GPU servers · NVIDIA Blackwell
RTX PRO 6000 dedicated server with the full 96 GB GPU
An RTX PRO 6000 dedicated server from us is one whole machine with an NVIDIA RTX PRO 6000 Blackwell passed straight to your own operating system: 96 GB of GDDR7 with ECC, more than the 80 GB on an A100 or an H100. You rent it by the month, nobody else uses it, and the port is unmetered, so large models and video move in and out at no extra cost.
- CUDA cores
- 24,064
- Memory speed
- 1,792 GB/s
- Video encoders
- 4 × AV1
- Board power
- 600 W
Fifth-generation Tensor cores. Fourth-generation RT cores.
Have an RTX PRO 6000 that is not busy?
Put the machine on the Primcast marketplace and earn a monthly rent from people who need the compute.
00Marketplace
Or rent an RTX PRO 6000 from its owner
Owners list their machines on our marketplace and set the monthly price themselves. A listing held by our sister company, Server Room, is rented on serverroom.net, not here.
01What it is for
A 96GB GPU server for models, rendering and video
You rent the RTX PRO 6000 when the job needs a lot of memory on one GPU and also needs to draw or encode. Here is what that looks like in practice.
01Capacity
A large model on one GPU
A 70-billion-parameter model fits at 4-bit with room for its context, and a 30-billion model fits at 8-bit. Everything sits in one address space, so you skip the sharding and the second machine.
02Video
Four AV1 encoders on board
Four NVENC and four NVDEC engines, all with AV1. You can transcode many live channels in hardware and keep the processor free. The A100 and the H100 have no video encoder at all.
03Rendering
Ray tracing with room for big scenes
Fourth-generation RT cores render, and 96 GB holds scenes and textures that would overflow a 48 GB GPU. You can render, encode and run a model on the same board.
04One tenant
The full GPU, on your own system
No vGPU profile, no time-sharing, no other customer on the machine. Some cloud offers for this GPU are a virtual machine on a shared host, so compare like with like before you compare prices.
02Before you order
More memory than an H100. Here is what you give up
The RTX PRO 6000 holds more memory than the data center accelerators we fit. Two things separate it from them, and one of them may decide it for you.
No NVLink
This GPU has no NVLink bridge. Two of them in one machine are two separate 96 GB pools talking over the PCIe slot, not one 192 GB pool. If your model needs more than 96 GB in one address space, look at the A100 and H100, which pool over NVLink.
GDDR7, not stacked HBM
GDDR7 is quick, 1,792 GB/s, which is more than the A100 40 GB and twice the L40S. The A100 80 GB still reads a little faster at 1,935 GB/s. For serving many users at a large batch size, read speed sets the pace, so check that column if throughput is your goal.
If your question is “will the model fit on one GPU”, the RTX PRO 6000 answers it with the most room. If it is “how many tokens a second for hundreds of users”, an accelerator may be worth a look. The comparison table puts memory and monthly cost side by side.
03Specifications
RTX PRO 6000 Blackwell specs: GDDR7, Tensor and RT cores
Figures from NVIDIA’s product page for the Workstation Edition. The Max-Q Edition is the same chip with the same 96 GB at 300 W.
RTX PRO 6000 Blackwell 96GB
Blackwell
96 GB GDDR7 ECC
24,064
1,792 GB/s
600 W
PCIe Gen 5 x16
4 NVENC, 4 NVDEC, AV1
Yes
- 125 TFLOPS FP32
- 380 TFLOPS RT
- Tensor cores with FP4
- MIG, up to 4 instances
04Choosing
RTX PRO 6000 vs H100, A100 and RTX 6000 Ada
The large-memory GPUs we fit, with the memory and the monthly cost next to each other. Memory and price do not rise together on this shelf, and that is the reason to read it.
57 GPU models in the catalogue today · none holds more than 96 GB
| Card | Card memory | The card | Cheapest complete machine |
|---|---|---|---|
| A40 | 48 GB GDDR6 | $499 | $572.47 |
| RTX 6000 ADA | 48 GB GDDR6 ECC | $625 | $698.47 |
| A100 80GB | 80 GB HBM2e | $899 | $972.47 |
| RTX PRO 6000 Blackwell 96GB | 96 GB GDDR7 ECC | $1,299 | $1,382.67 |
| H100 80GB | 80 GB HBM2e | $1,599 | $1,682.67 |
The RTX 6000 Ada is the previous professional GPU, with 48 GB of GDDR6. The RTX PRO 6000 doubles that on newer silicon. If 48 GB is enough for your work, the Ada is the cheaper way in. For every GPU we fit, priced the same way, see all GPU servers with prices.
05The machine
Which server it goes in, and the PCIe link it gets
The RTX PRO 6000 is a PCIe Gen 5 GPU. Each machine below shows the link its slot gives it, so you can see what you get before you pick.
Intel Xeon Silver / Gold
AMD EPYC
Once the weights are loaded on the GPU, the slot sits mostly idle, which covers most inference, most rendering and all transcoding. A training loop that sends a fresh batch across the slot on every step is the job that notices. Tell us what you will run and we will point you to the right machine.
06Sizing
What fits in 96 GB of GPU memory
Weights only. Leave room for the context window and the runtime on top.
| Model | 16-bit | 8-bit | 4-bit |
|---|---|---|---|
| 7-8 billion parameters | 16 GB — fits | 8 GB — fits | 5 GB — fits |
| 13-14 billion parameters | 28 GB — fits | 14 GB — fits | 8 GB — fits |
| 30-34 billion parameters | 68 GB — tight | 34 GB — fits | 19 GB — fits |
| 70 billion parameters | 140 GB — does not fit | 70 GB — tight | 38 GB — fits |
A 70-billion model at 8-bit loads but leaves little room for a long context or a second user. At 4-bit it fits comfortably. To work it out from your model to the GPU, use the LLM memory sizing page.
07Prices
Rent an RTX PRO 6000: prices with the machine
The GPU is one line on your invoice and the server is another. Both come from the order catalogue the checkout uses, so each price below belongs to the machine beside it.
| Card | The machine it goes in | The card | Complete, per month |
|---|---|---|---|
| RTX PRO 6000 Blackwell 96GB | Intel Xeon Silver / Gold | $1,299 | $1,382.67 Configure this build → |
| RTX PRO 6000 Blackwell 96GB | AMD EPYC | $1,299 | $1,516.59 Configure this build → |
The port is unmetered in both directions with no egress charge. An IPv4 and an IPv6 address come with the server, you pay one month at a time, and there is no contract. A build carries a one-time setup fee, shown on the order before you pay.
08Delivery and help
How soon you get it, and who answers
It depends on where the GPU is right now. You are told which of these applies before you pay.
- About 30 minutes
Already racked
A machine with the GPU fitted is in a rack, and it is yours about 30 minutes after the order is approved.
- 4 to 24 hours
On our shelf
We fit a GPU we already hold into the server you chose.
- 5 to 10 working days
Bought in
We order the GPU from a distributor, then fit it to your server.
- 10 to 15 working days
On allocation
When distributors ration a GPU, we name the date before you commit.
Questions go to our chat. An AI assistant answers in seconds and can check your account and your server, and a person from our team joins whenever you ask. Machines racked and ready right now are on instant servers.
09Questions
Questions about renting an RTX PRO 6000
01Is the RTX PRO 6000 Blackwell the same as the RTX 6000 Ada?
No, although the names are close. The RTX 6000 Ada Generation is the earlier professional GPU, with 48 GB of GDDR6 and ECC. The RTX PRO 6000 Blackwell, sometimes searched as the RTX 6000 Pro, has 96 GB of GDDR7 with ECC on the newer Blackwell chip. We fit both, and the comparison table shows each one priced with its machine.
02Does the RTX PRO 6000 support NVLink?
No. This GPU has no NVLink connector, so two of them in one server work as two separate 96 GB pools linked through PCIe. If you need more than 96 GB pooled together, the A100 and H100 support NVLink.
03How much does it cost to rent an RTX PRO 6000?
The cheapest complete build, the GPU and the server together, is $1,382.67 a month today. The price table on this page lists every machine the GPU goes into, read live from the order catalogue, with a link to configure each one. There is no contract and the port is unmetered.
04Should I rent an RTX PRO 6000 or an H100?
Rent the RTX PRO 6000 when the model has to fit on one GPU, or when the work also renders or encodes video. Rent an H100 when you need NVLink to pool several GPUs, or HBM memory for serving at large batch sizes. The comparison table shows the memory and the monthly cost of both.
05Is the GPU shared or split into virtual GPUs?
No. You get the full GPU passed through to your own operating system, on a machine nobody else uses: no hypervisor, no vGPU profile, no time-sharing. The GPU supports MIG with up to four instances, so you can split it yourself if your work suits that.
06Which NVIDIA driver and CUDA version will I get?
The ones you choose. The server comes with a clean operating system and root access, and you install and pin the driver, CUDA and framework versions your code needs. Nothing upgrades behind your back.
07Can I have two RTX PRO 6000 GPUs in one server?
Yes, as a quoted build rather than a checkout option. At up to 600 W each, the chassis and its power budget decide how many fit, and which edition goes in. Tell us what you plan to run and we reply with the chassis, the count and the lead time. For larger setups, see multi-GPU and multi-node servers.
08How soon can I have an RTX PRO 6000 server?
About 30 minutes after approval if a racked machine already carries the GPU, 4 to 24 hours if we hold one, 5 to 10 working days if we buy it in, and 10 to 15 working days when it is on allocation. You are told which applies before you pay. Racked machines are listed on instant servers.
Pick your machine, then add the GPU
Each RTX PRO 6000 build is priced above with its server. Choose one, configure it, and the order shows the full monthly total before you pay.