Choose the right GPU for you
Cluster training H200
⚪NVIDIA HGX H200 GPU
⚪1128 GB RAM
⚪20 vCPU
8 Card H200 GPU Node
- Equipped with 141GB of HBM3e memory with 4.8TB/sec bandwidth.
- Enhanced Tensor Core Architecture and Faster Memory Bandwidth for Large-Scale AI Deployments
- In the case of the full-blooded version of DeepSeek, for example, the inference throughput of a single 8-card H200 is expected to improve by about 30% or so over a 16-card H100.
- Bare metal delivery
Cluster Training H100
⚪NVIDIA HGX H100 GPU
⚪256GB RAM
⚪20 vCPU
8 Card H100 GPU Node
- Supports NVlink for multi-card training
- 25Gbps private network bandwidth
- Public network bandwidth 10Gbps or more
- Supports virtual machine or bare metal provisioning
Cluster AMD MI300X
⚪AMD MI300X
⚪1536 GB
⚪Intel® Xeon® Platinum 8568Y
8 Card AMD MI300X GPU Node
- 5.3 TB/s of HBM3 memory bandwidth
- 2048 GiB memory
- 6 TiB NVMe storage
- Bare metal offering
Cluster AMD MI300X
⚪NVIDIA A100 SXM4 GPU
⚪90 GB RAM
⚪12 vCPU
8 Card AMD MI300X GPU Node
- NVIDIA A100 SXM4 GPUs*8, 80GB*8 640GB HBM2 Memory
- Supports NVlink for multi-card training
- 10Gbps private network bandwidth
- Peak maximum bandwidth of 10Gbps on the public network, with 2Gbps bandwidth guaranteed
Inference
A100-80G
90 GB RAM
12 vCPU
A6000
45 GB RAM
8 vCPU
V100
16 GB RAM
30 GB RAM
8 vCPU
A5000
45 GB RAM
8 vCPU
A4000
45 GB RAM
8 vCPU
Compare these GPUs
| GPU | Architecture | FP16 Performance | FP32 Performance | Memory Capacity | Memory Type | Bandwidth |
|---|---|---|---|---|---|---|
|
A100 |
Ampere |
312 TFLOPS |
19.5 TFLOPS |
80 GB |
HBM2 |
2,039 GB/s |
|
V100 |
Volta |
125 TFLOPS |
15.7 TFLOPS |
32 GB |
HBM2 |
900 GB/s |
|
A6000 |
Ampere |
77.4 TFLOPS |
38.7 TFLOPS |
48 GB |
GDDR6 |
768 GB/s |
|
Ampere |
Ampere |
54.2 TFLOPS |
27.8 TFLOPS |
24 GB |
GDDR6 |
768 GB/s |
|
A4000 |
Ampere |
19.17 TFLOPS |
19.17 TFLOPS |
16 GB |
GDDR6 |
448 GB/s |
Quick Start
Quotation consultation, program design. We have certified product specialists ready to help.
