GPU server · RTX 30 series

RTX 3080 dedicated server plus four more RTX 30 GPUs, 8 GB to 24 GB

An RTX 3080 dedicated server from Primcast is a whole physical machine with an NVIDIA GeForce RTX 3080 handed to your own operating system: 10 GB GDDR6X and 8,704 CUDA cores, with no virtual machine and no neighbours. If your work needs more memory, our servers also take an RTX 3080 Ti or a 24 GB RTX 3090. You pay by the month, with no contract.

Build your RTX 3080 server See all five with prices
geforce-rtx-30 Ampere
  • RTX 30708 GB GDDR65,888
  • RTX 3070 Ti8 GB GDDR6X6,144
  • RTX 308010 GB GDDR6X8,704
  • RTX 3080 Ti12 GB GDDR6X10,240
  • RTX 309024 GB GDDR6X10,496

All five: PCIe Gen 4, one video encoder, no ECC memory.

01Pick one

Which RTX 30 series server fits your job

Start at the top of this table and stop at the first GPU with enough memory for what you load. The GPUs further down do the same work faster. The only thing they add is room for a bigger model or scene.

Cheapest complete machine first. Every figure read from the order catalogue when this page was served.
CardCard memoryThe cardCheapest complete machine
RTX 30708 GB GDDR6$139$209.93
RTX 3070 Ti8 GB GDDR6X$149$222.47
RTX 308010 GB GDDR6X$159$229.93
RTX 3080 Ti12 GB GDDR6X$199$272.47
RTX 309024 GB GDDR6X$239$309.93

Our RTX 3080 is the 10 GB model

NVIDIA made two RTX 3080s. Ours is the original, with 10 GB GDDR6X and 8,704 CUDA cores. The later version has 12 GB and 8,960 cores, and plenty of listings quote 12 GB for the cheaper GPU. Size your work against 10 GB. A 13-billion-parameter model at 8-bit needs about 14 GB, so it fits neither version.

The RTX 3090 is the big step

Going from the RTX 3080 Ti to the RTX 3090 doubles the memory, from 12 GB to 24 GB GDDR6X. It is the only RTX 30 GPU with 24 GB. If your model or scene needs that much, nothing cheaper here will run it. For more than 24 GB, see every GPU we fit, with prices.

02What it is for

What people run on an RTX 3080 GPU server

These GPUs came out in 2020 and we still rent them, because a lot of real work does not need anything newer. For these four jobs an older GPU gets it done for less each month.

01AI

Running a model that already fits

Serving a 7B model, testing prompts or fine-tuning something small. A 7-8B model loads at 4-bit on any of the five, and at 8-bit on the RTX 3080 with little room to spare. The RTX 3090 holds a 13B model at 8-bit with room left for context.

02Games

Game servers that need a real GPU

Games are written and tested on GeForce GPUs like these. If your game server renders or streams, the RTX 30 series runs the same GeForce drivers your players have at home.

03Rendering

Batch renders you can restart

Blender, Unreal or your own pipeline, frame after frame. If a job fails you run it again, so you do not need to pay for error-correcting memory. Spend the money on CUDA cores instead.

04Video

H.264 and HEVC encoding

Each GPU has one hardware encoder and one decoder for H.264 and HEVC. That covers live channels or a queue of uploads without loading the processor. For AV1, or many channels on one machine, see live video transcoding servers.

03Before you order

Two things the RTX 30 series cannot do

Both are in NVIDIA's own specification for these GPUs. Check them against your job before you order.

No AV1 encoding

None of the five can encode AV1 in hardware. NVIDIA added that with the RTX 40 series. If your pipeline sends AV1, choose an RTX 4090 D server or an L40S server, which has three AV1 encoders.

No ECC memory, no professional driver

A flipped bit goes unnoticed. That is fine for work you can check or repeat. For a training run that lasts weeks without checkpoints, or for software licensed only on professional GPUs, pick one from the workstation range on the full GPU list.

04Specification

RTX 3090, 3080 and 3070 specifications

Ampere architecture, second-generation RT cores, third-generation Tensor cores and PCIe Gen 4 on all five. The figures are NVIDIA's, from its GeForce comparison pages.

