Announced, not yet shipping: SambaNova expects SN50 deliveries in the second half of 2026, so these are vendor figures for an unreleased part. SambaNova sells the SN50 inside its SambaRack systems and cloud rather than as a standalone card, and publishes no memory bandwidth or per-chip power for it.
SambaNova logo

SambaNova SN50 NEW

Type: ASICArchitecture: RDUForm factor: CustomAnnounced: 2026Process: 5nmStatus: Announced, not shippingSpec confidence: Vendor claimed
FP8 (dense)
3,200
TFLOPS
Memory
64 GB
HBM2E

Overview

The SN50 is SambaNova's fifth-generation Reconfigurable Dataflow Unit, announced in February 2026 as a chip built for agentic inference, where many models and long contexts have to stay resident and token generation has to be fast. Like the SN40L before it, it runs a model as a dataflow pipeline laid out across the chip rather than as a sequence of kernels, and feeds it from three memory tiers: 432 MB of on-chip SRAM, 64 GB of HBM2E for active weights and hot KV cache, and a large DDR5 pool for cold caches and other models, quoted as up to 512 GB on SambaNova's product page and up to 2 TB on its datasheet. SambaNova rates it at 1,600 TFLOPS of BF16 and 3,200 TFLOPS of FP8, five times the SN40's 640 at its best format, on TSMC 5nm. SambaRacks of 16 SN50s can be joined into a scale-up domain of up to 256 chips, against 16 for the SN40L. SambaNova has published no memory bandwidth or chip power, and shipping begins in the second half of 2026.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
No verified data available
TF32
No verified data available
BF16
Brain Float 16
1,600TFLOPS
FP16
No verified data available
FP8
8-bit floating point
3,200TFLOPS
FP6
No verified data available
FP4
No verified data available
INT8
No verified data available

Specifications

Architecture

RDU

Form Factor

Custom

Launch Year

2026

Process Node

5nm

Memory

64 GB HBM2E

Bandwidth

No verified data available

TDP

No verified data available

Max power

No verified data available

Spec Confidence

Vendor claimed

Full Specifications

Memory
Memory 64 GB
Memory Type HBM2E
Bandwidth No verified data available
Interface Width No verified data available
On-Die SRAM 432 MB
Power & Thermal
TDP No verified data available
Max power No verified data available
General
Form Factor Custom
Architecture RDU
Process Node 5nm
Launch Year 2026

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
SambaNova
Published
No verified data available

Data Quality

Spec confidence
Vendor claimed
Clock basis
Boost
Core precisions with figures
2 of 9
Normalization
All values in TFLOPS

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Similar GPUs

SambaNova SN40L

·
FP32: No verified data available
Compare vs SN40L

Frequently Asked Questions

How many TFLOPS does the SambaNova SN50 have?

The SambaNova SN50 delivers 3,200 TFLOPS FP8 at peak. Flopper does not currently have verified FP32 and FP16 throughput figures for it.

What is the power consumption of the SambaNova SN50?

Flopper does not currently have a verified TDP (Thermal Design Power) figure for the SambaNova SN50.

How much memory does the SambaNova SN50 have?

The SambaNova SN50 is equipped with 64 GB of memory. Flopper does not currently have a verified memory bandwidth figure for it.

What architecture is the SambaNova SN50 based on?

The SambaNova SN50 is based on the RDU architecture, launched in 2026.

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

© 2026 Flopper.io - Compare the GPUs Powering AI