AMD CDNA 3 bare metal for memory-rich LLM inference, training, and FP64 HPC: 192 GB HBM3 at 5.3 TB/s per GPU, 1,307 TFLOPS FP16/BF16, ROCm 6, and up to 8 OAM GPUs with 1.5 TB of HBM3 per node. Reserve capacity now.
Thanks. An engineer will spec your MI300X build and reach out shortly.
Something went wrong. Please try again or email sales@redswitches.com.
MI300X servers are landing in our data centers soon. Leave your details and our engineers will reserve a configuration for you, confirm lead times, and quote single-GPU, 4x and 8x Infinity Fabric builds with volume and committed-term discounts.
Thanks for reaching out. An engineer will spec your build and get back to you shortly.
Something went wrong. Please try again or email sales@redswitches.com.
Every RedSwitches AMD Instinct MI300X server is single-tenant bare metal with full root and IPMI access, unmetered 1, 10 or 25 Gbps bandwidth, no setup fee, no bandwidth overage and a flat monthly price. Stocked builds are online in about 1 hour, larger configurations up to 8 GPUs per node are built to order, volume and 6/12-month committed-term discounts apply, payments include crypto, and engineers answer 24/7.
Reservations for AMD Instinct MI300X are free and non-binding. An engineer confirms the lead time, holds the hardware for you, and sends a quote before you commit to anything.
No setup fee and a flat monthly price for the whole server. No per-hour meter and no surprise line items, so GPU spend is forecastable.
Unmetered 1, 10 or 25 Gbps uplinks are included, and whatever port speed you choose you can use all of it: no overage charges and no egress bills, ever. Moving datasets, checkpoints and model weights in and out costs nothing extra.
Single-tenant hardware with root and IPMI access. You choose the OS, drivers, CUDA or ROCm version, and the NVLink or MIG layout, and a private VLAN can link your RedSwitches servers.
Single and multi-GPU nodes, up to 8 GPUs per server with NVLink where the card supports it. Volume discounts on multi-GPU and multi-server orders, plus 6 and 12-month term savings.
Live chat, Telegram and email answered by engineers around the clock. Pay by card, PayPal, bank wire or crypto with no KYC, in 20+ Tier III data centers across the EU, US and Asia.
Same accelerator, different economics: what changes when the GPU sits in a dedicated server you control instead of a metered instance.
| RedSwitches MI300X server | Typical cloud GPU instance | |
|---|---|---|
| Billing | Flat monthly price per server | Per-hour or per-second metering |
| Bandwidth | Unmetered 1/10/25 Gbps, use the full port, no overage or egress fees | Egress billed per GB |
| Private networking | Private VLAN between your servers on request | Paid VPC and peering constructs |
| Setup fee | $0 | Varies by instance and region |
| Tenancy | Single-tenant bare metal | Shared, virtualised hosts |
| Access | Root and IPMI, your OS and drivers | Hypervisor-managed images |
| Multi-GPU | Up to 8 per node, NVLink where supported, built to order | Fixed instance shapes |
| Payments | Card, PayPal, bank wire, crypto | Card or invoice |
| Support | 24/7 engineers on chat, Telegram and email | Ticket tiers, paid support plans |
CDNA 3 chiplet silicon with 192 GB HBM3, 5.3 TB/s bandwidth, 1,307 TFLOPS FP16/BF16, and Infinity Fabric links across 8 OAM GPUs.
192 GB of HBM3 per GPU, 5.3 TB/s of bandwidth, and an open ROCm stack: fewer GPUs per model and no CUDA lock-in.
192 GB of HBM3 per GPU is 2.4x the 80 GB on NVIDIA H100 SXM and more than the 141 GB on H200: a 70B-parameter model in FP16 fits on one MI300X, and Llama 3.1 405B runs on a single 8-GPU node without spanning hosts.
Token generation is limited by memory bandwidth, not compute. 5.3 TB/s per GPU, backed by 256 MB of Infinity Cache, keeps decode-phase inference and large-batch training moving with fewer stalls.
8 accelerator complex dies on 4 I/O dies, 153 billion transistors, 304 compute units and 1,216 matrix cores deliver 1,307.4 TFLOPS FP16/BF16 and 2,614.9 TFLOPS FP8 per GPU, doubled again with structured sparsity.
PyTorch, TensorFlow, JAX, vLLM, SGLang, Triton and Hugging Face run natively on ROCm 6, and hipify ports existing CUDA kernels, so teams diversifying off CUDA keep their code and their workflow.
81.7 TFLOPS of FP64 vector and 163.4 TFLOPS of FP64 matrix compute per GPU mean the same node that serves LLMs also runs CFD, molecular dynamics and climate models at full double precision.
Single-tenant servers with full root and IPMI access, dual AMD EPYC or Intel Xeon hosts, NVMe storage, unmetered 1/10/25 Gbps uplinks, and no noisy neighbours, with the 8-GPU Infinity Fabric topology under your control.
From memory-rich LLM inference to ROCm training and FP64 HPC, where 8x MI300X nodes pay for themselves.
Serve Llama, DeepSeek, Mixtral and proprietary models with vLLM or SGLang; 192 GB per GPU means fewer GPUs per replica and more replicas per node.
Pre-train and fine-tune with PyTorch on ROCm 6 across 8x Infinity Fabric nodes, with 1.5 TB of HBM3 for larger batches and longer sequences.
Keep embeddings, reranker and generator resident on one node and serve long-context RAG with room for large KV caches.
