NVIDIA B300 SXM 288GB

Name: NVIDIA B300 SXM 288GB
Brand: NVIDIA

Blackwell Ultra SXM 2025 4nm Spec confidence: Official

FP8 (dense)

4,500

TFLOPS

FP32

TFLOPS

VRAM

288 GB

HBM3e

Bandwidth

8.0 TB/s

memory

TDP

1.4 kW

53.6 TFLOPS/kW

Download datasheet Rental pricing from $3.75/hr →

Overview

The NVIDIA B300 is the Blackwell Ultra refresh of the B200, and it is a memory and inference part rather than a general uplift. Capacity rises to 288 GB of HBM3e at 8 TB/s, board power rises to 1,400 W, and dense NVFP4 rises by half again to 13.5 PFLOPS while sparse NVFP4 stays where the B200 left it. That asymmetry is the whole design: Blackwell Ultra spends its transistor budget on dense low-precision throughput and on attention, which NVIDIA rates at twice the B200's, because reasoning models generate long chains of tokens and are bound by decode rather than by training arithmetic. The trade is visible in what got cut. INT8 is not a serious datapath on this part, sitting far below FP8 because the silicon behind it was reallocated to NVFP4, and FP64 is similarly de-emphasised, so a B300 is a poor choice for classical HPC in a way a B200 is not. Read those two rows as deliberate, not as errors.

Performance

Peak theoretical throughput by precision type

FP6

4,500 TFLOPS

9,000 with sparsity

INT8

8-bit integer

188 TOPS

375 with sparsity

FP8

8-bit floating point

4,500 TFLOPS

9,000 with sparsity

FP4

4-bit floating point

13,500 TFLOPS

18,000 with sparsity

FP16

16-bit floating point

2,250 TFLOPS

4,500 with sparsity

BF16

Brain Float 16

2,250 TFLOPS

4,500 with sparsity

TF32

TensorFloat-32

1,125 TFLOPS

2,250 with sparsity

FP32

32-bit floating point

75 TFLOPS

FP64

64-bit floating point

1.3 TFLOPS

Dense Added with sparsity

3.214 TFLOPS/W FP8 at 1400 W

1.607 TFLOPS/W FP16 at 1400 W

0.054 TFLOPS/W FP32 at 1400 W

The B300 SXM 288GB in the GPU landscape

Peak FP16 TFLOPS (dense) against TDP, single-GPU parts tracked by Flopper

Higher and further left is better: more half-precision throughput for less power.

Specifications

Architecture

Blackwell Ultra

Form Factor

SXM

Launch Year

2025

Process Node

4nm

Memory

288 GB HBM3e

Bandwidth

8,000 GB/s

TDP

1.4 kW

Max Power

~1.6 kW

Spec Confidence

Official

Full Specifications

Memory
VRAM	288 GB
Memory Type	HBM3e
Bandwidth	8.0 TB/s
Power & Thermal
TDP	1.4 kW
Enterprise Features
Sparsity	Yes
General
Form Factor	SXM
Architecture	Blackwell Ultra
Launch Year	2025

Datasheet & Resources

Flopper Datasheet

NVIDIA B300 SXM 288GB specifications, generated from our database. Printable.

View PDF

Vendor product page

NVIDIA documentation for the B300 SXM 288GB

View

NVIDIA Developer Documentation

Technical specs, programming guides

Browse

Data Provenance

Every figure traced to a source

Primary Source

Document: --
Publisher: NVIDIA
Published: --

Data Quality

Spec confidence: Official
Clock basis: Boost
Precisions tracked: 9
Normalization: All values in TFLOPS

