Denvr

gpuperhour.com · Compute · 5 records · JSON

Atlantic.net Computepaidemail?card?R
Cirrascale Computepaidemail?card?R
Denvr Computepaidemail?card?R
Lyceum Computepaidemail?card?R
QuantaCloud Computepaidemail?card?R

Atlantic.net Compute

paidemail?card?R

gpuperhour.com/providers/atlanticnet · GPU cloud · compute/atlantic-net · JSON · compare

Free tier	USD 1.67 per GPU-hour for an NVIDIA L40S with 48 GB: 32 vCPU, 192 GB RAM, 1400 GB storage, in Iowa (a second row at the same price carries the region 'United States'). The next rows are H100 NVL 94 GB at USD 3.94 per GPU-hour with 28 vCPU, 240 GB RAM and 2400 GB storage in Virginia, and the same card at the same price with 1400 GB of storage. Four offers in total. unit hours; metered on wall-clock; none
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not published on the page read: the index lists GPU, VRAM, host specs, region, price and status only
Identity	email (placeholder) not stated
Binding	none No quota of any kind is stated on the page: the index publishes GPU model, VRAM, host specs, region, price and status, and no transfer allowance or port speed. The wall is the hourly rate.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated on the page read
Composes with	USD 1.67 an hour for a 48 GB L40S with 192 GB of system RAM is the cheapest 48 GB card, and an L40S at 48 GB holds a 70B model at 4-bit with room for the KV cache. The same provider appears on the VPS indexes as a small VPS host, so one account spans both a CPU box and a GPU box
Reliability	no reliability event recorded
Notes	Found on gpuperhour, which tracked it; it also appears on vpsbenchmarks as atlantic_net. Not in dossier/compute (searched: atlantic.net and atlanticnet, no record; the one grep hit is dossier/compute/hashbang.json, which mentions the name). The two L40S rows differ only in region label, so the city of the first is not asserted. would-settle: the vendor's GPU pricing page for the billing unit and the transfer terms.
Evidence rung	reported
Checked	2026-10-22 (live); stale on 2026-11-05

Evidence (2)

date	source	what it proved
2026-10-22	gpuperhour.com/providers/atlanticnet	row fields, transcribed and joined by me (unquoted values are not verbatim): Atlantic.net \| NVIDIA L40S 48GB VRAM \| 48GB \| 32 vCPU 192GB RAM 1400GB Storage \| Iowa \| $1.67/GPU/hr; a second identical row carries 'United States'
2026-10-22	gpuperhour.com/providers/atlanticnet	further rows, transcribed and joined by me: NVIDIA H100 NVL 94GB VRAM \| 94GB \| 28 vCPU 240GB RAM 2400GB Storage \| Virginia \| $3.94/GPU/hr and the same card at 1400 GB storage; the page header reads '4 offers available'

More fields (1)

{
 "wave2": {
  "class": null,
  "free_tier_kind": "none",
  "shape": {
   "vcpu": 32,
   "ram_gib": 192,
   "disk_gib": 1400,
   "gpu_class": "NVIDIA L40S 48GB",
   "gpu_count": 1
  }
 }
}

Unknown or withheld values (9)

$.egress.metered egress-unknown
$.identity_cost.detail not-stated
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable

Cirrascale Compute

paidemail?card?R

gpuperhour.com/providers/cirrascale · GPU cloud · compute/cirrascale · JSON · compare

Free tier	USD 0.27 per GPU-hour for an 8-GPU NVIDIA RTX A4000 node with 16 GB per card: 40 vCPU, 256 GB RAM, 2610 GB storage, in the United States. The 8x A5000 24 GB node is USD 0.41 per GPU-hour on the same host shape, and the page lists 11 GPU models across 49 instances in one region. The provider is described as 'An AI Innovation Cloud targeting deep learning and HPC research with dedicated performance on non-virtualized hardware', established 2010. unit hours; metered on wall-clock; none
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not published on the page read
Identity	email (placeholder) not stated
Binding	none No quota is stated on the page: price, VRAM, host specs, region and status are all it publishes. The practical wall is the shape, because the cheapest rate is on an eight-card node and the row's total is USD 2.16 an hour.
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	not stated on the page read
Composes with	USD 0.27 per GPU-hour is the cheapest GPU rate of any provider, on an A4000 with 16 GB and a 40-vCPU host, which is a batch-inference shape: many small workers with a big CPU headroom and no need for a large card. The eight-card minimum is what turns that rate into USD 2.16 an hour, so it suits a queue that keeps eight workers busy and wastes money on a single notebook session
Reliability	no reliability event recorded
Notes	Found via gpuperhour, which listed it among 28 extractable provider slugs. Not in dossier/compute (searched: cirrascale, no record; the one grep hit is dossier/compute/graphcore-ipu-cloud.json, which mentions the name). would-settle: the vendor's own pricing page, since every figure here is third-party, and whether a single A4000 can be booked rather than only a full node.
Evidence rung	reported
Checked	2026-10-22 (live); stale on 2026-11-05

