Accelerated Compute
Research-optimized GPU compute for AI Natives and enterprises
Train, fine-tune, and deploy on self-service GPU clusters optimized by frontier research — with flexible pricing, production reliability, and security built in.
Why Together Accelerated Compute?
Research-led performance, NVIDIA-validated architecture, and enterprise-grade reliability to train, fine-tune, and serve at any scale.
Faster training & inference
Together Kernel Collection delivers custom CUDA kernels from the FlashAttention team — 90% faster training on NVIDIA Blackwell GPUs with software-led optimization on top of NVIDIA accelerated hardware.
Predictable performance at scale
Go from zero to running workloads in minutes with batteries-included GPU clusters. Virtualized for elasticity, bare-metal where it matters — GPU and network performance without compromise. Pre-configured drivers, built-in observability, and managed orchestration included.
Production reliability & security
Self-healing infrastructure with 99.9% uptime, backed by SOC 2 Type II encryption in transit/at rest, and tenant-level isolation.
Performance-optimized NVIDIA hardware
- NVIDIA GB200 NVL72Memory
192GB HBM3e per GPU
Use CaseHeavy training and inference at frontier scale
AvailabilityAvailable now through AI Factory
Learn how - NVIDIA B200Memory
192GB HBM3e
Use CaseBlackwell architecture
AI reasoning and training workloadsAvailabilityAvailable now through GPU Clusters
Learn how - NVIDIA H200Memory
141GB HBM3e
Use CaseExtended memory
Large model training and inferenceAvailabilityAvailable now
Starting PriceFrom $3.99/GPU-hour
Learn how - NVIDIA H100Memory
80GB HBM2e
Use CaseProven performance
Foundation model workloadsAvailabilityAvailable now
Starting PriceFrom $3.09/GPU-hour
Learn how
GPU Clusters, built for production
Choose the right GPUs, deploy with the orchestration stack you prefer, and operate with the observability and security required for production.
Managed infrastructure
Pre-configured driversObservabilityZero setup overheadDeploy GPU clusters with integrated observability, managed orchestration, drivers, and networking entirely pre-configured. Run production workloads instantly without manual infrastructure setup.
Orchestration flexibility
KubernetesSlurmFully managedDeploy Kubernetes for open-source extensibility, or run Slurm for precise hardware control and gang scheduling. Both fully managed.
Self-healing infrastructure
Acceptance testingAutomated remediationHealth checksKeep workloads running through hardware events using automated remediation and continuous health checks. Every GPU passes rigorous acceptance testing before joining the cluster.
Flexible pricing modes
No commitmentsGuaranteed capacitySelf-serve reservationsOn-demand for flexibility, reserved for guaranteed capacity and better rates. Both fully self-serve.
Frontier research-powered training performance
The Together Kernel Collection, built by our Chief Scientist Tri Dao (creator of FlashAttention), delivers improved training and inference performance.
TKC
ThunderKittens
AI Training Performance: NVIDIA Hopper to Blackwell, with TKC
TKC vs SOTA Approaches
90% faster training
Training a 70B parameter Llama-architecture model (BF16) with an optimized TorchTitan + Together Kernel Collection (TKC) reached 15,264 tokens/second/GPU on NVIDIA HGX B200, up from 8,080 tokens/second on NVIDIA HGX H100—a 90% jump in training speed.
learn moreFP8 GEMM Performance (M x N x K)
- ThunderKittens B200
- cuBLAS H100
- cuBLASB200
ThunderKittens vs cuBLAS
~2× faster
ThunderKittens’ FP8 kernel for NVIDIA HGX B200 matches NVIDIA cuBLAS GEMM performance while delivering ~2× speedup over H100 FP8 GEMMs, leveraging Blackwell’s Tensor Core–accelerated matrix operations.
learn more
Deployments for any scale
- GPU ClustersGPU clusters at scale
Spin up clusters in minutes with Kubernetes or Slurm. Choose on-demand or reserved capacity. Scale from 8 GPUs to 4,000+.
Explore GPU Clusters - AI FactoryCustom infrastructure at frontier scale
Bespoke infrastructure at factory scale, starting from 1,000+ GPUs. Powered by NVIDIA accelerated compute with Together's research team continuously optimizing performance for your AI workloads.
Contact Sales
Regions and availability zones
Launch close to your users and data across 25+ cities.
USA2GW+ in the portfolio with 600MW of near-term capacity in US.
Europe150 MW+ available in Europe: France, Netherlands, Sweden, and Romania.
Asia & Middle EastOptions available based on the scale of the projects in Asia and the Middle East.
Choose from global regions to meet data residency and compliance requirements—HIPAA for healthcare, GDPR for Europe, or banking regulations.
Reference-architecture performance.
Production-grade security.
We take security and compliance seriously, with strict data privacy controls to keep your information protected. Your data and models remain fully under your ownership, safeguarded by robust security measures.
preferred partner
SOC 2 Type II
ISO 27001:2022
Customers running GPU Clusters in production
"Together GPU Clusters provided a combination of amazing training performance, expert support, and the ability to scale to meet our rapid growth to help us serve our growing community of AI creators."
Demi Guo
CEO, Pika
“Together AI provides the performance and reliability we need for real-time, high-quality image and video generation at scale. We value that Together AI is much more than an infrastructure provider — they're a true innovation partner, enabling us to push creative boundaries without compromise.”
Victor Perez
Co-Founder, Krea