NVIDIA RTX PRO 6000 Blackwell Max-Q GPU servers now available
The NVIDIA RTX PRO 6000 Blackwell Max-Q, a top-tier professional GPU, is now live. Located in the Netherlands, this high-performance configuration is engineered for elite AI, rendering, and HPC tasks. Deployment is available right now for any mission-critical project.
Full server configuration
CPU: AMD EPYC 9355P
32 Cores, 64 Threads, 3.55/4.4GHz
Memory: 128GB REG ECC DDR5 (768GB max.)
Storage: 960GB Enterprise NVMe SSD (Up to 2x PCIe & 6x SATA)
GPU: NVIDIA RTX PRO 6000 Blackwell Max-Q - 96GB DDR7 (Up to 4x)
Port: 1 Gbps - 50TB Traffic (Up to 10Gbps - Unmetered)
IP-KVM: Asus BMC
Starting at €1141.7/month (Ex. VAT)
The platform: ASUS ESC4000A-E12
The foundation of this build is the ASUS ESC4000A-E12 – a 2U GPU server engineered for dense, multi-GPU deployments. It supports up to four dual-slot GPUs with independent CPU and GPU airflow tunnels for thermal optimization. The chassis runs PCIe 5.0 across eight expansion slots, effectively doubling CPU-to-GPU data bandwidth compared to the previous generation – critical when moving massive AI datasets between memory spaces.
Enterprise-grade features include OCP 3.0 networking, ASUS ASMB11-iKVM remote management, and hardware-level Root-of-Trust security. The system runs on a redundant power supply, ensuring uninterrupted operation even in the event of a PSU failure – a must for production AI and HPC workloads. The single-socket design keeps power consumption focused while leaving full headroom for multi-GPU expansion.
The CPU: AMD EPYC 9355P
The server runs on the AMD EPYC 9355P – a 32-core Zen 5 processor with a 12-channel DDR5 memory controller supporting up to 768 GB of ECC-registered RAM at up to 6400 MHz. The CPU side never becomes a bottleneck when preparing inference queues or shuffling training batches. Current base configuration ships with 128 GB REG ECC DDR5, with in-place expansion available on request.
Storage starts at a 960 GB Enterprise NVMe SSD (2x 7.68TB max.) and scales to six 7.68 TB enterprise SSDs or six HDDs up to 16 TB each – fast local access to large datasets without relying on network storage.
The GPU: NVIDIA RTX PRO 6000 Blackwell Max-Q
The heart of this server is the NVIDIA RTX PRO 6000 Blackwell Max-Q – the density-optimized variant of NVIDIA's professional GPU flagship. Built on the Blackwell architecture, it brings 5th-generation Tensor Cores, 4th-generation RT Cores, and 24064 CUDA cores in a 300W dual-slot form factor.
The GPU carries 96 GB of GDDR7 memory at 1792 GB/s bandwidth. Large language models that previously required multiple L40s can now run inference on a single card. Scale up to four GPUs and you get a combined 384 GB of high-bandwidth memory in a single 2U chassis.
Key architectural features:
5th-gen Tensor Cores with FP4 precision and DLSS 4 Multi Frame Generation
4th-gen RT Cores – up to 100× more ray-traced triangles (RTX Mega Geometry)
Multi-Instance GPU (MIG): up to 4 fully isolated instances per card (24 GB each)
PCIe Gen 5 – 2× the bandwidth of PCIe Gen 4
9th-gen NVENC + 6th-gen NVDEC with 4:2:2 H.264/HEVC support
3511 AI TOPS at FP4 precision
What is this server built for?
AI inference & LLM serving
Run 70B parameter models in FP8/FP4 quantization on a single GPU – no multi-card setup required. Deploy vLLM or TensorRT-LLM with FP4 quantization for peak throughput. MIG partitioning allows up to 4 fully isolated model instances (24 GB each) to run simultaneously on a single GPU.
AI model fine-tuning
Fine-tune transformer models with full-precision (practical for models up to ~13B), LoRA, or QLoRA strategies for larger architectures. The 96 GB framebuffer fits large batches that smaller GPUs simply cannot handle, reducing multi-GPU synchronization overhead and complexity.
Generative AI & media production
Image, video, and 3D content generation workflows benefit directly from Blackwell's 5th-gen Tensor Cores. Text-to-video pipelines deliver up to 3.3× higher throughput compared to L40S. NVENC/NVDEC with 4:2:2 H.264 and HEVC support covers professional broadcast and streaming pipelines.
Life sciences & HPC
Genomic sequencing, drug discovery, and protein structure prediction achieve multi-fold speedups. NVIDIA Parabricks runs germline analysis pipelines up to 1.75× faster than H100 on the same tasks.
3D rendering & visualization
RTX Mega Geometry enables up to 100× more ray-traced triangles per scene. Complex architectural visualization, VFX compositing, and real-time virtual production workflows are all natively GPU-accelerated.
Data Science & analytics
RAPIDS and CUDA-X accelerated pipelines benefit from both the large VRAM buffer and PCIe 5.0 bandwidth. Process datasets that would exceed typical GPU memory limits – and scale to four GPUs as workloads grow.
Don't settle for standard specs. Tailor this Blackwell powerhouse to your exact requirements with expanded RAM, multi-GPU arrays, or massive storage upgrades. High-speed 10 Gbps unmetered bandwidth is available as an option to ensure your data never hits a bottleneck.
Explore the full GPU server lineup!
