GPU servers with NVIDIA A40

Servers with NVIDIA A40 (48 GB with ECC, Ampere) for large-model inference, rendering and virtual workstations, as a full card or a vGPU slice. Compare VRAM, CPU and price, cheapest first.

Filters 1
Processor
Disk
Disk type
Size, GB
RAM
Traffic per month
at least
Operating system
Country
Payment method
Virtualization
Terms
Found 10 plans — showing , first:
Host and plan ProcessorRAM Disk Traffic Where Payment Price
BitHost BitHostVultr GPU NVIDIA A40 · 1 vCPU / 3 GB / 40 GB
GPU NVIDIA A40, 1 GB Hourly billing No ID required Ready in minutes
1 coreCPU 3 GBRAM 40 GBSSD 2 TBtraffic not specified Cryptocurrency
60.00 $ Open
Hosthink HosthinkHT-C-013 Cloud GPU
GPU NVIDIA A40, 2 GB DDoS protection IPv6
1 coreCPU 5 GBRAM 90 GBSSD 3 TB10 Gbit/s Bank card (Visa/Mastercard)Invoice for companies
110 $ Open
Mag Host Mag Hostgpu.v36-a40-48
GPU NVIDIA RTX A40, 48 GB Money-back Ready in minutes IPv6
36 coresCPU 128 GBRAM 500 GBNVMe 1 Gbit/sport Romania CryptocurrencyBank card (Visa/Mastercard)PayPalInvoice for companies
397 $ when paying for 2 years: 310 $/⁠mo Open
H Hapih Host1xA40
GPU NVIDIA A40, 48 GB DDoS protection
8 coresCPU 32 GBRAM 500 GBSSD 2 TBtraffic India Bank card (Visa/Mastercard)
434 $ when paying for year: 302 $/⁠mo Open
Infinitive Host Infinitive HostGPU Chakra
GPU NVIDIA A40, 48 GB Money-back DDoS protection
16 coresCPU 100 GBRAM 250 GBSSD —traffic India Bank card (Visa/Mastercard)Invoice for companies
584 $ Open
MaxCloudON MaxCloudONGPU Server MAX9
GPU NVIDIA RTX A40 ×2, 48 GB Money-back
48 coresCPU 128 GBRAM 500 GBNVMe unlimitedtraffic Bulgaria CryptocurrencyBank card (Visa/Mastercard)Invoice for companies
980 $ Open
IPHOST IPHOSTGPU A40 48GB
GPU NVIDIA A40, 48 GB No ID required
64 coresCPU 256 GBRAM 7680 GBNVMe —traffic Moldova CryptocurrencyBank card (Visa/Mastercard)Invoice for companies
1,060 $ total when paying for 3 mo Open
Vultr VultrCloud GPU 1× NVIDIA A40 24 vCPU / 120 GB (vcg-a40-24c-120g-48vram)
GPU NVIDIA A40, 48 GB Hourly billing DDoS protection IPv6
24 coresCPU 120 GBRAM 1400 GBSSD 15 TBtraffic CryptocurrencyBank card (Visa/Mastercard)PayPal
1,250 $ Open
0xcloud 0xcloudvcg-a40-24c-120g-48vram
GPU NVIDIA A40, 48 GB No ID required Ready in minutes IPv6
24 coresCPU 120 GBRAM 1400 GBSSD 15360 GBtraffic Cryptocurrency
1,688 $ Open
SpaceWeb SpaceWebIntel Xeon Gold 5320 / DDR4 1024Gb / 2x4TB HDD, 2x2TB NVME / 1xTesla A40 48GB
GPU NVIDIA A40, 48 GB Trial Hourly billing
52 cores2.2 GHz 1024 GBRAM 4 TBNVMe unlimitedtraffic Russia Payment methods are not listed on the host’s website
2,138 $ Open

GPU servers with NVIDIA A40: prices, hosting providers and payment methods

The selection has 30 plans from 10 hosts. Prices start at 60.00 $ per month, a quarter of plans are cheaper than 420 $, half are cheaper than 980 $. NVMe: 17% of plans, unlimited traffic: 23%, hourly billing: 37%.

Cheapest hosting providers

  1. BitHostVultr GPU NVIDIA A40 · 1 vCPU / 3 GB / 40 GB · 1 core · 3 GB60.00 $
  2. HosthinkHT-C-013 Cloud GPU · 1 core · 5 GB110 $
  3. Mag Hostgpu.v36-a40-48 · 36 cores · 128 GB397 $
  4. Hapih Host1xA40 · 8 cores · 32 GB434 $
  5. Infinitive HostGPU Chakra · 16 cores · 100 GB584 $

Price by RAM size

RAMPlansPrice from
from 1 GB3060.00 $
from 2 GB3060.00 $
from 4 GB29110 $
from 8 GB27210 $
from 16 GB25397 $

Payment methods

  • Bank card7 70%
  • Cryptocurrency6 60%

How many hosting providers in the selection accept each payment method.

Where most plans are

Provider comparisons VPS price index VPS price calculator VPS guide

Server with NVIDIA A40: 48 GB for inference, rendering and virtual workstations

The NVIDIA A40 is an Ampere data-centre card. According to the NVIDIA specification, it has 48 GB of GDDR6 with error correction (ECC) and 696 GB/s of bandwidth, 10,752 CUDA cores, passive cooling and a power draw of up to 300 W. Two cards can be joined with NVLink (112.5 GB/s), giving 96 GB as a pair. The card supports NVIDIA vGPU, so hosts offer it both as a whole card and in slices.

What it is good for

  • Language models that do not fit in 24 GB: 30–34B parameters in 8-bit, 70B in 4-bit quantization with a short context.
  • Rendering and 3D: second-generation RT Cores and plenty of memory for heavy scenes.
  • Virtual workstations for CAD and video editing via vGPU profiles.

The A40's memory is slower than on gaming cards: when generating language model output, the RTX 4090 with its 1 TB/s is faster, as long as the model fits in 24 GB. A newer and faster card in the same 48 GB class is the L40S.

Whole card or vGPU slice

The cheapest rows in the table are slices of one card with 2, 4 or 8 GB of video memory. They suit a remote workstation and light graphics, but not language models. Inference and rendering need a whole 48 GB card, so check the memory size in the plan. The A40 has no MIG: it can only be shared through vGPU.

Monthly and hourly cost

In the catalogue as of 17 November 2026, the median price of a server with one whole A40 is about $1,050 a month, with half of the plans between $600 and $1,100. Over 730 hours that comes to about $1.44 an hour. There are both cloud plans and dedicated servers; only a few hosts offer hourly billing. If you need the card for a few hours, compare hourly plans and read how to work out the cost of an hour. How much memory a model takes is covered in the article on running your own language model on a server.

FAQ

8 questions with answers

How is the A40 different from the L40S?

Both have 48 GB, but the L40S is newer (Ada Lovelace), faster in compute, and supports FP8 and hardware AV1 encoding. The A40 is usually cheaper and can be paired with another card via NVLink.

What does a plan with an A40 and 4 GB of video memory mean?

It is a slice of one card via NVIDIA vGPU. It is enough for a remote workstation and graphics, while neural networks need a whole 48 GB card.

Can I run a 70B model on a single A40?

Only in 4-bit quantization: the weights take about 40 GB, leaving little room for context. For a service under load, two cards linked with NVLink are better.

Which CUDA version does the A40 need?

The card has compute capability 8.6 and works with CUDA 11.1 and later. On vGPU slices the driver is installed by the host.

How much does GPU servers with NVIDIA A40 cost?

From 60.00 $ per month. A quarter of plans are cheaper than 420 $, half are cheaper than 980 $. In total 30 plans from 10 hosts.

Can I pay for GPU servers with NVIDIA A40 by bank card or cryptocurrency?

By bank card at 7 hosts of 10 (70%), by cryptocurrency at 6 (60%). Payment methods for each plan are in the Payment column.

How much does GPU servers with NVIDIA A40 with 2 GB of RAM cost?

From 60.00 $ per month; with 4 GB from 110 $, with 8 GB from 210 $.

Which countries have the most plans?

Most plans are in in India, in the UK and in the USA. You can pick a country in the filter, and prices and payment methods will update.

Related

Guide
GPU Server: Which Graphics Card for Which Job — Memory, Bandwidth and PricesYour Own Language Model on a Server: How Much Video Memory and Which GPU You Need

Prices and specs are taken from hosting providers' websites; check them on the website before paying, as hosts change plans without notice. About the catalog and methodology