a3-megagpu-8g
Compute Engine · A3 family
The a3-megagpu-8g is a gpu & accelerated Compute Engine machine type with 208 vCPUs and 1,872 GiB of memory, 8 × NVIDIA H100, running on Intel Sapphire Rapids with up to 1600 gbps networking, starting at $92.94 per hour on demand.
- vCPUs
- 208
- memory
- 1,872 GiB
- gpu
- 8× H100
- on-demand
- $92.94
- spot vm
- $39.63
- 1-yr committed use
- $63.94
dedicated
9 GiB per vCPU
640 GiB total
$67,844/mo · us-east4
−57% vs on-demand
−31% vs on-demand
a3-megagpu-8g price by region
9 regions · hourly list prices in USD · On-demand · Spot VM · 1-yr Committed use · 3-yr Committed use
| Northern Virginiaus-east4cheapestNorth America | $92.94 | $39.63 | $63.94 | $40.49 |
| Columbusus-east5North America | $94.07 | $39.72 | $64.73 | $40.98 |
| Belgiumeurope-west1Europe | $102 | $45.78 | $70.32 | $44.52 |
| Las Vegasus-west4North America | $105 | $28.21 | $71.96 | $45.56 |
| Frankfurteurope-west3Europe | $110 | $64.81 | $75.41 | $47.75 |
| Sydneyaustralia-southeast1Oceania | $116 | $46.68 | $79.92 | $50.60 |
| Netherlandseurope-west4Europe | $118 | $70.87 | $81.34 | $51.49 |
| Tokyoasia-northeast1Asia | $119 | $31.03 | $82.07 | $51.96 |
| Singaporeasia-southeast1Asia | $120 | $66.84 | $82.55 | $52.26 |
Specifications
compute
- vCPUs
- 208
- Memory
- 1,872 GiB
- Memory per vCPU
- 9 GiB
- Processor
- Intel Sapphire Rapids
- CPU vendor
- