GPUBeacon
Language
NVIDIA

Cloud GPU · verified 2026-09-05

NVIDIA B300

Blackwell288 GB

Blackwell Ultra with a 288 GB tank and a power bill that looks like a small furnace. Rent it for memory-bound frontier work. Do not rent it to look current.

Author Valentin FourierLast modified

VRAM288 GB
From /GPU/hr$7.50
ArchitectureBlackwell
VendorNVIDIA

In brief

What to know before you rent

Catalog listings

B300 Pricing and Availability

Published on-demand rates first. Spot and waitlists are labeled. Confirm the live rate before you provision.

ProvidersRegionBilling typeInterconnect/GPU/hr
VerdaIn stockEU-NorthOn-demandNVLink$7.50Open provider
Packet.aiWaitlistUS-WestOn-demandNVLink$7.90Open provider
KoyebWaitlistEU-WestOn-demandNVLink$8.25Open provider
CoreWeaveWaitlistUS-EastOn-demandInfiniBand NDR$9.40Open provider

Our reading

Who this GPU is for

B300 is what you rent when 192 GB is the wall and the team already lives on Blackwell. I will not recommend it as a first cloud GPU. I will recommend it when the alternative is an ugly multi-node split you cannot operate.

Rent it if Labs training past Hopper envelopes, inference with gigantic KV caches, people who already survived B200 software.

Skip it if Everyone still shopping H100. Startups with a 4090-shaped job. Anyone who cannot name why 288 GB beats 141 GB.

How we compare hours

Datasheet

Technical specifications

NVIDIA B300 · Per GPU · Blackwell Ultra

Compute · dense

FP4
~15,000 TFLOPSSKU-dependent; sparsity doubles the slide
FP8
~7,000 TFLOPS
FP16 / BF16
~3,500 TFLOPS
Precision support
FP4, FP6, FP8, FP16, BF16, TF32, FP32, INT8

Memory

Capacity
288 GB HBM3eSome OEM tables list ~270 GB, confirm the listing
Bandwidth
~8,000 GB/s
ECC
Yes

Silicon

Architecture
Blackwell Ultra
Positioning
Above B200 on memory and FP4, not a quiet refresh

Fabric and host

GPU interconnect
NVLink 5 · 1.8 TB/s
Host interface
PCIe 5.0 x16

Power

Board power
1,200–1,400 WDense racks are liquid-cooled

B300 figures are still SKU-noisy across OEM decks. I publish the envelope clouds use on this index and I will not invent an NVIDIA PDF that is not on the listing. Confirm VRAM and TDP on the provider page.

Trade-offs

Pros and cons of B300

What this GPU does well, and when the hour is the wrong buy.

Strengths

  • 288 GB is a different class of “fits on one GPU”
  • NVLink 5 still at 1.8 TB/s, the fabric matches the memory
  • FP4 headroom above B200 if the stack keeps quality

Limits

  • Power, cooling, and price all scale faster than most jobs
  • Availability is a rumor until the catalog row is green
  • Easy to confuse with a GB300 rack SKU. You are buying a GPU hour, not a NVL72

Who this is not for

If you are comparing B300 to a 5090 because both say Blackwell, stop. One is a furnace with HBM. The other is a gaming card with 32 GB. The architecture name is not a product.

The target reader here already knows their KV cache size. If you do not, start on H100. Curiosity on B300 is how invoices happen.

Versus B200

You are buying more HBM and more FP4, not a new interconnect generation. If 192 GB fits, B200 is the saner hour, when it exists. If it does not fit, B300 is the point of the SKU, not a flex.

How I actually rent it

  1. 01

    Prove 192 GB fails

    If B200 fits, you do not need this page.

  2. 02

    Read power and cooling

    A 1.4 kW listing on a host that cannot cool it is a throttle, not a deal.

  3. 03

    Keep Hopper in the tab

    H200 141 GB still solves a surprising number of “we need Blackwell” tickets.

FAQ

B300 questions

Who has the cheapest B300?+

As of September 5, 2026, the cheapest verified in-stock on-demand B300 tracked by GPUBeacon is $7.50 per GPU per hour from Verda.

How much does the B300 cost per month?+

As of September 5, 2026, at 720 hours per month, one B300 costs an estimated $5,814 at the median on-demand price in this sample. Cheapest verified in stock: $5,400 per month on-demand.

Where can I rent an B300?+

GPUBeacon currently tracks B300 configs from 4 providers. The cheapest verified in-stock on-demand configs come from Verda. See the full price comparison above for every provider and config.

Are B300 prices going up or down?+

GPUBeacon publishes a catalog sample, not a 90-day price index. As of September 5, 2026, the median on-demand B300 in this sample is $8.08 per GPU per hour. The cheapest verified in-stock hour is $7.50.

Why choose the B300?+

Labs training past Hopper envelopes, inference with gigantic KV caches, people who already survived B200 software.

When is the B300 not a good fit?+

Everyone still shopping H100. Startups with a 4090-shaped job. Anyone who cannot name why 288 GB beats 141 GB.

Is B300 the same as GB300 NVL72?+

No. B300 is a GPU SKU. GB300 NVL72 is a rack. Do not budget a rack because you needed one more GPU of HBM.

Why is the hour so high?+

HBM, power, scarcity. If the job does not use the 288 GB, you are donating money to a datasheet.

Keep comparing

Alternatives to NVIDIA B300

Same job, different envelope. Compare VRAM, bandwidth and the hour before you lock a badge.

NVIDIA B200

180 GB HBM3e · similar bandwidth to the B300

From$3.75/ GPU / hr

Read the verdict

NVIDIA H200

141 GB HBM3e · 40% less bandwidth than the B300

From$3.00/ GPU / hr

Compare

AMD MI300X

192 GB HBM3 · 34% less bandwidth than the B300

From$1.99/ GPU / hr

Compare

NVIDIA H100

80 GB HBM3 · 58% less bandwidth than the B300

From$1.73/ GPU / hr

Compare

Browse all GPUs

Sources

Where these numbers come from

TFLOPS and TDP are from the vendor datasheet. Hourly prices are from the GPUBeacon catalog sample. Open the originals, then confirm the live SKU with the provider before you rent.