Run a model

A model's size, counted in billions of adjustable numbers it learned during training. 'Qwen3-235B' means 235 billion parameters. More parameters usually means a smarter model — and, crucially, more GPU memory needed to run it. The 'B' is your first clue to how big a machine you'll need.

When it matters: It's the starting point for every buying decision: parameters tell you the memory, memory tells you the hardware.

235B parameters × 1 GB = 235 GB, plus 20% working room (47 GB) = 282 GB of GPU memory required. The single most important number. A model's 'weights' — its entire brain — must fit inside the GPU's own memory all at once, or the model simply will not run. If a model needs 282 GB and your cards add up to 240 GB, it doesn't run slowly, it doesn't run at all. This is the number that decides what you can buy.

When it matters: Always. It's the first thing to check for any model. Bigger model = more GPU memory, no exceptions.

This build gives 288 GB — enough, with 6 GB of headroom to spare.

When it matters: Always. It's the first thing to check for any model. Bigger model = more GPU memory, no exceptions.

Big models don't fit on one card, so you link several together and their memory adds up. Four 96 GB cards give you 384 GB of usable pool. More GPUs also means more speed for more users at once.

When it matters: When one card isn't big enough for the model, or when you need to serve many users at the same time.

1,800 watts — about 2 average homes' worth of power, running around the clock.

When it matters: Every day you own it. Two builds can cost the same to buy but very different amounts to run.

1,800 watts — about 2 average homes' worth of power, running around the clock.

When it matters: When you're deciding whether your building (and budget) can even feed the machine.

The power bill, turned into money. We assume the machine runs around the clock at an average US electricity price of about $0.17 per kilowatt-hour. Formula: kilowatts × 24 hours × 30 days × $0.17. Your local rate will differ, but this makes 'watts' feel like what it really is: a monthly bill.

When it matters: When comparing builds. A cheap-to-buy machine that burns twice the power can cost more within a couple of years.

Running 24/7, this build uses about 0.5 full electric-car batteries (~90 kWh each) of energy every day.

When it matters: When you want a gut sense of energy over time, not just an instant number.

What you pay once, up front, to own the hardware. It does not include the electricity to run it, the cooling, or the room to put it in — those are ongoing.

When it matters: At purchase. But always read it next to the power draw: cheap-to-buy can be expensive-to-run.

A model's size, counted in billions of adjustable numbers it learned during training. 'Qwen3-235B' means 235 billion parameters. More parameters usually means a smarter model — and, crucially, more GPU memory needed to run it. The 'B' is your first clue to how big a machine you'll need.

When it matters: It's the starting point for every buying decision: parameters tell you the memory, memory tells you the hardware.

671B parameters × 1 GB = 671 GB, plus 20% working room (134 GB) = 805 GB of GPU memory required. The single most important number. A model's 'weights' — its entire brain — must fit inside the GPU's own memory all at once, or the model simply will not run. If a model needs 282 GB and your cards add up to 240 GB, it doesn't run slowly, it doesn't run at all. This is the number that decides what you can buy.

When it matters: Always. It's the first thing to check for any model. Bigger model = more GPU memory, no exceptions.

This build gives 846 GB — enough, with 41 GB of headroom to spare.

When it matters: Always. It's the first thing to check for any model. Bigger model = more GPU memory, no exceptions.

Big models don't fit on one card, so you link several together and their memory adds up. Four 96 GB cards give you 384 GB of usable pool. More GPUs also means more speed for more users at once.

When it matters: When one card isn't big enough for the model, or when you need to serve many users at the same time.

4,200 watts — about 4 average homes' worth of power, running around the clock.

When it matters: Every day you own it. Two builds can cost the same to buy but very different amounts to run.

4,200 watts — about 4 average homes' worth of power, running around the clock.

When it matters: When you're deciding whether your building (and budget) can even feed the machine.

The power bill, turned into money. We assume the machine runs around the clock at an average US electricity price of about $0.17 per kilowatt-hour. Formula: kilowatts × 24 hours × 30 days × $0.17. Your local rate will differ, but this makes 'watts' feel like what it really is: a monthly bill.

When it matters: When comparing builds. A cheap-to-buy machine that burns twice the power can cost more within a couple of years.

Running 24/7, this build uses about 1.1 full electric-car batteries (~90 kWh each) of energy every day.

When it matters: When you want a gut sense of energy over time, not just an instant number.

What you pay once, up front, to own the hardware. It does not include the electricity to run it, the cooling, or the room to put it in — those are ongoing.

When it matters: At purchase. But always read it next to the power draw: cheap-to-buy can be expensive-to-run.

A model's size, counted in billions of adjustable numbers it learned during training. 'Qwen3-235B' means 235 billion parameters. More parameters usually means a smarter model — and, crucially, more GPU memory needed to run it. The 'B' is your first clue to how big a machine you'll need.

When it matters: It's the starting point for every buying decision: parameters tell you the memory, memory tells you the hardware.

1000B parameters × 1 GB = 1000 GB, plus 20% working room (200 GB) = 1200 GB of GPU memory required. The single most important number. A model's 'weights' — its entire brain — must fit inside the GPU's own memory all at once, or the model simply will not run. If a model needs 282 GB and your cards add up to 240 GB, it doesn't run slowly, it doesn't run at all. This is the number that decides what you can buy.

When it matters: Always. It's the first thing to check for any model. Bigger model = more GPU memory, no exceptions.

This build gives 1,269 GB — enough, with 69 GB of headroom to spare.

When it matters: Always. It's the first thing to check for any model. Bigger model = more GPU memory, no exceptions.

Big models don't fit on one card, so you link several together and their memory adds up. Four 96 GB cards give you 384 GB of usable pool. More GPUs also means more speed for more users at once.

When it matters: When one card isn't big enough for the model, or when you need to serve many users at the same time.

6,300 watts — about 5 average homes' worth of power, running around the clock.

When it matters: Every day you own it. Two builds can cost the same to buy but very different amounts to run.

6,300 watts — about 5 average homes' worth of power, running around the clock.

When it matters: When you're deciding whether your building (and budget) can even feed the machine.

The power bill, turned into money. We assume the machine runs around the clock at an average US electricity price of about $0.17 per kilowatt-hour. Formula: kilowatts × 24 hours × 30 days × $0.17. Your local rate will differ, but this makes 'watts' feel like what it really is: a monthly bill.

When it matters: When comparing builds. A cheap-to-buy machine that burns twice the power can cost more within a couple of years.

Running 24/7, this build uses about 1.7 full electric-car batteries (~90 kWh each) of energy every day.

When it matters: When you want a gut sense of energy over time, not just an instant number.

What you pay once, up front, to own the hardware. It does not include the electricity to run it, the cooling, or the room to put it in — those are ongoing.

When it matters: At purchase. But always read it next to the power draw: cheap-to-buy can be expensive-to-run.