Skip to page content
SAVRN
Search Plan a project

SAVRN Model Hub · Step-5-Preview-NVFP4-Local

Step-5-Preview-NVFP4-Local GPU Requirements

Step-5-Preview-NVFP4-Local needs about 812 GB of GPU memory at 16-bit, 406 GB at 8-bit and 203 GB at 4-bit: its 338.5B parameters plus a 20% working margin. The cheapest setup at 16-bit is 3x MI355X from $7.77 an hour, about $5,672 a month around the clock. None of the 9 accelerators the SAVRN Index prices holds it on one card at 16-bit.

Parameters338.5B
Memory at 16-bit812 GB
Memory at 8-bit406 GB
Memory at 4-bit203 GB
Context1,048,576

Every Accelerator, Every Precision

How many cards of each accelerator the SAVRN Index prices it takes to hold Step-5-Preview-NVFP4-Local, and what that many cards cost an hour at the lowest listed on-demand price. Memory needed: 812 GB at 16-bit, 406 GB at 8-bit, 203 GB at 4-bit.

AcceleratorMemory per cardLowest price per card16-bit8-bit4-bit
H100
Voltage Park
80 GB $1.99 More than 8 6 cards
$11.94/hr
3 cards
$5.97/hr
H200
GMI Cloud
141 GB $2.60 6 cards
$15.60/hr
3 cards
$7.80/hr
2 cards
$5.20/hr
B200
Vultr
180 GB $3.50 5 cards
$17.50/hr
3 cards
$10.50/hr
2 cards
$7.00/hr
GB200 NVL72
GMI Cloud
186 GB $8.00 5 cards
$40.00/hr
3 cards
$24.00/hr
2 cards
$16.00/hr
MI300X
Vultr
192 GB $1.85 5 cards
$9.25/hr
3 cards
$5.55/hr
2 cards
$3.70/hr
MI325X
Vultr
256 GB $2.00 4 cards
$8.00/hr
2 cards
$4.00/hr
1 card
$2.00/hr
B300
Massed Compute
268 GB $6.60 4 cards
$26.40/hr
2 cards
$13.20/hr
1 card
$6.60/hr
GB300 NVL72
Verda
279 GB $11.17 3 cards
$33.51/hr
2 cards
$22.34/hr
1 card
$11.17/hr
MI355X
Vultr
288 GB $2.59 3 cards
$7.77/hr
2 cards
$5.18/hr
1 card
$2.59/hr

Running It Around the Clock

PrecisionCheapest setupPer hourPer month (730 hours)
16-bit3x MI355X (Vultr) $7.77$5,672
8-bit2x MI325X (Vultr) $4.00$2,920
4-bit1x MI325X (Vultr) $2.00$1,460

One copy of the model on rented cards, busy or idle. Serving more users at once takes more copies or more memory for their contexts.

Questions

How much VRAM does Step-5-Preview-NVFP4-Local need?

About 812 GB at 16-bit; about 406 GB at 8-bit; about 203 GB at 4-bit: the weights plus 20% for the runtime and a short context. A long context needs more.

What is the cheapest GPU setup to run Step-5-Preview-NVFP4-Local?

At 16-bit, 3x MI355X from $7.77 an hour, at the lowest on-demand price the SAVRN Index lists.

Can Step-5-Preview-NVFP4-Local run on a single H100?

No. An H100 has 80 GB, and Step-5-Preview-NVFP4-Local needs 203 GB even at 4-bit, so it takes more than one card.

Memory is the weights at that precision plus 20% for the runtime and a short context. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 20, 2026. Setups beyond eight cards, one server, are not listed.