B300
NVIDIA · Blackwell Ultra
288 GBHBM3e
8 TB/sBandwidth
4.5 PFFP8 dense
$4,899 /GPU/mo$6,552 · −25%
GPU catalog
10 data-center GPUs. One flat monthly price each.
NVIDIA Blackwell, Hopper and Ada, AMD Instinct CDNA 4 and CDNA 3. Dedicated GPUs from 1 to 8 per server, at least 25% below the market median.
- 204 GPUs free now
- 8 of 10 models in stock
- From $849/GPU/mo
Only 9 left
B200
NVIDIA · Blackwell
180 GBHBM3e
8 TB/sBandwidth
4.5 PFFP8 dense
$4,449 /GPU/mo$5,979 · −25%
MI355X
AMD · CDNA 4
288 GBHBM3E
8 TB/sBandwidth
5 PFFP8 dense
$2,449 /GPU/mo$3,285 · −25%
MI350X
AMD · CDNA 4
288 GBHBM3E
8 TB/sBandwidth
4.6 PFFP8 dense
$2,249 /GPU/mo$3,022 · −25%
H200
NVIDIA · Hopper
141 GBHBM3e
4.8 TB/sBandwidth
1.98 PFFP8 dense
$2,699 /GPU/mo$3,606 · −25%
MI325X
AMD · CDNA 3
256 GBHBM3E
6 TB/sBandwidth
2.6 PFFP8 dense
$2,299 /GPU/mo$3,071 · −25%
MI300X
AMD · CDNA 3
192 GBHBM3
5.3 TB/sBandwidth
2.6 PFFP8 dense
$1,749 /GPU/mo$2,351 · −25%
H100
NVIDIA · Hopper
80 GBHBM3
3.35 TB/sBandwidth
1.98 PFFP8 dense
$1,999 /GPU/mo$2,716 · −26%
RTX PRO 6000
NVIDIA · Blackwell
96 GBGDDR7
1.6 TB/sBandwidth
1 PFFP8 dense
$1,199 /GPU/mo$1,599 · −25%
L40S
NVIDIA · Ada Lovelace
48 GBGDDR6
864 GB/sBandwidth
0.73 PFFP8 dense
$849 /GPU/mo$1,146 · −25%
Specifications
Every GPU side by side
Per-GPU figures from vendor datasheets. Tensor throughput is dense. Click a model for its full page, pricing and FAQ.
| GPU | Architecture | Memory | Bandwidth | FP8 | FP4 | Interconnect | Per GPU / mo | Stock |
|---|---|---|---|---|---|---|---|---|
| NVIDIA B300 | Blackwell Ultra | 288 GB HBM3e | 8 TB/s | 4.5 PF | 13.5 PF | NVLink 5 · 1.8 TB/s | $4,899 | Only 9 left |
| NVIDIA B200 | Blackwell | 180 GB HBM3e | 8 TB/s | 4.5 PF | 9 PF | NVLink 5 · 1.8 TB/s | $4,449 | Back Nov 20 |
| AMD MI355X | CDNA 4 | 288 GB HBM3E | 8 TB/s | 5 PF | 10.1 PF | Infinity Fabric · 1.08 TB/s | $2,449 | 21 available |
| AMD MI350X | CDNA 4 | 288 GB HBM3E | 8 TB/s | 4.6 PF | 9.2 PF | Infinity Fabric · 1.08 TB/s | $2,249 | Only 6 left |
| NVIDIA H200 | Hopper | 141 GB HBM3e | 4.8 TB/s | 1.98 PF | — | NVLink 4 · 900 GB/s | $2,699 | 34 available |
| AMD MI325X | CDNA 3 | 256 GB HBM3E | 6 TB/s | 2.6 PF | — | Infinity Fabric · 896 GB/s | $2,299 | Back Dec 2 |
| AMD MI300X | CDNA 3 | 192 GB HBM3 | 5.3 TB/s | 2.6 PF | — | Infinity Fabric · 896 GB/s | $1,749 | 41 available |
| NVIDIA H100 | Hopper | 80 GB HBM3 | 3.35 TB/s | 1.98 PF | — | NVLink 4 · 900 GB/s | $1,999 | Only 3 left |
| NVIDIA RTX PRO 6000 | Blackwell | 96 GB GDDR7 | 1.6 TB/s | 1 PF | 2 PF | PCIe 5.0 x16 | $1,199 | 52 available |
| NVIDIA L40S | Ada Lovelace | 48 GB GDDR6 | 864 GB/s | 0.73 PF | — | PCIe 4.0 x16 | $849 | 38 available |
How to choose
Start from your workload
Pre-training
B300, B200 and MI355X with the most HBM per node.
See configurationsFine-tuning
H200, MI350X and B300 to fit 70B+ models and big batches.
See configurationsInference
From L40S to B300, sized to your model and traffic.
See configurationsRendering
RTX PRO 6000 and L40S with RT cores and video engines.
See configurationsFAQ
Choosing a GPU
Another question? Our engineers answer within one business day.
Contact usWhich GPU should I rent for LLM training?
Which GPU is best for inference?
It depends on model size: L40S and RTX PRO 6000 for models up to ~70B, H200 and MI300X for 70B–405B, and B300 for FP4 serving at scale.
NVIDIA or AMD?
NVIDIA runs every CUDA-only kernel and library. AMD Instinct GPUs give more memory per dollar and run PyTorch, JAX, vLLM and SGLang on ROCm. If your stack is framework-level, AMD is usually the better value.
Are all GPUs available in 8-GPU servers?
Yes, every model can be rented from 1 to 8 GPUs per server. SXM and OAM GPUs are connected by NVLink or Infinity Fabric; RTX PRO 6000 and L40S are PCIe cards.
Ready when you are: live in under 60 seconds
Pick a GPU and 1 to 8 units. Month to month, cancel anytime.