RTX 3090

Architecture

Ampere

Card memory

24 GB GDDR6X

CUDA cores

10,496

Memory bandwidth

936 GB/s

Board power

350 W

Card interface

PCIe Gen 4 x16

Hardware video encoders

1 NVENC, 1 NVDEC, no AV1 encode

Error-correcting memory

No

RTX 3080 Ti

Architecture

Ampere

Card memory

12 GB GDDR6X

CUDA cores

10,240

Memory bandwidth

912 GB/s

Board power

350 W

Card interface

PCIe Gen 4 x16

Hardware video encoders

1 NVENC, 1 NVDEC, no AV1 encode

Error-correcting memory

No

RTX 3080

Architecture

Ampere

Card memory

10 GB GDDR6X

CUDA cores

8,704

Memory bandwidth

760 GB/s

Board power

320 W

Card interface

PCIe Gen 4 x16

Hardware video encoders

1 NVENC, 1 NVDEC, no AV1 encode

Error-correcting memory

No

RTX 3070 Ti

Architecture

Ampere

Card memory

8 GB GDDR6X

CUDA cores

6,144

Memory bandwidth

608 GB/s

Board power

290 W

Card interface

PCIe Gen 4 x16

Hardware video encoders

1 NVENC, 1 NVDEC, no AV1 encode

Error-correcting memory

No

RTX 3070

Architecture

Ampere

Card memory

8 GB GDDR6

CUDA cores

5,888

Memory bandwidth

448 GB/s

Board power

220 W

Card interface

PCIe Gen 4 x16

Hardware video encoders

1 NVENC, 1 NVDEC, no AV1 encode

Error-correcting memory

No

05Model sizing

Will your model fit on an RTX 3080 or RTX 3090?

On the RTX 3080's 10 GB, a 7-8B model fits at 4-bit and is tight at 8-bit. On the RTX 3090's 24 GB, a 13B model fits at 8-bit and a 30B model loads at 4-bit, with little room left for context.

RTX 3080, 10 GB

Weights only, rounded up. Fits means the card’s 10 GB holds the weights 1.5× over — room for the KV cache, the activations and the runtime, and the only verdict the LLM sizing page recommends a card on. Tight clears 1.25× only: the model loads, and a long context or a second user runs the card out.
Model16-bit8-bit4-bit
7-8 billion parameters16 GB — does not fit8 GB — tight5 GB — fits
13-14 billion parameters28 GB — does not fit14 GB — does not fit8 GB — tight
30-34 billion parameters68 GB — does not fit34 GB — does not fit19 GB — does not fit
70 billion parameters140 GB — does not fit70 GB — does not fit38 GB — does not fit

RTX 3090, 24 GB

Weights only, rounded up. Fits means the card’s 24 GB holds the weights 1.5× over — room for the KV cache, the activations and the runtime, and the only verdict the LLM sizing page recommends a card on. Tight clears 1.25× only: the model loads, and a long context or a second user runs the card out.
Model16-bit8-bit4-bit
7-8 billion parameters16 GB — fits8 GB — fits5 GB — fits
13-14 billion parameters28 GB — does not fit14 GB — fits8 GB — fits
30-34 billion parameters68 GB — does not fit34 GB — does not fit19 GB — tight
70 billion parameters140 GB — does not fit70 GB — does not fit38 GB — does not fit

The RTX 3080 Ti sits between the two with 12 GB. The RTX 3070 and 3070 Ti have 8 GB, which holds a 7-8B model at 4-bit and not at 8-bit. Starting from the model instead? LLM server sizing works out how much GPU memory a model needs and which GPU clears it.

06The server around it

Which server each RTX 30 GPU goes in

Every RTX 30 GPU is a PCIe Gen 4 part. A Gen 4 slot gives it the full link. A Gen 3 slot gives it half, and the list below shows which server has which.

Intel Xeon E5-2600 v1/v2

RTX 3090, RTX 3080, RTX 3070

Slot: PCIe 3.0, 40 lanes per socket. Card: PCIe Gen 4.

About half the bus the card was designed for. That costs a training loop streaming batches off the host; it costs an inference server or a transcoder nothing, because the weights are already on the card.

Intel Xeon E5-2600 v3/v4

RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3070

Slot: PCIe 3.0, 40 lanes per socket. Card: PCIe Gen 4.

About half the bus the card was designed for. That costs a training loop streaming batches off the host; it costs an inference server or a transcoder nothing, because the weights are already on the card.

Intel Xeon Silver / Gold

RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3070

Slot: PCIe 3.0, 48 lanes per socket. Card: PCIe Gen 4.

About half the bus the card was designed for. That costs a training loop streaming batches off the host; it costs an inference server or a transcoder nothing, because the weights are already on the card.

AMD EPYC

RTX 3080

Slot: PCIe 4.0, 128 lanes per socket. Card: PCIe Gen 4.

Full width. The slot is at least the generation the card was designed for, so nothing is left on the table.

For most RTX 30 work the slot hardly matters. Once a model or a scene is loaded on the GPU, very little crosses the slot. It matters for training that streams new batches from the server on every step.

07Compare

RTX 3090 vs RTX 4090 D and RTX 5090

The RTX 3090 has the same 24 GB as the RTX 4090 D, one generation later, and the RTX 5090 two generations on has 32 GB. This table shows what the newer GPUs cost, each with the cheapest complete machine it goes in.

Cheapest complete machine first. Every figure read from the order catalogue when this page was served.
CardCard memoryThe cardCheapest complete machine
RTX 309024 GB GDDR6X$239$309.93
RTX 4090D24 GB GDDR6X$299$372.47
RTX 509032 GB GDDR7$399$472.47

Choose the newer GPU when speed per job or AV1 encoding is what you need. When memory is what you are paying for, the RTX 3090 usually does the job. Why we still rent older hardware explains the thinking.

08Prices

RTX 3080 dedicated server prices, and the other four

The GPU is one line on your invoice and the server is another. Each row shows the full machine, what the GPU adds and the monthly total, read from the same catalogue the checkout uses.

Read from the order catalogue when this page was served. The card is a separate line; the machine does not get dearer because of what is plugged into it.
CardThe machine it goes inThe cardComplete, per month
RTX 3070Intel Xeon E5-2600 v1/v2
Intel Xeon E5-2630L Hex Core 2.00 GHz · 32 GB DDR3 · 2 × SATA 500 GB (RAID 1) · 500 Mbps · PCIe 3.0		$139$209.93
Configure this build →
RTX 3070Intel Xeon E5-2600 v3/v4
Intel Xeon E5-2620 v4 Octo Core 2.10 GHz · 32 GB DDR4 · 2 × SATA 500 GB (RAID 1) · 1 Gbps · PCIe 3.0		$139$212.47
Configure this build →
RTX 3070 TiIntel Xeon E5-2600 v3/v4
Intel Xeon E5-2620 v4 Octo Core 2.10 GHz · 32 GB DDR4 · 2 × SATA 500 GB (RAID 1) · 1 Gbps · PCIe 3.0		$149$222.47
Configure this build →
RTX 3070Intel Xeon Silver / Gold
Intel Xeon Silver 4110 8 Core 2.10 GHz · 16 GB DDR4 · 2 × SATA-SSD 240 GB (RAID 1) · 1 Gbps · PCIe 3.0		$139$222.67
Configure this build →
RTX 3080Intel Xeon E5-2600 v1/v2
Intel Xeon E5-2630L Hex Core 2.00 GHz · 32 GB DDR3 · 2 × SATA 500 GB (RAID 1) · 500 Mbps · PCIe 3.0		$159$229.93
Configure this build →
RTX 3080Intel Xeon E5-2600 v3/v4
Intel Xeon E5-2620 v4 Octo Core 2.10 GHz · 32 GB DDR4 · 2 × SATA 500 GB (RAID 1) · 1 Gbps · PCIe 3.0		$159$232.47
Configure this build →
RTX 3070 TiIntel Xeon Silver / Gold
Intel Xeon Silver 4110 8 Core 2.10 GHz · 16 GB DDR4 · 2 × SATA-SSD 240 GB (RAID 1) · 1 Gbps · PCIe 3.0		$149$232.67
Configure this build →
RTX 3080Intel Xeon Silver / Gold
Intel Xeon Silver 4110 8 Core 2.10 GHz · 16 GB DDR4 · 2 × SATA-SSD 240 GB (RAID 1) · 1 Gbps · PCIe 3.0		$159$242.67
Configure this build →
RTX 3080 TiIntel Xeon E5-2600 v3/v4
Intel Xeon E5-2620 v4 Octo Core 2.10 GHz · 32 GB DDR4 · 2 × SATA 500 GB (RAID 1) · 1 Gbps · PCIe 3.0		$199$272.47
Configure this build →
RTX 3080 TiIntel Xeon Silver / Gold
Intel Xeon Silver 4110 8 Core 2.10 GHz · 16 GB DDR4 · 2 × SATA-SSD 240 GB (RAID 1) · 1 Gbps · PCIe 3.0		$199$282.67
Configure this build →
RTX 3090Intel Xeon E5-2600 v1/v2
Intel Xeon E5-2630L Hex Core 2.00 GHz · 32 GB DDR3 · 2 × SATA 500 GB (RAID 1) · 500 Mbps · PCIe 3.0		$239$309.93
Configure this build →
RTX 3090Intel Xeon E5-2600 v3/v4
Intel Xeon E5-2620 v4 Octo Core 2.10 GHz · 32 GB DDR4 · 2 × SATA 500 GB (RAID 1) · 1 Gbps · PCIe 3.0		$239$312.47
Configure this build →
RTX 3090Intel Xeon Silver / Gold
Intel Xeon Silver 4110 8 Core 2.10 GHz · 16 GB DDR4 · 2 × SATA-SSD 240 GB (RAID 1) · 1 Gbps · PCIe 3.0		$239$322.67
Configure this build →
RTX 3080AMD EPYC
AMD EPYC 7413 24 CORE 2.65 GHz 128MB L3 CACHE · 32 GB DDR4 · 1 × SATA-SSD 240 GB · 1 Gbps · PCIe 4.0		$159$376.59
Configure this build →
  • PortUnmetered traffic in both directions, on a port that serves your machine alone.
  • AddressOne IPv4 address included.
  • TermMonthly by default, with no contract.
  • HelpOur chat assistant answers in seconds and can check your account and server. Ask for a person any time.

Need a machine today? Instant servers lists what is already racked. To compare every GPU we fit, see GPU servers, all priced.

09Questions

RTX 3080 server questions people ask before ordering

01Is your RTX 3080 the 10 GB or the 12 GB version?

It is the 10 GB version, the original RTX 3080 with 8,704 CUDA cores. The 12 GB version has 8,960 cores, and we do not fit it. If you need 12 GB, the RTX 3080 Ti has it.

02RTX 3080 or RTX 3090 server?

Pick the RTX 3080 if your model or scene fits in 10 GB. Between 10 and 12 GB, take the RTX 3080 Ti. Pick the RTX 3090 if it needs more than 12 GB: it has 24 GB GDDR6X, the most of any RTX 30 GPU, and more CUDA cores. Both are in the price table above, each with the complete machine.

03Is the GPU shared with anyone else?

No. You are the only tenant on the physical machine, and the GPU goes straight to your own operating system. There is no hypervisor, no partition and no time-slicing. You install the driver and CUDA version you want, and nothing updates without you.

04Can I put two RTX 30 GPUs in one server?

Yes, as a build we quote rather than a checkout option, because it depends on the chassis, the width of the GPU and the power budget. Ask in chat and we name the chassis and the lead time. Two GPUs are two separate memory pools: a model that does not fit on one RTX 3090 has to be split across both by your software.

05How soon can I have an RTX 3080 server?

It depends on where the GPU is, and you are told before you pay. A GPU we already hold is fitted within four to twenty-four hours. One we have to buy in takes five to ten working days. To start today, instant servers lists the machines already racked.

06What happens if the GPU fails?

We replace it. The agreement gives us four hours from the moment you report it by phone or live chat, and past that you are owed service credit on a published schedule.

07Who helps me if something goes wrong?

Open the chat on any page. Our assistant replies in seconds, can look at your account and your server, and passes you to one of our team whenever you ask. You keep full root access to the machine the whole time.

Get an RTX 3080 server for your project

Pick the server, add the RTX 3080 or any of the other four, and see the monthly total before you pay. No contract, and the port is unmetered.

Build your RTX 3080 server Compare every GPU