Seeweb
computeprices.com · Compute · 7 records · JSON
- Avian Compute Computepaidemail?card?R
- Nscale Computepaidemail?card?R
- QuickPod Computepaidemail?card?R
- Seeweb Computepaidemail?card?R
- Sesterce Computepaidemail?card?R
- SwissGPU Computepaidemail?card?R
- TensorWave Computepaidemail?card?R
Avian Compute Compute
paidemail?card?R
| Free tier | USD 4.32 per GPU-hour on an 8-GPU NVIDIA B200 node with 180 GB per card, sold as a 12-month reservation in North America: 128 vCPU, 2048 GB RAM. The second row is an 8-GPU HGX B300 with 288 GB per card at USD 4.95 per GPU-hour on 192 vCPU and 4096 GB. Two GPU models, both reserved rather than on-demand. 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 transfer quota is stated. The binding term is the commitment instead: both rows are marked 'Reserved - 12mo', so the cost of the cheapest offer is an annual contract on eight cards, and the index's per-hour figure is a division of that commitment rather than an hourly price a client can walk away from. |
| Interface | rest inferred from the tooling the provider documents; not probed |
| Tooling | the provider description states it offers 'a range of AI models accessible via a single API key, billed on a pay-per-token basis or through long-term reservations' |
| Composes with | USD 4.32 per GPU-hour for a B200 with 180 GB is the cheapest Blackwell rate, and it is also the least reachable: the commitment is twelve months on eight cards. An 8x B200 node at that rate is the machine for a training run that will not fit on a rented 48 GB card, not a machine for a single inference job, and the pay-per-token API on the same provider is the row to price for the small job |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices. Not in dossier/compute (searched: avian, no record). Both rows are reserved-term, so the per-hour figure is not an on-demand rate and is recorded as such. would-settle: the vendor's own compute page for the reservation terms and whether shorter terms exist. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/avian | row fields, transcribed and joined by me (unquoted values are not verbatim): [B200] | $4.32/hr | 180GB | 8x | 128 | 2048 GB | 10/21/2026, with the rows tags reading Reserved - 12mo' and 'North America' |
| 2026-10-22 | computeprices.com/providers/avian | second row fields, transcribed and joined by me (unquoted values are not verbatim): [HGX B300] | $4.95/hr | 288GB | 8x | 192 | 4096 GB | 10/21/2026 with the same Reserved - 12mo and 'North America' tags; provider described as 'Avian Compute provides AI inference and dedicated GPU compute services through NVIDIA B200 and B300 HGX nodes' |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": 128,
"ram_gib": 2048,
"disk_gib": null,
"gpu_class": "NVIDIA B200 180GB",
"gpu_count": 8
}
}
}Unknown or withheld values (8)
$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable
Nscale Compute
paidemail?card?R
| Free tier | No price: the index records 'Nscale does not publish public hourly pricing. Contact the provider directly for a quote.' The provider is described as 'a vertically integrated AI hyperscaler offering GPU clusters and managed AI services from data centers in the UK, Europe and the US, with a focus on Blackwell-generation reservations and large-scale training capacity'. No rate row exists to record. unit none; 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 and no price are stated, so nothing in this record binds. The page's own framing is 'Reservation-led capacity', so the wall is a reservation rather than a metered limit. |
| Interface | unknown inferred from the tooling the provider documents; not probed |
| Tooling | not stated on the page read |
| Composes with | Sovereign UK and EU capacity for Blackwell-generation reservations is the compliance-shaped end of the market, and with no published rate it is a lead for the identity-and-procurement lane rather than a price floor. Recorded so the name is in the corpus with its actual state: quote-only, reservation-led |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices. Not in dossier/compute (searched: nscale, no record). Like TensorWave this is a partial finding: no number exists to record and none was invented. would-settle: a quote request or the vendor's own pricing page, neither of which attempted. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/nscale | page quoted: 'Nscale is a vertically integrated AI hyperscaler offering GPU clusters and managed AI services from data centers in the UK, Europe and the US, with a focus on Blackwell-generation reservations and large-scale training capacity.' and 'Nscale does not publish public hourly pricing. Contact the provider directly for a quote.' |
| 2026-10-22 | computeprices.com/providers/nscale | the advantages list states 'Reservation-led capacity for newest NVIDIA Blackwell systems' and 'Vertically integrated control of power, networking and cooling' |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": null,
"ram_gib": null,
"disk_gib": null,
"gpu_class": "NVIDIA Blackwell generation (model per configuration not stated)",
"gpu_count": null
}
}
}Unknown or withheld values (9)
$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.interfaceinterface-unknown$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable
QuickPod Compute
paidemail?card?R
| Free tier | USD 0.100 per GPU-hour for a single RTX A4000 with 16 GB in Kansas, US: 12 vCPU, 41 GB RAM. Next rows: the same card at the same rate as a 2x configuration on 8 vCPU and 125 GB in Kentucky; RTX 3060 12 GB at USD 0.120 on 10 vCPU and 35 GB in Indiana; RTX 3090 24 GB at USD 0.180 on 4 vCPU and 31 GB in Kentucky; RTX 3080 10 GB at USD 0.230 in Oxfordshire, GB; RTX 5090 32 GB at USD 0.570 in Queensland, AU. Five models across four US states, the UK and Australia. 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: the page publishes a per-GPU hourly rate, VRAM, vCPU count, RAM and a region. The wall is the hourly rate, and the rate is low enough that the machine-hours are the constraint rather than a byte counter. |
| Interface | unknown inferred from the tooling the provider documents; not probed |
| Tooling | the provider is tagged 'docker' and the description says 'docker pod rentals through a streamlined interface'; a console URL at console.quickpod.io is the linked order path |
| Composes with | Ten cents an hour for a 16 GB card is the cheapest GPU rate, half of Cirrascale's A4000 rate and a third of SwissGPU's RTX 4080. The region is the other half of the offer: US, UK and Australian hosts under one account makes it the row to price when the crawler's source address and the GPU have to sit in the same country |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices. Not in dossier/compute (searched: quickpod, no record). The $0.100 row is dated 10/17/2026 on the index while the page's own 'last updated' line reads 21 October 2026, so the per-row date is the one quoted. would-settle: console.quickpod.io for the disk size, the billing minimum and whether the A4000 row is still stocked at ten cents. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/quickpod | row fields, transcribed and joined by me (unquoted values are not verbatim): [RTX A4000] | $0.100/hrKansas, US | 16GB | 1x | 12 | 41 GB | 10/17/2026; the page header reads 'GPU Models 5', 'From / hour $0.10' and 'Prices last updated: October 21, 2026' |
| 2026-10-22 | computeprices.com/providers/quickpod | provider quoted: 'QuickPod offers CPU and GPU docker pod rentals through a streamlined interface', tagged 'GPU-focused', 'budget', 'docker'; further rows RTX 3090 at $0.180/hr in Kentucky, RTX 3060 at $0.120/hr in Indiana, RTX 3080 at $0.230/hr in Oxfordshire GB, RTX 5090 at $0.570/hr in Queensland AU |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": 12,
"ram_gib": 41,
"disk_gib": null,
"gpu_class": "NVIDIA RTX A4000 16GB",
"gpu_count": 1
}
}
}Unknown or withheld values (9)
$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.interfaceinterface-unknown$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable
Seeweb Compute
paidemail?card?R
| Free tier | USD 0.325 per hour for a single Quadro RTX 6000 with 24 GB, the cheapest row on the page and consistent with its header 'From / hour $0.33'. Next rows: A30 24 GB at USD 0.326; L4 24 GB at USD 0.427; RTX A6000 48 GB at USD 0.831; L40S 48 GB at USD 0.955; A100 PCIe 80 GB at USD 1.11; H100 SXM 80 GB at USD 2.12; H200 141 GB at USD 2.92; MI300X 192 GB at USD 2.64. Ten GPU models, prices dated 22 October 2026. unit GiB; 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: the page publishes a per-GPU hourly rate, VRAM and a count per row. The wall is the hourly rate and whatever minimum the vendor's own order page sets. |
| Interface | unknown inferred from the tooling the provider documents; not probed |
| Tooling | not stated on the page read |