Evidence (2)

date	source	what it proved
2026-10-22	gpuperhour.com/providers/cirrascale	row fields, transcribed and joined by me (unquoted values are not verbatim): Cirrascale \| 8xNVIDIA RTX A4000 16GB VRAM \| 16GB \| 40 vCPU 256GB RAM 2610GB Storage \| United States \| $0.27/GPU/hr $2.16/hr total (8x)
2026-10-22	gpuperhour.com/providers/cirrascale	page, transcribed and joined by me (unquoted values are not verbatim): 'An AI Innovation Cloud targeting deep learning and HPC research with dedicated performance on non-virtualized hardware', '49 Instances', '1 Regions', '49 Available', and 'Cirrascale offers 11 GPU models'; the A5000 row reads $0.41/GPU/hr $3.28/hr total (8x)

More fields (1)

{
 "wave2": {
  "class": null,
  "free_tier_kind": "none",
  "shape": {
   "vcpu": 40,
   "ram_gib": 256,
   "disk_gib": 2610,
   "gpu_class": "NVIDIA RTX A4000 16GB",
   "gpu_count": 8
  }
 }
}

Unknown or withheld values (8)

$.egress.metered egress-unknown
$.identity_cost.detail not-stated
$.identity_cost.kind identity-defaulted
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable

Denvr Compute

paidemail?card?R

gpuperhour.com/providers/denvr · GPU cloud · compute/denvr · JSON · compare

Free tier	USD 0.65 per GPU-hour for a 4-GPU NVIDIA A40 node with 48 GB per card: 128 vCPU, 512 GB RAM, 7600 GB storage, in Virginia. Next rows: 8x Intel Gaudi 2 at USD 1.10 per GPU-hour with 160 vCPU and 1024 GB; 4x A100 PCIe 80 GB at USD 1.15 with 64 vCPU and 512 GB; 8x A100 SXM4 80 GB at USD 1.15 with 128 vCPU and 1024 GB. Seven offers in total. unit hours; metered on wall-clock; none
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not published on the page read
Identity	email (placeholder) not stated
Binding	none No quota is stated on the page: it publishes price, VRAM, host specs, region and status. The wall is the node shape, because the cheapest rate is a four-card node at USD 2.60 an hour total.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated on the page read
Composes with	The Gaudi 2 row at USD 1.10 per GPU-hour with 96 GB of accelerator memory is the cheapest large-memory device and the only non-NVIDIA entry besides Sesterce's AMD lines, so it is the row that matters if the workload can use a non-CUDA toolchain. 30 TB of local storage on that node also makes it a candidate for a job that reads a large corpus once
Reliability	no reliability event recorded
Notes	Found via gpuperhour. Not in dossier/compute (searched: denvr, no record; the one grep hit is dossier/compute/shadeform.json, which mentions the name). would-settle: the vendor's own pricing page for the billing unit and whether a single A40 can be booked.
Evidence rung	reported
Checked	2026-10-22 (live); stale on 2026-11-05

Evidence (2)

date	source	what it proved
2026-10-22	gpuperhour.com/providers/denvr	row fields, transcribed and joined by me (unquoted values are not verbatim): Denvr \| 4xNVIDIA A40 48GB VRAM \| 48GB \| 128 vCPU 512GB RAM 7600GB Storage \| Virginia \| $0.65/GPU/hr $2.60/hr total (4x)
2026-10-22	gpuperhour.com/providers/denvr	further rows, transcribed and joined by me (unquoted values are not verbatim): 8xIntel Gaudi 2 96GB VRAM \| 96GB \| 160 vCPU 1024GB RAM 30400GB Storage \| Virginia \| $1.10/GPU/hr; 4xNVIDIA A100 PCIe 80GB \| 80GB \| 64 vCPU 512GB RAM 7600GB Storage \| Virginia \| $1.15/GPU/hr; '7 offers available'

More fields (1)

{
 "wave2": {
  "class": null,
  "free_tier_kind": "none",
  "shape": {
   "vcpu": 128,
   "ram_gib": 512,
   "disk_gib": 7600,
   "gpu_class": "NVIDIA A40 48GB",
   "gpu_count": 4
  }
 }
}

Unknown or withheld values (9)

$.egress.metered egress-unknown
$.identity_cost.detail not-stated
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable

Lyceum Compute

paidemail?card?R

gpuperhour.com/providers/lyceum · GPU cloud · compute/lyceum · JSON · compare

Free tier	USD 1.99 per GPU-hour for a single NVIDIA L40S with 48 GB: 12 vCPU, 128 GB RAM, 100 GB storage, in Europe. The same per-GPU rate holds at 2x, 4x and 8x, with the 8x configuration marked Sold Out and the 1x and 2x marked Available; an A100 PCIe 80 GB follows at USD 2.29 per GPU-hour on a 16 vCPU, 120 GB host. Twenty-four offers in total. unit hours; metered on wall-clock; none
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not published on the page read
Identity	email (placeholder) not stated
Binding	none No quota is stated on the page: price, VRAM, host specs, region and status are all it publishes. The page's own differentiator is per-second billing and a spot book exposed through an API, neither of which is a quota.
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	the about block states 'On-demand and spot price book exposed through the API' and 'Email and webhook GPU availability r', the sentence truncated in the extract; no API documentation URL is given
Composes with	The 100 GB disk is small for a training job, so checkpoints have to go elsewhere
Reliability	no reliability event observed the 8x configuration of the same card was Sold Out while the 1x and 2x were Available on the read, which is the kind of movement the index records
Notes	Found via gpuperhour. Not in dossier/compute (searched: lyceum, no record). would-settle: the vendor's own price book through its API, which the about block says exists but does not link.
Evidence rung	reported
Checked	2026-10-22 (live); stale on 2026-11-05

Evidence (2)

date	source	what it proved
2026-10-22	gpuperhour.com/providers/lyceum	row fields, transcribed and joined by me (unquoted values are not verbatim): Lyceum \| NVIDIA L40S 48GB VRAM \| 48GB \| 12 vCPU 128GB RAM 100GB Storage \| Europe \| $1.99/GPU/hr \| Available; the 8x row at the same rate reads 'Sold Out'
2026-10-22	gpuperhour.com/providers/lyceum	about block, transcribed and joined by me: 'Founded 2024', 'Berlin, Germany and Zurich, Switzerland', Best For: European GPU data residency, On-demand and spot H100, H200, and A100 virtual machines, Per-second GPU workloads, Reserved multi-node InfiniBand clusters; '24 offers available'

More fields (1)

{
 "wave2": {
  "class": null,
  "free_tier_kind": "none",
  "shape": {
   "vcpu": 12,
   "ram_gib": 128,
   "disk_gib": 100,
   "gpu_class": "NVIDIA L40S 48GB",
   "gpu_count": 1
  }
 }
}

Unknown or withheld values (8)

$.chaining partly-held
$.egress.metered egress-unknown
$.identity_cost.detail not-stated
$.identity_cost.kind identity-defaulted
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.retention.notes process
$.retention.stated not-applicable

QuantaCloud Compute

paidemail?card?R

gpuperhour.com/ · GPU cloud · compute/quantacloud · JSON · compare

Free tier	Named by a live stock-and-price tracker as the cheapest H100 PCIe offer with stock on 2026-10-22: H100 PCIe 80GB at USD 2.59 per GPU-hour, across five providers that report stock on that model. The tracker counts 20,068 configurations from 31 providers, reads secure on-demand listings only, and requires a stock observation within 15 minutes of its refresh. unit hours; metered on wall-clock; none
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	email (placeholder) unverified: the vendor page was not read and no signup route was established, so the identity kind is unknown. Recorded as email because the corpus vocabulary has no unknown value; the reconcile pass should treat this as a gap rather than a finding.
Binding	none Nothing is known about this provider's terms beyond the rate; the entry exists so the name is in the corpus and the next session can price it directly.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	A 2.59 PCIe H100 undercuts the USD 3.29 H100 median and is close to the USD 2.27 preemptible H100 on Daytona, but with no preemption penalty if the listing is a secure offer.
Reliability	no reliability event recorded
Notes	It is recorded because the PCIe part sits between the two well-covered shapes, a 1.80 PCIe H100 at HyperAI and a 2.73 SXM H100 at Massed Compute, and a session pricing a PCIe H100 otherwise has no third point. The tracker's own caveat applies: an offer count is a count of configurations, not of physical GPUs.
Evidence rung	reported
Checked	2026-10-22 (live); stale on 2026-11-05

Evidence (2)

date	source	what it proved
2026-10-22	gpuperhour.com/	quoted: 'The cheapest current H100 PCIe is $2.59 per GPU-hour on QuantaCloud', with 5 providers reporting stock on that model
2026-10-22	gpuperhour.com/	quoted method: 'All figures are the cheapest current secure on-demand price per GPU-hour tracked by GPUPerHour, excluding spot and peer-to-peer marketplace listings', 31 providers and 20,068 configurations, stock observation within 15 minutes

More fields (1)

{
 "wave2": {
  "class": "gpu-broker-rate-floor",
  "free_tier_kind": "none",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": "H100 PCIe",
   "gpu_count": 1,
   "vram_gib": 80
  }
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated