The ready model sits on one local address, and any app on his network calls it as easily as ChatGPT. And here is how he described, in plain words, what this pair of boxes does: "this is a pair of boxes that holds the huge Qwen3-235B model and serves it to one network. the model is split in half, and each box owns its half. parts: // Box 1 (holds the first half of the model and starts the answer fast, the first word appears in under a second) // Box 2 (holds the second half and writes out the rest, about 10 tokens a second) // Cable (connects the 2 boxes and moves data between them on every step, with no lag) // Address (one local address where any app sends its request, like to a cloud model) // Test (a script that runs big prompts through and measures speed and delays) // Monitor (checks temperature, power draw, and load on both boxes every 2 seconds). the model never goes to the cloud. he only steps in when a box runs hotter than 80 degrees or the cable between them starts dropping data." So the system knows exactly what it is, what it is for, and where its limits are. It knows it has to hold the whole huge model across 2 boxes on its own. It knows it has to answer every request locally, with no meter, no limits, and no internet. It knows the human is only needed when a box overheats or the link between them stalls. → The setup runs around the clock on 2 boxes, each pulling under 60 watts → However many requests he sends, the monthly bill is $0, only electricity → The first box starts the answer in under a second → The second writes text at about 10 tokens a second → One request at a time: 838 tokens in 85 seconds, first word in 0.8s → Two requests at once: 697 tokens in 108 seconds, first word in 0.7s → Both boxes sit at 96% load and warm up to 76-78 degrees And only when a chip in a box runs hotter than 80 degrees or the cable between the 2 Sparks drops data does the system call the owner. And when he himself is out on a run or in a coffee shop, he still reaches his own model at home from his phone: sends a big prompt to the local Qwen3-235B, gets the full answer back in under a minute and a half, with no token meter ticking and no limit to hit. Here is what the test shows on his screen during one of the night runs: "one request at a time: 838 tokens in 84.9 seconds, first word in 0.8s, then 0.1s per token." "two requests at once: 697 tokens in 107.6 seconds, first word in 0.7s, then 0.15s per token." "Box 1: chip at 96% load, 76 degrees, 56 watts, 73GB used in memory." "Box 2: chip at 96% load, 78 degrees, 56 watts, the Qwen3-235B model fully loaded." And while everyone around is paying for AI by the month and bumping into limits, his top-tier model just sits on the desk and works as much as he wants: his own little power plant instead of a forever meter. He has no server rack of his own and no cloud account behind it. Just 2 DGX Spark boxes on a desk, one model split in half between them, one local address, and a folder of prompts next to it. Out of everything I have seen this year, this is the cleanest way to stop paying for AI: $5,998 of hardware on the desk once, $0 a month to the cloud, unlimited forever, and between them 2 gold boxes, 1 cable, and the full Qwen3-235B answering at home with no internet.
1:16