| Composes with | Ten models from one Italian vendor, including the only AMD MI300X with 192 GB on this index at USD 2.64 an hour, makes this the single account to price against when the job needs a specific accelerator rather than a specific provider. The 24 GB floor at USD 0.325 is competitive with QuickPod's A4000 and more expensive than SwissGPU's RTX 4080 |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices, which already appears on docs/compute/watchlist.md. Not in dossier/compute (searched: seeweb, no record). The index row carries no vCPU, RAM or disk, so those shape fields are null rather than guessed, and the vendor page it links was not fetched. would-settle: seeweb.it's GPU cloud page for the host shape that comes with each card. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/seeweb | row fields, transcribed and joined by me (unquoted values are not verbatim): [Quadro RTX 6000] | $0.325/hr | 24GB | 1x | 10/22/2026; the page header reads 'From / hour $0.33' and 'Prices last updated: October 22, 2026' |
| 2026-10-22 | computeprices.com/providers/seeweb | provider quoted: 'Italian cloud provider with NVIDIA and AMD GPU cloud servers', 'part of the Dominion Hosting Holding group, offering GPU-accelerated cloud servers with NVIDIA and AMD accelerators, pay-per-hour billing, and data centers across Europe'; other rows as listed above, including MI300X at $2.64/hr |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": null,
"ram_gib": null,
"disk_gib": null,
"gpu_class": "NVIDIA Quadro RTX 6000 24GB",
"gpu_count": 1
}
}
}Unknown or withheld values (9)
$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.interfaceinterface-unknown$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable
Sesterce Compute
paidemail?card?R
| Free tier | USD 1.42 per GPU-hour for a single A10 with 24 GB: 30 vCPU, 200 GB RAM, region labelled 'Zone 7'. The A100 PCIe 80 GB follows at USD 1.49 on a 28 vCPU, 120 GB host in Zone 5 and in an unlabelled zone, and A100 SXM at USD 1.52 on 14 vCPU and 100 GB in Zone 13. Seventeen GPU models are listed across zones in North America, Europe and Asia-Pacific. 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: the page publishes a per-GPU hourly rate, VRAM, vCPU count, RAM and a zone label. The page header also prints 'From / hour $0.17' without naming the card or zone, so the true floor is below the cheapest row that carries a specification and is unverified. |
| Interface | unknown inferred from the tooling the provider documents; not probed |
| Tooling | not stated on the page read |
| Composes with | An A10 with 30 vCPU and 200 GB of system RAM for USD 1.42 an hour is the most CPU-per-GPU shape, which is what a data pipeline that touches the accelerator intermittently wants: the 24 GB card is small enough for embedding and inference work, and the 200 GB of RAM is where the batch lives. The zone labels are the reason not to read this as one price, because the same card is printed at different rates per zone |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices. Not in dossier/compute (searched: sesterce, no record). The zone naming ('Zone 5', 'Zone 7', 'Zone 13') is the index's own, and the vendor's zone-to-city mapping was not read. would-settle: cloud.sesterce.com's price list for what each zone is and the disk that ships with the A10 row. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/sesterce | row fields, transcribed and joined by me (unquoted values are not verbatim): [A10] | $1.42/hrZone 7 | 24GB | 1x | 30 | 200 GB | 10/22/2026; the page header reads 'GPU Models 17' and 'From / hour $0.17' |
| 2026-10-22 | computeprices.com/providers/sesterce | provider, transcribed and joined by me: 'Sovereign European AI factories and on-demand GPU cloud', 'a self-serve GPU cloud with on-demand VMs, bare-metal servers and dedicated clusters. Its self-serve on-demand catalog lists regions across North America, Europe and Asia-Pacific'; further rows [A100 PCIe 80GB] | $1.49/hr | 80GB | 1x | 28 | 120 GB and [A100 SXM] | $1.52/hrZone 13 | 80GB | 1x 8x | 14 | 100 GB |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": 30,
"ram_gib": 200,
"disk_gib": null,
"gpu_class": "NVIDIA A10 24GB",
"gpu_count": 1
}
}
}Unknown or withheld values (9)
$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.interfaceinterface-unknown$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable
SwissGPU Compute
paidemail?card?R