Train and serve vision-language, diffusion and video-generation models whose activations outgrow 80 GB cards.
81.7 TFLOPS FP64 vector per GPU for CFD, climate, molecular dynamics and finite element simulation on bare metal.
Build inference and training platforms on an open stack: ROCm, PyTorch, Triton and Hugging Face on dedicated hardware you control.
Trusted by Enterprise Teams Worldwide
Real reviews from Google, HostAdvice, and Cryptwerk.
Port speeds on cloud were capped and shady. With RS I actually get the 25Gbps they say. No throttle bs.
Set up multiple Solana + Avalanche validators through them. Hardware was clean, latency was low (especially in Europe), and uptime’s been 100% so far.
As a sysadmin, I care more about control than flashy dashboards. RedSwitches gives me root access, IPMI, and actual hardware specs I can configure. Good for serious users.
RedSwitches gave us full control over our infrastructure without the vendor lock-in we kept running into with cloud hosts. Customizable builds, fast provisioning, and actual humans handling support tickets. It’s refreshing.
My website works smoothly, thanks to their 99.99% uptime guarantee. The support team is another plus for me, as I always have been able to get help whenever I needed it in just 5 minutes on average.
Deployed a few Ethereum and Bitcoin nodes here. Uptime’s been flawless, and sync speed was great thanks to their storage config.
Been using RedSwitches for 6 months now for my small game server biz. Uptime has been great, and I haven’t run into any hidden charges.
Using the storage servers to archive logs and snapshots from our AI pipeline… I also love that I could pick the datacenter closest to our team.
Support is super responsive. I had an issue with an OS reinstall and they jumped in within 10 minutes… Transparent pricing = win.
They have an instant delivery section… they delivered it within 120 mins with all my requirements fulfilled (OS/RAID/Software configured etc).
With servers available in numerous strategic locations, RedSwitches offers exceptional versatility and performance for our company’s diverse hosting needs. Plus, their no setup fee policy really helps keep costs down.
The dedicated server I got from RedSwitches has been incredibly reliable and fast. Their bare metal cloud solutions offer excellent performance, and the cloud VPS options are perfect for scaling. Highly recommended!
Absolutely delighted with RedSwitches! The setup was quick and free, and the fact that they accept all major payment gateways made the process seamless.
Very good experience using their bare metal servers. Their customer service is one of the finest I have experienced - always prompt at resolving troubles. Highly recommend.
Availability and reservations, MI300X vs H100/H200 and MI325X, ROCm and CUDA porting, 8-GPU nodes, pricing, power, and drivers.
MI300X capacity is being installed across our data centers now. Submit the reservation form with your preferred configuration and region; our engineers confirm the lead time for that build, hold the hardware for you, and send a quote. Reservations are free and non-binding until you approve the quote.
NVIDIA H100 SXM has 80 GB of HBM3 at 3.35 TB/s, H200 has 141 GB at 4.8 TB/s, and MI300X has 192 GB of HBM3 at 5.3 TB/s. Memory per GPU decides how many GPUs a model needs: a 70B-parameter model in FP16 needs about 140 GB of weights, which fits on one MI300X but must be split across two H100s, and an 8x MI300X node holds 1.5 TB, enough for Llama 3.1 405B in FP16 with room for KV cache. Fewer GPUs per model means less inter-GPU communication and more replicas per node, which is why MI300X is a strong fit for memory-bound inference, while teams with deep CUDA-only dependencies may still prefer H100 or H200.
Both are AMD CDNA 3 GPUs that run the same ROCm 6 software stack. MI325X moves to 256 GB of HBM3E at 6 TB/s with 1,000 W peak board power; MI300X has 192 GB of HBM3 at 5.3 TB/s at 750 W. Reserve MI325X when model weights plus KV cache per GPU exceed 192 GB; otherwise MI300X delivers the same programming model at lower power per node. We quote both, so tell us your model sizes and we will recommend a configuration.
Pricing is quoted per configuration because host CPU, memory, storage and uplink choices vary widely. Volume discounts apply to multi-GPU and multi-server orders, and every GPU server qualifies for committed-term discounts on 6 and 12-month billing. Crypto, card, PayPal and bank wire are accepted.
For most workloads, yes. ROCm 6 ships native PyTorch, TensorFlow and JAX builds, and vLLM, SGLang, Triton and Hugging Face Transformers support MI300X, so Python-level code typically runs unchanged. Custom CUDA kernels are ported with hipify, which translates CUDA source to HIP, the portable C++ dialect that compiles for both AMD and NVIDIA GPUs. Our engineers can review your stack before you commit to a configuration.
Yes. MI300X ships as an OAM module on the AMD Instinct MI300X platform baseboard, so 8-GPU configurations with Infinity Fabric all-to-all links and 1.5 TB of HBM3 are the native form factor. We also quote single-GPU and 4x builds. Tell us your target topology in the reservation form and we will size the host CPUs, RAM, NVMe and uplinks around it.
Our GPU racks are provisioned for high-density, high-power accelerators; an 8x MI300X node is deployed with the power feeds and cooling it needs. You do not need to plan facilities, only the workload.
Ubuntu, Debian, Rocky/AlmaLinux or your own ISO, with ROCm 6, the AMD GPU driver and a container runtime pre-installed on request, so the node is ready for your PyTorch or vLLM containers on day one.
No problem. Our talented engineers will consult, architect, migrate, manage, and do whatever it takes to help your business grow and succeed.