Sophgo publishes no compute figure for this card. Its own brochure row headed 深度学习算力, which translates as deep learning compute, reads "支持INT8/FP8/FP16/BF16/FP32/TF32" and contains no number, and the product page says the same thing in English. The six datatypes are therefore listed here as supported with no rate. A ladder does exist and it is Sophgo's own writing, in the product submission it filed to the China Security and Protection Industry Association in September 2024: more than 400 TOPS of INT8 and FP8, more than 200 TFLOPS of FP16 and BF16, more than 100 TFLOPS of TF32, and more than 25 TFLOPS of FP32. Every one of those is prefixed 超过, meaning more than, so all four are lower bounds rather than measurements, and none is stored as a value. The memory bandwidth is the one number Sophgo states both ways: the English product page gives a flat 1.1 TB/s, which is what is stored, while the association submission writes 超过1.1TB/s. Power is held as a maximum, not a TDP, because the brochure files 300 W under 最大功耗. The card also ships in a 128 GB configuration, held on this one row at the headline 256 GB. Sophgo names no processor for this card anywhere; its corporate timeline records a BM1690 released in 2024 and the connection is never made in print, so no chip and no architecture are asserted here. The 256 decode and 128 encode figures are concurrent 1080p30 streams, not engine counts.
SOPHGO logo

SOPHGO SC11 FP300

Type: ASICForm factor: PCIeReleased: 2024Spec confidence: Official
Memory
256 GB
LPDDR5X
Bandwidth
1.1 TB/s
memory

Overview

The SC11 FP300 is Sophgo's cloud accelerator card and the first part in this catalogue from the Chinese TPU and RISC-V designer. It is a full-length full-height dual-width PCIe Gen5 card weighing 1.5 kg and drawing at most 300 W, and its defining choice is memory: 256 GB of LPDDR5X at 8,533 Mbps giving about 1.1 TB/s, where almost every competitor at this capacity uses HBM. That trade buys capacity and cost at the expense of bandwidth, which is exactly the argument Sophgo makes for it, pitching the card on inference cost and power per token rather than on peak arithmetic. A 128 GB option exists and both carry ECC. Eight cards in a 4U server reach 2 TB of accelerator memory, linked by Sophgo's own SG-Link card-to-card fabric at 256 GB/s. The card also carries an unusually heavy media block for a datacentre accelerator, decoding 256 and encoding 128 concurrent 1080p30 streams in H.264, H.265, VP9 and AV1, and Sophgo embeds RISC-V cores on it so that operators its compiler cannot map can be compiled rather than dropped. Sophgo publishes no throughput figure of any kind for the card, in any document, so the six datatypes it names are recorded here without rates.

Performance

Peak theoretical throughput by precision type

PrecisionPeak
FP64
No verified data available
FP32
32-bit floating point
No verified data available
TF32
TensorFloat-32
No verified data available
BF16
Brain Float 16
No verified data available
FP16
16-bit floating point
No verified data available
FP8
8-bit floating point
No verified data available
FP6
No verified data available
FP4
No verified data available
INT8
8-bit integer
No verified data available

Flopper holds no verified throughput figures for this part yet. The rows stay so the gaps are visible.

Specifications

Architecture

No verified data available

Form Factor

PCIe

Launch Year

2024

Process Node

No verified data available

Memory

256 GB LPDDR5X

Bandwidth

1,100 GB/s

TDP

No verified data available

Max power

300 W

Interconnect

256 GB/s SG-Link

direction and scope not stated by vendor

Spec Confidence

Official

Full Specifications

Memory
Memory 256 GB
Memory Type LPDDR5X
Bandwidth 1.1 TB/s
Interface Width No verified data available
Memory Clock 8.533 GT/s
Interconnect & I/O
GPU-to-GPU SG-Link
Interconnect Bandwidth 256 GB/s direction and scope not stated by vendor
PCIe 5.0
Power & Thermal
TDP No verified data available
Max power 300 W
Cooling Passive
Enterprise Features
ECC Memory Yes
Compute APIs SOPHON SDK, TPU-MLIR, PyTorch, TensorFlow, Caffe, ONNX, vLLM, TGI
Physical & Media
Card Length 269.6 mm
Width 2-slot
General
Form Factor PCIe
Architecture No verified data available
Process Node No verified data available
Launch Year 2024

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
SOPHGO
Published
No verified data available

Data Quality

Spec confidence
Official
Clock basis
Boost
Core precisions with figures
0 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

Frequently Asked Questions

How many TFLOPS does the SOPHGO SC11 FP300 have?

Flopper does not currently have any verified throughput figures for the SOPHGO SC11 FP300.

What is the power consumption of the SOPHGO SC11 FP300?

Flopper does not currently have a verified TDP (Thermal Design Power) figure for the SOPHGO SC11 FP300.

How much memory does the SOPHGO SC11 FP300 have?

The SOPHGO SC11 FP300 is equipped with 256 GB of memory with 1,100 GB/s of memory bandwidth.

What architecture is the SOPHGO SC11 FP300 based on?

The SOPHGO SC11 FP300 was launched in 2024. Flopper does not currently have a verified architecture name for it.

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