| Free tier | USD 0.303 per hour for a single RTX 4080 with 16 GB: 12 vCPU, 32 GB RAM, in Switzerland. Next rows: RTX 5080 16 GB at USD 0.364 on 16 vCPU and 32 GB; RTX 4090 24 GB at USD 0.486 on 16 vCPU and 32 GB; RTX 5090 32 GB at USD 0.607; RTX 6000 Ada 48 GB at USD 1.82 on 20 vCPU and 64 GB. Five models, prices dated 21 October 2026. unit GiB; 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: the page publishes a per-GPU hourly rate, VRAM, vCPU count and RAM. 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 | Sixteen gigabytes holds a 7B model at 4-bit, which is the useful floor for local inference experiments |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices. Not in dossier/compute (searched: swissgpu, no record). The description calls the servers bare-metal while the index prices them per GPU-hour on a vCPU count, which is not a contradiction only if the host is dedicated and the billing is hourly; the page does not resolve it. would-settle: swissgpu.ch/pricing, which the index links, for the billing unit and the disk size. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/swissgpu | row fields, transcribed and joined by me (unquoted values are not verbatim): [RTX 4080] | $0.303/hr | 16GB | 1x | 12 | 32 GB | 10/21/2026; the page header reads 'GPU Models 5', 'From / hour $0.30' and 'Prices last updated: October 21, 2026' |
| 2026-10-22 | computeprices.com/providers/swissgpu | provider quoted: 'Swiss-based GPU servers with low latency', 'Provider of bare-metal GPU servers fully hosted in Switzerland.'; further rows RTX 5080 at $0.364/hr, RTX 4090 at $0.486/hr, RTX 5090 at $0.607/hr, RTX 6000 Ada at $1.82/hr |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": 12,
"ram_gib": 32,
"disk_gib": null,
"gpu_class": "NVIDIA RTX 4080 16GB",
"gpu_count": 1
}
}
}Unknown or withheld values (10)
$.chainingpartly-held$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.interfaceinterface-unknown$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable
TensorWave Compute
paidemail?card?R
| Free tier | No price: the index records the provider as quote-only, stating 'TensorWave does not publish public hourly pricing. Contact the provider directly for a quote.' The provider is described as 'an AMD-first AI cloud with bare-metal and managed clusters built on Instinct MI300X/MI325X/MI355X/MI455X accelerators with large VRAM and ROCm'. It is tracked for AMD hardware and carries no rate row, so no figure is recorded here. unit GiB; 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 and no price are stated, so nothing in this record binds. The wall is a sales conversation: the provider publishes no hourly rate at all. |
| Interface | rest inferred from the tooling the provider documents; not probed |
| Tooling | the page states the software stack is ROCm; no API documentation is named |
| Composes with | The corpus has AMD capacity only at second hand, through providers that happen to stock MI300X, so this is the vendor to price when a job needs a 192 GB or 288 GB device and the CUDA toolchain is not mandatory. With no published rate it cannot be ranked against Cirrascale or Seeweb, and the record exists to stop a later session re-discovering it as new |
| Reliability | no reliability event recorded |
| Notes | Found via computeprices. Not in dossier/compute (searched: tensorwave, no record; the one grep hit is dossier/compute/test-flight.json, which mentions the name). This is a partial finding by construction: the provider is real, sells the capacity, and publishes no number, so the price fields are empty rather than estimated. would-settle: a quote request, which does not make because it would require an account. |
| Evidence rung | reported |
| Checked | 2026-10-22 (live); stale on 2026-11-05 |
Evidence (2)
| date | source | what it proved |
|---|---|---|
| 2026-10-22 | computeprices.com/providers/tensorwave | page, transcribed and joined by me (unquoted values are not verbatim): TensorWave is an AMD-first AI cloud with bare-metal and managed clusters built on Instinct MI300X/MI325X/MI355X/MI455X accelerators with large VRAM and ROCm. and 'TensorWave does not publish public hourly pricing. Contact the provider directly for a quote.' |
| 2026-10-22 | computeprices.com/providers/tensorwave | the pages advantages list states First-to-market with latest AMD accelerators (MI455X, MI355X, MI325X)', 'High VRAM capacity up to 288GB per GPU', 'SOC 2 Type II and ISO 27001 certified and HIPAA compliant', 'Utilizes an open-source software stack (ROCm)' |
More fields (1)
{
"wave2": {
"class": null,
"free_tier_kind": "none",
"shape": {
"vcpu": null,
"ram_gib": null,
"disk_gib": null,
"gpu_class": "AMD Instinct MI300X / MI325X / MI355X / MI455X (model per configuration not stated)",
"gpu_count": null
}
}
}Unknown or withheld values (8)
$.egress.meteredegress-unknown$.identity_cost.detailnot-stated$.identity_cost.kindidentity-defaulted$.per_object_cap.notesprocess$.per_object_cap.statednot-applicable$.reliability_notenone-observed$.retention.notesprocess$.retention.statednot-applicable