This Chinese developer linked two $2,999 NVIDIA DGX Sparks into one box and runs the full Qwen3-235B at home, after dropping his $1,999-a-month cloud bill to zero. He wired 2 small boxes into a single computer, split a giant 235-billion-parameter model in half between them, and serves it across his own network at about 10 tokens a second, with no internet, no cloud, right there on the desk. No data center, no thousand-dollar graphics cards, no monthly cloud bill. Just him, 2 gold boxes the size of a sandwich, one cable between them, and 1 power strip. And here is the whole payoff. He used to pay the cloud $1,999 a month for the same model, and the meter ticked on every request. Now he paid $5,998 once for 2 boxes, they covered their cost in 3 months, and after that he sends as many requests as he wants for free, only electricity. The two Sparks talk over one fast cable, each holds 128GB of memory, and together they carry the whole model, about 73GB loaded per box, with the chip inside pinned near the limit at 96%. Both boxes work as one and keep trading data over the cable, with no cloud in the loop and no single word leaking out. The ready model sits on one local address, and any app on his network calls it as easily as ChatGPT. And here is how he described, in plain words, what this pair of boxes does: "this is a pair of boxes that holds the huge Qwen3-235B model and serves it to one network. the model is split in half, and each box owns its half. parts: // Box 1 (holds the first half of the model and starts the answer fast, the first word appears in under a second) // Box 2 (holds the second half and writes out the rest, about 10 tokens a second) // Cable (connects the 2 boxes and moves data between them on every step, with no lag) // Address (one local address where any app sends its request, like to a cloud model) // Test (a script that runs big prompts through and measures speed and delays) // Monitor (checks temperature, power draw, and load on both boxes every 2 seconds). the model never goes to the cloud. he only steps in when a box runs hotter than 80 degrees or the cable between them starts dropping data." So the system knows exactly what it is, what it is for, and where its limits are. It knows it has to hold the whole huge model across 2 boxes on its own. It knows it has to answer every request locally, with no meter, no limits, and no internet. It knows the human is only needed when a box overheats or the link between them stalls. → The setup runs around the clock on 2 boxes, each pulling under 60 watts → However many requests he sends, the monthly bill is $0, only electricity → The first box starts the answer in under a second → The second writes text at about 10 tokens a second → One request at a time: 838 tokens in 85 seconds, first word in 0.8s → Two requests at once: 697 tokens in 108 seconds, first word in 0.7s → Both boxes sit at 96% load and warm up to 76-78 degrees And only when a chip in a box runs hotter than 80 degrees or the cable between the 2 Sparks drops data does the system call the owner. And when he himself is out on a run or in a coffee shop, he still reaches his own model at home from his phone: sends a big prompt to the local Qwen3-235B, gets the full answer back in under a minute and a half, with no token meter ticking and no limit to hit. Here is what the test shows on his screen during one of the night runs: "one request at a time: 838 tokens in 84.9 seconds, first word in 0.8s, then 0.1s per token." "two requests at once: 697 tokens in 107.6 seconds, first word in 0.7s, then 0.15s per token." "Box 1: chip at 96% load, 76 degrees, 56 watts, 73GB used in memory." "Box 2: chip at 96% load, 78 degrees, 56 watts, the Qwen3-235B model fully loaded." And while everyone around is paying for AI by the month and bumping into limits, his top-tier model just sits on the desk and works as much as he wants: his own little power plant instead of a forever meter. He has no server rack of his own and no cloud account behind it. Just 2 DGX Spark boxes on a desk, one model split in half between them, one local address, and a folder of prompts next to it. Out of everything I have seen this year, this is the cleanest way to stop paying for AI: $5,998 of hardware on the desk once, $0 a month to the cloud, unlimited forever, and between them 2 gold boxes, 1 cable, and the full Qwen3-235B answering at home with no internet.

Blaze

93,871 次观看 • 2 个月前

This guy plugged a DGX Spark (the $3K Nvidia box) and a Mac Mini M4 together to run AI and what happened next surprised everyone > the Nvidia box handles the hard part - processing your prompt in milliseconds > Mac Mini M4 handles the fast part - generating the response at memory bandwidth speeds nothing else can match > together they hit 84 tokens per second on Llama - 6x faster than the Spark alone > running compute agents locally on this setup means your data never leaves your hardware > two boxes. two different architectures. one AI system that Deepseek runs at data center scale he ran it on his desk save this. the way we build local AI is about to change
18:41

This guy plugged a DGX Spark (the $3K Nvidia box) and a Mac Mini M4 together to run AI and what happened next surprised everyone > the Nvidia box handles the hard part - processing your prompt in milliseconds > Mac Mini M4 handles the fast part - generating the response at memory bandwidth speeds nothing else can match > together they hit 84 tokens per second on Llama - 6x faster than the Spark alone > running compute agents locally on this setup means your data never leaves your hardware > two boxes. two different architectures. one AI system that Deepseek runs at data center scale he ran it on his desk save this. the way we build local AI is about to change

Mr. Buzzoni

153,768 次观看 • 2 个月前

i'm running a 397 billion parameter model on a amd ai max box that sits on my desk and pulls less power than a gaming laptop. the model is Nex-N2-Pro, 397B-A17B, the open weight release people are putting next to gpt-5.5 on coding. i have it quantized to IQ1_M, 1.75 bits per weight, 90gb of weights loaded into the 128gb of unified memory on amd's strix halo igpu. watch the gpu in this recording. it spikes, it sustains, it does not fall over. that is the part the spec sheets never show you, not just that a 400b model loads, but that an integrated graphics chip holds the load and generates token after token, stable, no crash, no thermal cliff. and it is not a slideshow. roughly 18 tokens a second, faster than you can read. a frontier scale model producing usable output, fully local. no datacenter, no rented h100s, no api key, no permission. three years ago a model this size meant a server room and a budget to match. tonight it is a quiet box on my desk. this is the accessible tier almost nobody benchmarks honestly, and it is further along than the timeline thinks. the full breakdown is coming, rocm vs vulkan on this chip, and this little amd box head to head against the nvidia equivalent. stay tuned.
2:18

i'm running a 397 billion parameter model on a amd ai max box that sits on my desk and pulls less power than a gaming laptop. the model is Nex-N2-Pro, 397B-A17B, the open weight release people are putting next to gpt-5.5 on coding. i have it quantized to IQ1_M, 1.75 bits per weight, 90gb of weights loaded into the 128gb of unified memory on amd's strix halo igpu. watch the gpu in this recording. it spikes, it sustains, it does not fall over. that is the part the spec sheets never show you, not just that a 400b model loads, but that an integrated graphics chip holds the load and generates token after token, stable, no crash, no thermal cliff. and it is not a slideshow. roughly 18 tokens a second, faster than you can read. a frontier scale model producing usable output, fully local. no datacenter, no rented h100s, no api key, no permission. three years ago a model this size meant a server room and a budget to match. tonight it is a quiet box on my desk. this is the accessible tier almost nobody benchmarks honestly, and it is further along than the timeline thinks. the full breakdown is coming, rocm vs vulkan on this chip, and this little amd box head to head against the nvidia equivalent. stay tuned.

Sudo su

32,163 次观看 • 2 个月前

APPLE RESEARCH SCIENTIST JUST SHOWED HOW 4 MAC STUDIOS RUN A TRILLION PARAMETER MODEL LOCALLY ZERO COSTS 13:18 she shows the main thing - connect 4 Mac Studios and you get 1TB of shared memory - exactly enough to run a trillion parameter model right on your desk Apple's library - and four machines start working as one cluster tensor parallelism: every machine holds part of every layer - all process the same token simultaneously - speed increases 3x compared to a single device fine-tuning: one Mac Studio processes 180 tokens per second four together process 600 and not a single byte of data leaves the room one command in the terminal - and a trillion parameter model answers from your desk and runs 24/7 data centers took years to build to run models like this - Apple did it with four Thunderbolt cables
21:59

APPLE RESEARCH SCIENTIST JUST SHOWED HOW 4 MAC STUDIOS RUN A TRILLION PARAMETER MODEL LOCALLY ZERO COSTS 13:18 she shows the main thing - connect 4 Mac Studios and you get 1TB of shared memory - exactly enough to run a trillion parameter model right on your desk Apple's library - and four machines start working as one cluster tensor parallelism: every machine holds part of every layer - all process the same token simultaneously - speed increases 3x compared to a single device fine-tuning: one Mac Studio processes 180 tokens per second four together process 600 and not a single byte of data leaves the room one command in the terminal - and a trillion parameter model answers from your desk and runs 24/7 data centers took years to build to run models like this - Apple did it with four Thunderbolt cables

Noisy

48,615 次观看 • 2 个月前

CHINESE AI TEACHER STACKED 4 NVIDIA DGX SPARK CHIPS INTO A CLUSTER - AND NOW RUNS A BUSINESS ON IT FROM HIS DESK 4 chips at $2,999 each - one time investment - and now he has a cluster that handles client inference requests around the clock while he does something else he's running a Qwen 32 35B model with 4 concurrent requests at the same time and the speed barely drops compared to a single request what used to require a $10,000+/month cloud setup now sits on his desk and costs him $40-60/month in electricity 4 chips at $2,999 each - one time investment - and now he has a cluster that handles client inference requests around the clock while he does something else the people still paying per token are financing someone else's infrastructure
1:42

CHINESE AI TEACHER STACKED 4 NVIDIA DGX SPARK CHIPS INTO A CLUSTER - AND NOW RUNS A BUSINESS ON IT FROM HIS DESK 4 chips at $2,999 each - one time investment - and now he has a cluster that handles client inference requests around the clock while he does something else he's running a Qwen 32 35B model with 4 concurrent requests at the same time and the speed barely drops compared to a single request what used to require a $10,000+/month cloud setup now sits on his desk and costs him $40-60/month in electricity 4 chips at $2,999 each - one time investment - and now he has a cluster that handles client inference requests around the clock while he does something else the people still paying per token are financing someone else's infrastructure

Sprytix

51,010 次观看 • 2 个月前

THIS AI DEVELOPER RUNS 120B PARAMETER MODELS LOCALLY AND HIS MACBOOK PRO CAN’T EVEN OPEN THEM he bought a DGX Spark, gave it the model, went back to his desk by end of day the inference was done, tested, committed his team is still waiting for the cloud GPU queue to free up the only thing between them is 128GB of unified memory in a box smaller than your lunch one Spark is a datacenter node, two Sparks is your own cluster bookmark this and send it to whoever is still paying for cloud GPUs
2:54

THIS AI DEVELOPER RUNS 120B PARAMETER MODELS LOCALLY AND HIS MACBOOK PRO CAN’T EVEN OPEN THEM he bought a DGX Spark, gave it the model, went back to his desk by end of day the inference was done, tested, committed his team is still waiting for the cloud GPU queue to free up the only thing between them is 128GB of unified memory in a box smaller than your lunch one Spark is a datacenter node, two Sparks is your own cluster bookmark this and send it to whoever is still paying for cloud GPUs

leopardracer

85,726 次观看 • 2 个月前

THIS DEVELOPER CONNECTED 8 NVIDIA DGX SPARKS INTO ONE CLUSTER - AND RAN AN 800GB MODEL THAT MADE HIM 10X MORE PRODUCTIVE 21:47 he says it straight - "this is a terabyte of VRAM - we ran Quen 3.5, 800GB on disk, a model that doesn't even fit on a single Mac Studio - 24 tokens per second - I'd say that's a win" 8 Sparks connected through a $1,300 switch via RDMA over Ethernet - each node adding 128GB of memory into one unified pool of 1TB started with one Spark at 3 tokens per second - every added node doubled the speed - and eight together deliver 24 tokens on a model that physically cannot run anywhere else Kimi K2 at 600GB loaded in 15 minutes, 115GB per node, 13 tokens per second - a model that simply cannot run on anything smaller Claude helped configure the entire cluster - SSH mesh across all 8 machines, network config, jumbo frames, QSFP port speeds - all from one terminal most people rent cloud compute for models this size at $2,000+/month - he built the cluster once and now every token costs 20x less
24:17

THIS DEVELOPER CONNECTED 8 NVIDIA DGX SPARKS INTO ONE CLUSTER - AND RAN AN 800GB MODEL THAT MADE HIM 10X MORE PRODUCTIVE 21:47 he says it straight - "this is a terabyte of VRAM - we ran Quen 3.5, 800GB on disk, a model that doesn't even fit on a single Mac Studio - 24 tokens per second - I'd say that's a win" 8 Sparks connected through a $1,300 switch via RDMA over Ethernet - each node adding 128GB of memory into one unified pool of 1TB started with one Spark at 3 tokens per second - every added node doubled the speed - and eight together deliver 24 tokens on a model that physically cannot run anywhere else Kimi K2 at 600GB loaded in 15 minutes, 115GB per node, 13 tokens per second - a model that simply cannot run on anything smaller Claude helped configure the entire cluster - SSH mesh across all 8 machines, network config, jumbo frames, QSFP port speeds - all from one terminal most people rent cloud compute for models this size at $2,000+/month - he built the cluster once and now every token costs 20x less

Noisy

185,632 次观看 • 2 个月前

THIS DEVELOPER JUST RAN A TRILLION PARAMETER MODEL ON 4 MAC STUDIOS - 10X FASTER AND 5X CHEAPER THAN CLOUD CODE 19:00 he says it out loud. "we just ran a trillion parameter model. 30 something tokens per second. wow." RDMA over Thunderbolt made the cluster 10x faster than before. tensor parallelism splits the model between machines in parallel instead of sequentially DeepSeek V3 full size without quantization. 183GB per machine. still 25 tokens per second. faster than most people can read Kimi K2 with trillion parameters and 256,000 context window. loaded, responded and adapted memory usage dynamically based on prompt size 4 machines consuming 66 watts total. cloud GPU for the same workload costs $1,900/month a cluster that used to cost millions now assembled from Mac Studios in a few hours
22:35

THIS DEVELOPER JUST RAN A TRILLION PARAMETER MODEL ON 4 MAC STUDIOS - 10X FASTER AND 5X CHEAPER THAN CLOUD CODE 19:00 he says it out loud. "we just ran a trillion parameter model. 30 something tokens per second. wow." RDMA over Thunderbolt made the cluster 10x faster than before. tensor parallelism splits the model between machines in parallel instead of sequentially DeepSeek V3 full size without quantization. 183GB per machine. still 25 tokens per second. faster than most people can read Kimi K2 with trillion parameters and 256,000 context window. loaded, responded and adapted memory usage dynamically based on prompt size 4 machines consuming 66 watts total. cloud GPU for the same workload costs $1,900/month a cluster that used to cost millions now assembled from Mac Studios in a few hours

Noisy

79,665 次观看 • 2 个月前

48GB of VRAM rents for $500 a month, and this rig has room to double it. A vertical frame. A bare motherboard. One graphics card, exposed and humming. No rack. No server room. Just a desk. Ryzen 9 9950x. AsRock Taichi X870E. 96GB of DDR5 at 6000MT/s. A Thermalright Royal Pretor 130 Ultra on top. A 1300-watt Super Flower supply feeding all of it. One card. 48GB of VRAM. On the open market, that card alone rents for $0.58 to $0.82 an hour. Run it around the clock, and it clears $500 a month, crunching someone else's AI workloads. The motherboard and the power supply were not picked at random. Both, for one reason: a second GPU. Slide it in, and VRAM doubles to 96GB. So does the money: $1,000 a month, from a machine that fits under a desk. Most people rent that power by the hour. This desk prints it.
0:54

48GB of VRAM rents for $500 a month, and this rig has room to double it. A vertical frame. A bare motherboard. One graphics card, exposed and humming. No rack. No server room. Just a desk. Ryzen 9 9950x. AsRock Taichi X870E. 96GB of DDR5 at 6000MT/s. A Thermalright Royal Pretor 130 Ultra on top. A 1300-watt Super Flower supply feeding all of it. One card. 48GB of VRAM. On the open market, that card alone rents for $0.58 to $0.82 an hour. Run it around the clock, and it clears $500 a month, crunching someone else's AI workloads. The motherboard and the power supply were not picked at random. Both, for one reason: a second GPU. Slide it in, and VRAM doubles to 96GB. So does the money: $1,000 a month, from a machine that fits under a desk. Most people rent that power by the hour. This desk prints it.

AiMind

30,460 次观看 • 2 天前

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.
0:42

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

Akshay 🚀

294,838 次观看 • 2 天前

people burn $1,900 a month on cloud, this $2,999 NVIDIA box runs 5 AI models at once and costs $10 to power 128GB of unified memory, stack two together and you double it, enough to run models a $2,000 graphics card cannot even open in the demo it ran Qwen3 Vision, Nemotron, a voice model, text to speech and speech to text all at once, and only used half the box here is the math that makes people switch: > cloud GPUs were costing $1,900 every single month, billed forever > the box is $2,999 one time, then about $10 of electricity a month > it paid for itself in two months, $22,000 stayed in the business in year one > zero data ever leaves your desk, no rental, no rate limits most people are still renting what this owns outright the full breakdown of the box and the exact math is in the article below
1:04

people burn $1,900 a month on cloud, this $2,999 NVIDIA box runs 5 AI models at once and costs $10 to power 128GB of unified memory, stack two together and you double it, enough to run models a $2,000 graphics card cannot even open in the demo it ran Qwen3 Vision, Nemotron, a voice model, text to speech and speech to text all at once, and only used half the box here is the math that makes people switch: > cloud GPUs were costing $1,900 every single month, billed forever > the box is $2,999 one time, then about $10 of electricity a month > it paid for itself in two months, $22,000 stayed in the business in year one > zero data ever leaves your desk, no rental, no rate limits most people are still renting what this owns outright the full breakdown of the box and the exact math is in the article below

John Doe

48,135 次观看 • 2 个月前

AMD CEO Lisa Su showcased a mini PC that could replace your entire AI stack what's changed: - 128GB of unified memory on a single chip - the CPU and GPU share the same memory space - a 235-billion-parameter model can be loaded entirely locally - no cloud, no subscriptions, no logs an active AI user spends around $420 a month on subscriptions - that's $5,280 a year the EVO-X2 costs $1,800 nstall ollama, pull the model, point claude code at localhost. same interface, nothing leaves the machine, nothing costs per request bookmark this and read the article below
0:59

AMD CEO Lisa Su showcased a mini PC that could replace your entire AI stack what's changed: - 128GB of unified memory on a single chip - the CPU and GPU share the same memory space - a 235-billion-parameter model can be loaded entirely locally - no cloud, no subscriptions, no logs an active AI user spends around $420 a month on subscriptions - that's $5,280 a year the EVO-X2 costs $1,800 nstall ollama, pull the model, point claude code at localhost. same interface, nothing leaves the machine, nothing costs per request bookmark this and read the article below

kepo

15,451 次观看 • 2 个月前

watch and let this sit for a second anon, a 35B AI model, building a complete multi-file game by itself on hermes agent. no cloud or api. not one line of code written by a human. running on a DGX spark in 2026. because not long ago this was science fiction. this is Ornith, a new open agentic coding model, running near lossless on a single DGX Spark, a 128GB box that sits on your desk. i hand it one spec and step back. it reasons about the architecture, creates the project structure, writes every module one by one, and here's the part that got me, it checks its own work as it goes. makes a call, reads the result, decides if it's right, corrects itself before moving on. no steering from me. this is what people keep underestimating about local AI. they hear "local" and think weaker, slower, a toy. watch the clip. a 35B at near full precision, planning and self correcting on a single desktop. the open models got genuinely good at agentic work, not benchmark good, actually builds the thing good. sped 5x so you can watch the whole build start to finish. this is where local AI actually is now. not coming soon. here.
2:29

watch and let this sit for a second anon, a 35B AI model, building a complete multi-file game by itself on hermes agent. no cloud or api. not one line of code written by a human. running on a DGX spark in 2026. because not long ago this was science fiction. this is Ornith, a new open agentic coding model, running near lossless on a single DGX Spark, a 128GB box that sits on your desk. i hand it one spec and step back. it reasons about the architecture, creates the project structure, writes every module one by one, and here's the part that got me, it checks its own work as it goes. makes a call, reads the result, decides if it's right, corrects itself before moving on. no steering from me. this is what people keep underestimating about local AI. they hear "local" and think weaker, slower, a toy. watch the clip. a 35B at near full precision, planning and self correcting on a single desktop. the open models got genuinely good at agentic work, not benchmark good, actually builds the thing good. sped 5x so you can watch the whole build start to finish. this is where local AI actually is now. not coming soon. here.

Sudo su

30,299 次观看 • 1 个月前

AMD CEO LISA SU HELD A MINI PC ON STAGE THAT RUNS A 235B MODEL AND REPLACES YOUR $440/MONTH AI STACK amd's ryzen ai max+ 395 is the first x86 chip that runs a 200 billion parameter model on one piece of silicon. cpu and gpu share 128gb of unified memory, no separate graphics card needed the gmktec evo-x2 runs qwen3 235b fully, deepseek v3 comfortably and llama 3.3 70b with headroom. on linux you get 110gb of usable vram out of 128gb amd claimed the chip beat an nvidia rtx 5080 by more than 3x on deepseek r1 inference. a lunchbox sized pc outrunning a $1,000 discrete gpu on a real ai workload a heavy ai user pays $200 for claude code max, $200 for chatgpt pro, $20 for cursor and $20 for gemini. that's $5,280 a year and the box pays itself off in 9 to 10 months install ollama, pull the model, point claude code at localhost. same interface, nothing leaves the machine, nothing costs per request bookmark this and read the article below
0:59

AMD CEO LISA SU HELD A MINI PC ON STAGE THAT RUNS A 235B MODEL AND REPLACES YOUR $440/MONTH AI STACK amd's ryzen ai max+ 395 is the first x86 chip that runs a 200 billion parameter model on one piece of silicon. cpu and gpu share 128gb of unified memory, no separate graphics card needed the gmktec evo-x2 runs qwen3 235b fully, deepseek v3 comfortably and llama 3.3 70b with headroom. on linux you get 110gb of usable vram out of 128gb amd claimed the chip beat an nvidia rtx 5080 by more than 3x on deepseek r1 inference. a lunchbox sized pc outrunning a $1,000 discrete gpu on a real ai workload a heavy ai user pays $200 for claude code max, $200 for chatgpt pro, $20 for cursor and $20 for gemini. that's $5,280 a year and the box pays itself off in 9 to 10 months install ollama, pull the model, point claude code at localhost. same interface, nothing leaves the machine, nothing costs per request bookmark this and read the article below

starmex

980,353 次观看 • 2 个月前

Your cloud agent runs $10–20 a day. This box runs OpenAI's 120B model for $0. Same workload. It just sits in a closet on AMD silicon. A Minisforum mini PC, Strix Halo, 128GB unified memory. The model lives entirely on-device. 56 tokens a second. No API key. Nothing leaves the network. Cloud only handles the exceptions now. Claude or OpenAI by routing, not by default. Everyone calls this "free." Wrong word. It's not free. It's owned. The bill just stops existing. The cloud was never the only option. It was the default nobody bothered to question.
0:21

Your cloud agent runs $10–20 a day. This box runs OpenAI's 120B model for $0. Same workload. It just sits in a closet on AMD silicon. A Minisforum mini PC, Strix Halo, 128GB unified memory. The model lives entirely on-device. 56 tokens a second. No API key. Nothing leaves the network. Cloud only handles the exceptions now. Claude or OpenAI by routing, not by default. Everyone calls this "free." Wrong word. It's not free. It's owned. The bill just stops existing. The cloud was never the only option. It was the default nobody bothered to question.

RetroChainer

13,925 次观看 • 2 个月前

8 DGX Sparks on a desk just ran a frontier model at 400K context. A friend who owns the inference budget at a Berlin AI startup, $52K a month in API bills, sent me this at 2 a.m. with one line: "we're buying the hardware." It starts as a disaster. A mountain of cables burying $32K of NVIDIA silicon, nothing talking to anything. 54 minutes later the whole stack answers as 1 machine. 1TB of unified memory. GLM-5.2 loaded across all 8 nodes. The part that made him rewatch twice sits around the 25-minute mark, where 1 networking fix turns 8 boxes into a cluster. Most people quit exactly there. He paused it 4 times. Screenshotted the interface setup. Forwarded it to his CTO without a caption. The cloud bill was never the moat. The cables were.
54:24

8 DGX Sparks on a desk just ran a frontier model at 400K context. A friend who owns the inference budget at a Berlin AI startup, $52K a month in API bills, sent me this at 2 a.m. with one line: "we're buying the hardware." It starts as a disaster. A mountain of cables burying $32K of NVIDIA silicon, nothing talking to anything. 54 minutes later the whole stack answers as 1 machine. 1TB of unified memory. GLM-5.2 loaded across all 8 nodes. The part that made him rewatch twice sits around the 25-minute mark, where 1 networking fix turns 8 boxes into a cluster. Most people quit exactly there. He paused it 4 times. Screenshotted the interface setup. Forwarded it to his CTO without a caption. The cloud bill was never the moat. The cables were.

the lich

56,893 次观看 • 1 个月前

$5,280 a year on AI subscriptions. Claude Max, ChatGPT Pro, Gemini, Cursor. $440 every month going to the cloud. a mini PC for $1,700 does the same thing on your desk. GMKtec EVO-X2 - a box the size of a hardcover book. inside it the chip that Lisa Su personally brought on stage at CES and signed with her name. 128GB of memory in one pool. runs models that cloud subscriptions behind $200 paywalls throttle during peak hours. here there are no limits - your hardware, your rules. NVIDIA's RTX 5080 at $1,000+ lost to this box by 3x on a real AI workload. a discrete GPU beaten by a mini PC the size of a lunchbox. electricity - $9 a month. pays for itself in 10 months. after that $440 every month stays in your pocket. the cloud made sense when nothing on a desk could compete. now it can.
0:32

$5,280 a year on AI subscriptions. Claude Max, ChatGPT Pro, Gemini, Cursor. $440 every month going to the cloud. a mini PC for $1,700 does the same thing on your desk. GMKtec EVO-X2 - a box the size of a hardcover book. inside it the chip that Lisa Su personally brought on stage at CES and signed with her name. 128GB of memory in one pool. runs models that cloud subscriptions behind $200 paywalls throttle during peak hours. here there are no limits - your hardware, your rules. NVIDIA's RTX 5080 at $1,000+ lost to this box by 3x on a real AI workload. a discrete GPU beaten by a mini PC the size of a lunchbox. electricity - $9 a month. pays for itself in 10 months. after that $440 every month stays in your pocket. the cloud made sense when nothing on a desk could compete. now it can.

Ventry

27,360 次观看 • 2 个月前

i built a full game on a single GPU with a 3B model and this is the worst local AI will ever be. this was supposed to be a benchmark test. load the model, measure tokens per second, write it up, move on. instead i spent 20 minutes playing Octopus Invaders because the game is genuinely fun and i couldn't stop. a model with 3B active parameters built this from a single prompt. it debugged its own collision system when bullets were phasing through enemies. read the error, found the fix, kept building. this is not a frontier API. this is a quantized open source model running on hardware you can buy used for $800-$1200. no cloud. no subscription. no API costs. just a mass produced consumer GPU doing things that would have been absurd 12 months ago. and here's the part that should keep you up at night: every month the models get smaller and smarter. the quants get tighter. the context windows get longer. the tooling gets cleaner. what 3B active parameters does today on 24gb, a 1B model will do on 8gb within a year. you are looking at the floor. not the ceiling.
0:27

i built a full game on a single GPU with a 3B model and this is the worst local AI will ever be. this was supposed to be a benchmark test. load the model, measure tokens per second, write it up, move on. instead i spent 20 minutes playing Octopus Invaders because the game is genuinely fun and i couldn't stop. a model with 3B active parameters built this from a single prompt. it debugged its own collision system when bullets were phasing through enemies. read the error, found the fix, kept building. this is not a frontier API. this is a quantized open source model running on hardware you can buy used for $800-$1200. no cloud. no subscription. no API costs. just a mass produced consumer GPU doing things that would have been absurd 12 months ago. and here's the part that should keep you up at night: every month the models get smaller and smarter. the quants get tighter. the context windows get longer. the tooling gets cleaner. what 3B active parameters does today on 24gb, a 1B model will do on 8gb within a year. you are looking at the floor. not the ceiling.

Sudo su

36,251 次观看 • 5 个月前

A 24 BAY DELL POWEREDGE T550 TOWER SERVER FITS IN A HOME OFFICE WITHOUT A RACK, 384TB OF RAW SAS STORAGE HOSTS EVERY OPEN SOURCE LLM EVER RELEASED LOCALLY, THE HOME AI LIBRARY TIER UNDER YOUR MAP 00:25 the reseller spins the chassis around, "it's a 24 bay Dell T550 tower server man, look it's got a boss card for dual NVMe boot SSDs" a refurbished T550 ships at $900-1,100 with single Xeon Silver, 64GB DDR4 ECC and 24 hot swap SAS bays in a tower form factor that fits next to a desk without a server rack 24x 16TB SAS drives at $80 each on eBay equals 384TB raw storage for $1,920, the same array on enterprise Mac Studio Pro tier would cost $35,000 in thunderbolt JBOD enclosures Llama 3.3 70B is 42GB, Qwen3-235B is 110GB, DeepSeek-V3 is 100GB, Mistral Large is 80GB, the entire ollama public model library fits inside 4TB, you can mirror every open weights release for the next decade your map covers compute boxes that swap one model at a time, the T550 is the model warehouse tier that feeds them, a homelab operator wires the tower over 10GbE to a mac mini and the mac never waits 30 seconds to pull a swap the window is open, follow and bookmark before it closes
0:53

A 24 BAY DELL POWEREDGE T550 TOWER SERVER FITS IN A HOME OFFICE WITHOUT A RACK, 384TB OF RAW SAS STORAGE HOSTS EVERY OPEN SOURCE LLM EVER RELEASED LOCALLY, THE HOME AI LIBRARY TIER UNDER YOUR MAP 00:25 the reseller spins the chassis around, "it's a 24 bay Dell T550 tower server man, look it's got a boss card for dual NVMe boot SSDs" a refurbished T550 ships at $900-1,100 with single Xeon Silver, 64GB DDR4 ECC and 24 hot swap SAS bays in a tower form factor that fits next to a desk without a server rack 24x 16TB SAS drives at $80 each on eBay equals 384TB raw storage for $1,920, the same array on enterprise Mac Studio Pro tier would cost $35,000 in thunderbolt JBOD enclosures Llama 3.3 70B is 42GB, Qwen3-235B is 110GB, DeepSeek-V3 is 100GB, Mistral Large is 80GB, the entire ollama public model library fits inside 4TB, you can mirror every open weights release for the next decade your map covers compute boxes that swap one model at a time, the T550 is the model warehouse tier that feeds them, a homelab operator wires the tower over 10GbE to a mac mini and the mac never waits 30 seconds to pull a swap the window is open, follow and bookmark before it closes

Antid

18,526 次观看 • 1 个月前

watch anon! 27 billion parameter model dancing at 42 tokens a second on a used rtx 3060ti, that costs 200 dollars. consuming 159 watts. temp 62 degrees. pushing almost 100% on the gpu. 6.8 of 8 gigs used with a 128k context window loaded. and every one of those spikes is the model reasoning and firing tool calls. the model is bonsai 27b at 1 bit. PrismML took qwen 3.6 27b, and crushed it to 3.9 gigabytes, about a single sign bit per weight. it drops onto an 8gb rtx 3060 ti, an ampere card from 2020, and it doesn't just load, it serves, it holds the full 128k window, it runs a hermes agent for half an hour straight, it hits its own failures and recovers, it never once drops the loop. the absurd part is the power meter. that same 27b, in its normal quant 4, wants a 24gb card and nearly 17 gigs of weights. this is it at 3.9, doing agentic work at 160 watts, on the tier people call budget. the numbers are sitting right there on nvtop, not in a slide, not a claim, the meter reading while it works. so i keep landing on one question. the model is open, apache 2.0. the card is old and cheap. the whole thing runs offline. why is almost nobody actually running this. is it the hardware, the setup, or did nobody tell you a 27b fits on the card already in your closet.
0:26

watch anon! 27 billion parameter model dancing at 42 tokens a second on a used rtx 3060ti, that costs 200 dollars. consuming 159 watts. temp 62 degrees. pushing almost 100% on the gpu. 6.8 of 8 gigs used with a 128k context window loaded. and every one of those spikes is the model reasoning and firing tool calls. the model is bonsai 27b at 1 bit. PrismML took qwen 3.6 27b, and crushed it to 3.9 gigabytes, about a single sign bit per weight. it drops onto an 8gb rtx 3060 ti, an ampere card from 2020, and it doesn't just load, it serves, it holds the full 128k window, it runs a hermes agent for half an hour straight, it hits its own failures and recovers, it never once drops the loop. the absurd part is the power meter. that same 27b, in its normal quant 4, wants a 24gb card and nearly 17 gigs of weights. this is it at 3.9, doing agentic work at 160 watts, on the tier people call budget. the numbers are sitting right there on nvtop, not in a slide, not a claim, the meter reading while it works. so i keep landing on one question. the model is open, apache 2.0. the card is old and cheap. the whole thing runs offline. why is almost nobody actually running this. is it the hardware, the setup, or did nobody tell you a 27b fits on the card already in your closet.

Sudo su

23,391 次观看 • 29 天前