Rent this GPU

$3.75 / GPU-hr

Lowest of 11 live listings across 7 providers· median $7.50

See all rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

Similar GPUs

NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip 252GB

Superchip · 2025

FP32: 80 TFLOPS

Compare vs GB300

NVIDIA B30A

· 2025

FP32: --

Compare vs B30A

Systems Using This GPU

Pre-configured systems featuring the NVIDIA B300

System	GPU Count	Peak Performance	Total Power	Interconnect
HPE Compute XD690 Compute XD · rack · 2025	8x B300	108.00 PFLOPS	14.3 kW	NVLink 5.0 / NVSwitch 4.0 across the HGX B300 baseboard; 8x NVIDIA ConnectX-8 for InfiniBand XDR or Spectrum-X Ethernet scale-out	View System
NVIDIA DGX B300 DGX · rack · 2025	8x B300	108.00 PFLOPS	14.0 kW	NVSwitch + NVLink	View System
NVIDIA DGX GB300 NVL72 NVL · rack · 2025	72x B300	1080.00 PFLOPS	--	NVLink + NVSwitch	View System
NVIDIA HGX B300 8-GPU HGX · baseboard · 2025	8x B300	108.00 PFLOPS	--	NVLink 5.0 / NVSwitch 4.0	View System
HPE ProLiant Compute XD685 (HGX B300) ProLiant Compute XD · rack · 2025	8x B300	108.00 PFLOPS	--	NVLink 5.0 / NVSwitch 4.0 across the HGX B300 baseboard; 4x FHHL PCIe Gen5 x16 slots for fabric, NIC or DPU	View System
Inventec Artemis II Rack Artemis · rack ·	72x B300	--	136.0 kW	NVLink across all 72 GPUs over passive copper cable cartridges, with nine 1RU NVLink switch trays; scale-out is ConnectX-8 at 800 Gb/s InfiniBand east-west, carried on two I/O boards holding two SuperNICs each, plus a BlueField-3 DPU at 400 Gb/s north-south	View System
Lenovo Lenovo NVIDIA GB300 NVL72 NVL · rack ·	72x B300	--	135.0 kW	Fifth-generation NVLink across all 72 GPUs through nine switch trays, joined by a rear cable cartridge backplane; the NVLink domain is contained within one rack, with no inter-rack scaling, and each compute tray adds up to three PCIe Gen 5 x16 slots for networking	View System
HPE NVIDIA GB300 NVL72 by HPE NVL · rack ·	72x B300	--	132.0 kW	Fifth-generation NVLink across all 72 GPUs through nine NVLink switch trays; each compute tray carries two ConnectX-8 ports at 800 Gbps and one BlueField-3 DPU, with two Spectrum SN2201 top-of-rack switches for management	View System
Inventec P5800G7 P-Series · rack ·	8x B300	--	--	One NVIDIA HGX B300 NVL8 liquid-cooled baseboard, with eight 800Gb OSFP connectors driven by NVIDIA ConnectX-8 SuperNICs for east-west traffic	View System
Inventec P9000G7 (AC) P-Series · rack ·	8x B300	--	--	One NVIDIA HGX B300 air-cooled baseboard, with eight 800Gb OSFP connectors for east-west traffic	View System
Fujitsu PRIMERGY GX2580 M8s PRIMERGY GX · rack ·	8x B300	--	--	One NVIDIA SXM HGX B300 8-GPU baseboard integrating eight Blackwell Ultra SXM modules with high-speed interconnect; scale-out is eight NVIDIA ConnectX-8 OSFP 800GbE ports, alongside two 10GbE RJ45 on an Intel X710	View System
QCT QuantaGrid D75H-10U QuantaGrid · rack ·	8x B300	--	--	One NVIDIA HGX B300 8-GPU baseboard; scale-out is eight ConnectX-8 OSFP ports carrying 800G Ethernet or InfiniBand at one port per GPU for GPU-Direct RDMA	View System
QCT QuantaGrid D75L-3U QuantaGrid · rack ·	8x B300	--	--	One NVIDIA HGX B300 8-GPU baseboard with fifth-generation NVLink at 1.8 TB/s; eight onboard ConnectX-8 SuperNICs at up to 800Gb/s for east-west traffic, with up to three further NICs for north-south	View System
Penguin Solutions Relion XE4418GTS-DTC Relion · rack ·	8x B300	--	--	Penguin publishes no interconnect figure; expansion is 4x PCIe Gen5 x16 FHHL double-width	View System
Penguin Solutions Relion XE8418GTS Relion · rack ·	8x B300	--	--	Penguin publishes no interconnect figure; expansion is 4x PCIe Gen5 x16 FHHL double-width	View System
Lenovo ThinkSystem SR680a V4 ThinkSystem · rack ·	8x B300	--	--	One NVIDIA HGX B300 SXM baseboard, which Lenovo pairs with fast interconnects between GPUs, CPUs and networking	View System

Frequently Asked Questions

How many TFLOPS does the NVIDIA B300 have?

The NVIDIA B300 delivers 75 TFLOPS for FP32 operations, 2250 TFLOPS for FP16, and 4500 TFLOPS for FP8 precision.

What is the power consumption of the NVIDIA B300?

The NVIDIA B300 has a TDP (Thermal Design Power) rating of 1400 watts.

How much memory does the NVIDIA B300 have?

The NVIDIA B300 is equipped with 288 GB of VRAM with 8000 GB/s memory bandwidth.

What architecture is the NVIDIA B300 based on?

The NVIDIA B300 is based on the Blackwell Ultra architecture, launched in 2025.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs