GPU servers · NVIDIA RTX PRO 6000 Blackwell
NVIDIA RTX PRO 6000 Blackwell servers with 96 GB of VRAM
Price list updated
You can rent an NVIDIA RTX PRO 6000 Blackwell Server Edition server from RyzenRDP in Frankfurt, Germany: 1, 2, 4 or 8 GPUs per server with 96 GB of GDDR7 memory each, from $1,951.09 a month, billed monthly. The single-GPU server has 16 vCPU, 128 GB RAM, 500 GB disk and a 10 Gbps port; spread over the 730 hours of an average month, its price works out to about $2.67 per GPU-hour, but there is no hourly billing. The card is NVIDIA's Blackwell-generation data center GPU.
- 96 GB of VRAM per GPU
- Up to 8 GPUs in one server
- Ubuntu 24.04 LTS with NVIDIA CUDA
RTX PRO 6000 Blackwell Server Edition specifications
These are NVIDIA's published figures for the card, from its datasheet. They describe the GPU itself, not a measurement of a RyzenRDP server.
| Specification | NVIDIA's figure |
|---|---|
| GPU memory | 96 GB GDDR7 with error correction (ECC) |
| Memory bandwidth | 1,597 GB/s |
| CUDA cores | 24,064 |
| Tensor Cores | 752, fifth generation |
| RT Cores | 188, fourth generation |
| Single-precision performance (FP32) | 120 TFLOPS |
| Graphics bus | PCI Express 5.0 x16 |
| Thermal design | Passive: cooled by the server's airflow |
| Power consumption | Up to 600 W, configurable |
Source: NVIDIA, RTX PRO 6000 Blackwell Server Edition and its datasheet. We publish no benchmark of our own for these servers.
What fits in 96 GB of VRAM?
VRAM decides which models you can load at all, and 96 GB on one GPU is a lot of room. A model's weights take roughly its parameter count times the bytes stored per parameter:
- a 32-billion-parameter model is about 64 GB at 16-bit, so it fits on one GPU unquantized;
- a 70-billion-parameter model is about 35 GB at 4-bit and about 70 GB at 8-bit, so both fit on one GPU, while its 16-bit weights (about 140 GB) need the two-GPU server;
- image and video generation models, with their text encoders, can stay loaded together instead of being swapped in and out.
Add headroom on top of the weights: the context cache grows with the length of the context and the number of simultaneous requests, and the framework needs some memory of its own. These are estimates from arithmetic, not measurements. The LLM hosting page has the full table.
When one GPU is not enough, the larger servers add whole GPUs: 192 GB, 384 GB or 768 GB of VRAM in total. A serving engine such as vLLM splits one model across them.
Server Edition or workstation card?
NVIDIA sells the RTX PRO 6000 Blackwell in more than one form. The Server Edition is the one built for data centers: it has no fans of its own and is cooled by the airflow of the server chassis, and NVIDIA describes it as designed for 24/7 data center operation. The Workstation Edition carries its own cooler and is meant for a desktop workstation under a desk.
Our GPU servers use the Server Edition. For you the practical difference is where the card lives: in a data center in Frankfurt, reachable over SSH, rather than in a machine you have to house, power and cool yourself.
RTX PRO 6000 Blackwell vs RTX 6000 Ada, RTX 5090, L40S and H100: memory
For AI work the first question is whether the model fits, so memory is the number to compare first. The RTX PRO 6000 Blackwell has twice the memory of the RTX 6000 Ada Generation, the professional card of the generation before it, and three times that of a GeForce RTX 5090.
| GPU | Memory |
|---|---|
| NVIDIA RTX PRO 6000 Blackwell Server Edition | 96 GB |
| NVIDIA H100 | 80 GB |
| NVIDIA RTX 6000 Ada Generation | 48 GB |
| NVIDIA L40S | 48 GB |
| NVIDIA GeForce RTX 5090 | 32 GB |
Memory is not speed. Cards with less memory can be faster or slower on a model that fits on both, and we publish no speed comparison. What the table tells you is which models fit on a single card.
Rent an RTX PRO 6000 Blackwell server
Servers with 1, 2, 4 or 8 NVIDIA RTX PRO 6000 Blackwell SE GPUs in Frankfurt, Germany, from $1,951.09 a month. The price is for the whole server for a month: its GPUs, vCPU, RAM, disk, a 10 Gbps port and 20 TB of outbound transfer.
| GPUs | VRAM per GPU | Total VRAM | vCPU | RAM | Disk | Monthly | Stock and order |
|---|---|---|---|---|---|---|---|
| 1 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 96 GB | 16 | 128 GB | 500 GB | $1,951.09 | |
| 2 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 192 GB | 32 | 256 GB | 1,000 GB | $3,900.60 | |
| 4 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 384 GB | 64 | 512 GB | 2,000 GB | $7,799.63 | |
| 8 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 768 GB | 128 | 1,024 GB | 4,000 GB | $15,597.68 |
Stock follows free GPU capacity and is shown live: a row marked Out of stock cannot be ordered until capacity returns. If the plan you need is out of stock, ask the sales team and tell us which plan and how many servers you need. Servers run Ubuntu 24.04 LTS with NVIDIA CUDA and give you root access over SSH. Everything about ordering and billing is on the GPU servers page.
RTX PRO 6000 Blackwell FAQ
What is the NVIDIA RTX PRO 6000 Blackwell Server Edition?
It is NVIDIA's Blackwell-generation professional GPU for data center servers. NVIDIA specifies it with 96 GB of GDDR7 memory with error correction, 24,064 CUDA cores and a passive thermal design that relies on the server's airflow. NVIDIA positions it for AI inference, generative AI, rendering, 3D graphics and video.
How much VRAM does the RTX PRO 6000 Blackwell have?
96 GB per GPU. A RyzenRDP server has 1, 2, 4 or 8 of them, which is 96 GB to 768 GB of VRAM in total.
How much does it cost to rent an RTX PRO 6000 Blackwell?
A server with one NVIDIA RTX PRO 6000 Blackwell SE, with 16 vCPU, 128 GB RAM, 500 GB disk and a 10 Gbps port, costs $1,951.09 a month. Servers with more GPUs cost $3,900.60, $7,799.63 and $15,597.68 a month. Billing is monthly; there is no hourly rate. Over a 730-hour month that is about $2.67 per GPU-hour.
Can I rent an RTX PRO 6000 Blackwell by the hour?
No. Every GPU server is billed by the month. Spread over the 730 hours of an average month, the single-GPU server costs about $2.67 per GPU-hour and the eight-GPU server about $2.67 per GPU-hour, so it suits a workload that runs all month, such as an inference endpoint, rather than a job of a few hours.
What is the difference between the RTX PRO 6000 Blackwell and the RTX 6000 Ada?
Generation and memory. The RTX 6000 Ada Generation is built on NVIDIA's Ada Lovelace architecture, the generation before Blackwell, with 48 GB of GDDR6; the RTX PRO 6000 Blackwell Server Edition has 96 GB of GDDR7, both with error correction, by NVIDIA's figures. RyzenRDP rents the RTX PRO 6000 Blackwell only.
Is the RTX PRO 6000 Blackwell good for running LLMs?
Its 96 GB of memory is what makes it useful for language models: a 70-billion-parameter model fits on one GPU at 4-bit or 8-bit, and a 32-billion-parameter model fits unquantized at 16-bit. How fast a given model runs depends on the model, the serving engine and the load, and we publish no benchmark.
What is the difference between the RTX PRO 6000 Blackwell and the RTX 5090?
Memory, first of all: 96 GB against 32 GB. The RTX PRO 6000 Blackwell Server Edition is a passively cooled data center card, while the GeForce RTX 5090 is a consumer card for desktop PCs. A model that needs more than 32 GB of VRAM fits on one RTX PRO 6000 Blackwell and not on one RTX 5090.
Can I rent a server with several RTX PRO 6000 Blackwell GPUs?
Yes. Plans have 1, 2, 4 or 8 GPUs. The largest has 8 GPUs with 768 GB of VRAM in total and 128 vCPU, 1,024 GB RAM, 4,000 GB disk and a 10 Gbps port.
Do the servers come with CUDA?
Yes. GPU servers run Ubuntu 24.04 LTS with NVIDIA CUDA. Run nvidia-smi after you sign in to see the GPUs and their memory.
Where are the RTX PRO 6000 Blackwell servers located?
In Frankfurt, Germany.
Rent an RTX PRO 6000 Blackwell server
96 GB of VRAM per GPU, 1, 2, 4 or 8 GPUs per server, in Frankfurt, Germany. From $1,951.09 a month.