Rubin vs B300
Rubin and B300 both have 288 GB per GPU; Rubin's HBM4 runs at 22 TB/s, against 8 TB/s for B300's HBM3e. B300 rents from $7.10/GPU-hour on demand (13 November 2026); we have found no public hourly Rubin price at the clouds we track.
Specifications, side by side
| Specification | Rubin | B300 |
|---|---|---|
| Exact GPU | Rubin | B300 |
| Memory per GPU | 288 GB | 288 GB |
| Memory generation | HBM4 | HBM3e |
| Memory bandwidth per GPU | 22 TB/s | 8 TB/s |
| NVLink bandwidth per GPU | 3.6 TB/s | 1.8 TB/s |
| FP16/BF16 compute per GPU, dense | 4 PFLOPS | 2.25 PFLOPS ² |
| Availability | Announced; NVIDIA says partners offer it from the second half of 2026 · No public hourly price found (13 November 2026) | Rentable now: 30 offers from 16 providers (13 November 2026) |
The two GPUs have equal published memory capacity. Specifications checked 5 October 2026.
These are NVIDIA's published peak figures, not measured speed on any workload.
² Our calculation. Calculation: 36 PFLOPS sparse ÷ 2 ÷ 8 GPUs (NVIDIA HGX B200/B300 table).
Recorded on-demand rental rates
| Rate | Rubin | B300 |
|---|---|---|
| Lowest recorded on-demand | No recorded on-demand rate | $7.10/GPU-hour (recorded 13 November 2026) |
Every rate comes from our daily price collection and shows the date it was recorded.
Check Rubin cloud announcements and availability.
Current B300 rental offers
More filters: provider, memory, node size, stock, place, CPU and RAM, sort, price unit
“Machines in” filters the 9 of 26 offers whose provider says where the machine is; the others are under “Place not stated”.
“vCPUs per GPU” and “RAM per GPU” use the 5 of 26 offers whose provider publishes the machine's vCPUs or RAM for a stated number of GPUs (DigitalOcean, Latitude.sh and RunPod); an offer that does not publish them never matches a minimum. A GiB counts as 1.07 GB.
Crusoe, Daytona, DigitalOcean, fal, Hyperstack, Modal, Nebius, RunPod, Spheron and Synteq put GPUs in North America on their own pages, but none of their offers here says so.
Crusoe, Daytona, DigitalOcean, fal, Hyperstack, Modal, Nebius, Spheron and Verda (DataCrunch) put GPUs in Europe on their own pages, but none of their offers here says so.
Modal, RunPod and Spheron put GPUs in Asia and Oceania on their own pages, but none of their offers here says so.
| GPU / provider | Billing | $/GPU-hour | Node · total per hour | Rent |
|---|---|---|---|---|
| B300 · Daytona288 GB · B300; GPU rate only, CPU/memory/disk billed separately | on-demand | $7.10 | – | Rent → |
| B300 · Hyperstack288 GB | on-demand | $7.40 | – | Rent →checked 13 NovDetails
|
| B300 · Scaleway288 GB | on-demand | $8.31 | 8× $66.49 | Rent → |
| B300 · Runcrate288 GB · B300; lowest live rate; billed per second | on-demandStarting rate | $8.72 | – | Rent → |
| B300 · Crusoe288 GB | on-demand | $8.96 | – | Rent → |
| B300 · RunPod288 GB | secure cloud | $8.99 | – | In stock: lowRent →checked 13 NovDetails
|
| B300 · Nebius288 GB | on-demand | $9.50 | – | Rent → |
| B300 · DigitalOcean288 GB · ×2 listings | on-demand | $11.19 | 1× $11.19 | Rent →checked 13 NovDetails
|
| B300 · Spheron288 GB | on-demand | $11.47−7.6% in 30 days | – | Rent → |
| B300 · Verda (DataCrunch)268 GB | on-demand | $12.29+25.3% in 30 days | 1× $12.29 | Rent → |
| B300 · Oracle Cloud263 GB · BM.GPU.B300.8; bare metal, 8 GPUs; billed per second after the first hour | on-demand | $15.00 | 8× $120.00 | Rent →checked 13 NovDetails
|
| B300 · Latitude.sh288 GB · g4-b300-large | on-demand | $16.00 | 8× $128.00 | Out of stockchecked 13 NovDetails
|
| B300 · Daytona288 GB · B300; GPU rate only, CPU/memory/disk billed separately | spot / interruptible | $4.08 | – | Rent → |
| B300 · Spheron288 GB | spot / interruptible | $5.84−6.7% in 30 days | – | Rent → |
| B300 · Verda (DataCrunch)268 GB | spot / interruptible | $6.14 | 1× $6.14 | Rent → |
| B300 · Lium269 GB · 1-GPU host | marketplace host | $6.58 | 1× $6.58 | In stockRent →checked 13 NovDetails
|
| B300 · Lium269 GB · ×2 hosts | spot / interruptible | $6.90 | 8× $55.20 | In stockRent →checked 13 NovDetails
|
| B300 · RunPod288 GB | community cloud | $6.94 | – | Rent → |
| B300 · Modal288 GB | serverless | $7.10$0.001972/sec | – | Rent →checked 13 NovDetails
|
| B300 · Lium269 GB · ×3 hosts | marketplace host | $11.90 | 8× $95.20 | In stockRent →checked 13 NovDetails
|
| B300 · RunPod288 GB · Flex worker | serverless | $11.99$0.003331/sec | – | Rent → |
| B300 · Lium269 GB · 1-GPU host | marketplace host | $12.67 | 1× $12.67 | In stockRent →checked 13 NovDetails
|
| B300 · fal288 GB | serverless | $12.99$0.003608/sec | – | Rent → |
| B300 · Latitude.sh288 GB · g4-b300-large | reserved / committed12 months, annual plan; normalized over 8760 hours | $5.60−31.0% in 30 days | 8× $44.80 | Out of stockchecked 13 NovDetails
|
| B300 · Synteq288 GB · nvidia-b300; full passthrough GPU; setup fee not reported | reserved / committed1 month, $4555/month ÷ 730 hours; monthly billing | $6.24 | – | Out of stock |
| B300 · Latitude.sh288 GB · g4-b300-large | reserved / committed1 month, monthly plan; normalized over 720 hours | $8.11 | 8× $64.89 | Out of stockchecked 13 NovDetails
|
Try another provider or reduce the minimum memory.
A spot machine can be stopped whenever the provider needs it back: how much notice each provider gives, and what happens to the disk.
Fresh prices come first. Within each billing type, prices run lowest first, with unavailable offers last. Commercial relationships never affect ordering. How we collect and rank prices.
Same GPU, provider, billing and price on several hosts or listings: one row, with the count. – means the provider does not state it.
Monthly view assumes 730 hours of continuous use. Daily view assumes 24 hours.
Cite this data
GPU Rental Prices, "Rubin vs B300: GPU memory and rental prices", prices checked 13 November 2026, https://gpurentalprices.com/vs/rubin-vs-b300
Quote it with a link to this page. Downloads of the whole dataset are on Data & API.
Specifications and sources
Rubin values come from NVIDIA's individual GPU specification table. B300 capacity comes from its NVIDIA specification. Checked 5 October 2026. Quoted evidence is listed under Sources in the footer.
B300 memory type · NVIDIA DGX B200 · NVIDIA HGX B200/B300 · Blackwell Ultra specifications · NVIDIA H200
NVIDIA publishes two sets of per-GPU Rubin figures. Its HGX table lists 22 TB/s of memory bandwidth and 3.6 TB/s of NVLink bandwidth; its Vera Rubin NVL72 table lists 19.2 TB/s and 3 TB/s, for an at-scale AI factory. Ask the provider which system it rents before comparing performance.
B300 rental prices and history · Rubin specifications and cloud availability · All GPU comparisons