TATC
Services
Token API
Explore model access, integration and usage-based billing
Bare Metal
Dedicated GPU servers for AI training, inference and content-generation workloads
Configurations for EnquiryComing Soon
Applicable Scenarios
AI Vision
Real-time visual analytics for retail and operations
Cross-Border E-Commerce AI
Potential AI workflows for recommendations, listings and customer assistance
MCN & Digital Live Streaming
GPU configuration options for digital media and AIGC workloads
News & Events
Help Center
Support & Tickets
Support enquiries and agreement-based response arrangements
Docs
Technical documentation, FAQ and changelog
Company
About Us
Who we are and where we operate
Responsible Use
Customer review and acceptable use
Contact Us
Sales and technical support
enEnglishthไทย
View ProductsContact Us
TATC

Advanced Computing for Southeast Asia

9/11 Moo 12, Khlong Song, Khlong Luang District, Pathum Thani 12120, Thailand
info@tatc.cloud

Services

  • Token API
  • Bare Metal

Applicable Scenarios

  • AI Vision
  • Cross-Border E-Commerce AI
  • AIGC

News & Events

  • News
  • Events

Help Center

  • Support & Tickets
  • Docs

Company

  • About Us
  • Responsible Use
  • Contact Us

© 2026 TATC TECHNOLOGY CO., LTD. Thailand Registration No. 0135567035437. 9/11 Moo 12, Khlong Song, Khlong Luang District, Pathum Thani 12120, Thailand

Privacy PolicyCookie Policy
  1. Services
  2. /Next-generation GPUs
  3. /AMD Instinct MI355X
AMD
Roadmap

AMD Instinct MI355X

288GB HBM3E accelerator with fourth-generation AMD CDNA architecture for generative AI, training, inference and HPC

AMD MI350 Series

Specifications (manufacturer data)

AMD Instinct MI355X combines 288GB HBM3E memory, fourth-generation AMD CDNA architecture and native MXFP4 processing for generative AI, model training, inference and HPC workloads.

Figures marked TBC / est. are preliminary manufacturer specifications and remain subject to confirmation. Roadmap status and manufacturer specifications do not represent available capacity or a committed delivery date.

Discuss this GPUBrowse GPUs available for enquiry
Architecture
4th Gen AMD CDNA
VRAM per card
288GB HBM3E
Memory bandwidth
8 TB/s
FP16 / BF16 tensor
2.5166 / 5.0332 PFLOPS (matrix, dense / with sparsity)
FP8 tensor
5.0332 / 10.0664 PFLOPS (OCP-FP8, dense / with sparsity)
FP4 tensor
10.0663 PFLOPS (MXFP4)
TDP
1400W maximum TBP
Interconnect
7× Infinity Fabric links · 153.6 GB/s each

Why it matters

  • 288GB HBM3E with 8 TB/s on-package memory bandwidth
  • 256 compute units, 1,024 matrix cores and 16,384 stream processors
  • Native MXFP4 compute delivering up to 10.0663 PFLOPS

Built for

  • High-capacity large-model inference
  • Mixture-of-Experts hosting
  • Cost-optimised ROCm training runs

Software stack

ROCm
PyTorch
JAX
Triton
vLLM-ROCm
Compare accelerators

NVIDIA BLACKWELL ULTRA — NVIDIA HGX B300

NVIDIA BLACKWELL ULTRA

NVIDIA HGX B300

Eight-GPU Blackwell Ultra platform with 2.1TB HBM3E for AI reasoning, training and inference

Memory
270GB HBM3E
Bandwidth
7.7 TB/s per GPU
TDP
Up to 1,100W per GPU (configurable)
View details
Configurations for enquiry

Reference NVIDIA configurations

Discuss the selected GPU, quantity, node design, deployment location and commercial terms with our team.

View enquiry configurations

Evaluate the right accelerator for your workload

Tell us your target model, workload profile and infrastructure requirements to compare accelerator options.

Discuss GPU requirementsView Products