Hardware reference / Consumer
NVIDIA RTX 5090
Blackwell architecture. Device specifications and exact provider configurations, kept separate.
- VRAM
- 32 GB
- Source-verified
- Memory bandwidth
- 1,792 GB/s
- Source-verified
- CUDA cores
- 21,760
- Source-verified
RTX 5090 Provider Offer Comparison
View approved price historyHourly prices recorded; units unverified
2 recorded offers; not live inventory. USD quotes; availability is not live. Compare matching variants and billing units.
| Provider | Configuration / billing | Recorded hourly price | GPU count | Recorded VRAM | Availability | Provenance |
|---|---|---|---|---|---|---|
| saladcloud | RTX 5090; billing model unknown | Recorded $0.27/hour (unit unverified) USD | 1 GPU | 32 GB (scope unverified) | Recorded availability: Available |
Price unverified Source unknown EvidenceHistorical catalogue price; verification date unknown |
| gpuhub | RTX 5090; billing model unknown | Recorded $0.36/hour (unit unverified) USD | 1 GPU | 32 GB (scope unverified) | Recorded availability: Available |
Price unverified Source unknown EvidenceHistorical catalogue price; verification date unknown |
Specifications & reference
Recorded specifications describe the device, not necessarily every provider configuration. Check source scope before comparing.
Recorded Technical Specifications
- CUDA cores
- 21,760
- Base clock
- 1,300 MHz
- Boost clock
- 2,600 MHz
- Memory
- 32 GB; GDDR7
- Memory bus
- 512 bit
- Recorded TDP / board power
- 575 W
Recorded Performance
- Recorded FP32
- Unknown
- Recorded FP16
- Unknown
Only source-verified benchmark fields are shown. Match precision, sparsity, form factor and test conditions before comparison; unknown means no verified measurement is recorded.
Detailed Specifications
| Architecture | Blackwell |
|---|---|
| GPU code name | Unknown |
| Recorded release date | 2025-01-30 |
| Recorded launch price (not rental) | 1,999 USD |
| Manufacturing process | 4nm |
| Transistors | 100B |
Specification review: 2026-11-01 · 10 source-verified fields · Cloud availability: unknown
Specification sources & scope
Only the listed fields are source-verified; other specifications remain unverified. Source review does not establish live availability.
- Source reviewed
- 2026-11-01
- Source-verified fields
- accelerator type, architecture, cuda cores, manufacturer, market segment, memory bandwidth, memory capacity, memory type, model, power consumption tdp
- Specification scope
- Single GPU; manufacturer reference specifications. Board-partner configurations may differ. Memory uses catalogue MiB units; bandwidth uses decimal GB/s; power is published board maximum.
- Lifecycle evidence
- Documented in manufacturer source; not a stock or shipping-status claim
- Lifecycle note
- Official manufacturer specifications reviewed; this is not a stock check, release-date verification or end-of-sale determination.
- Cloud rental availability
- Unknown — catalogue inclusion is not confirmation of rental stock
Manufacturer source links
https://www.nvidia.com/en-us/geforce/graphics-cards/compare/Recorded AI features
Only source-verified fields are shown; unknown does not mean unsupported. Confirm hardware and software versions.
Tensor Cores
Unknown
Transformer Engine
Unknown
Flash Attention
Unknown
Physical specifications
Physical Specifications
Dimensions
11in
Length
4.5in
Width
3-slot
Height
Software & workload notes
Allow memory for weights, optimizer state, KV cache and framework overhead. Core counts and nominal throughput are not directly comparable across architectures or manufacturers.
Recorded CUDA compute capability: 12.0. CUDA is NVIDIA-specific; confirm driver, toolkit and framework requirements against NVIDIA's documentation.
Manufacturer & software references
Confirm support for the exact device, operating system and software version.
NVIDIA CUDA documentation ↗NVIDIA GPU compute capability reference ↗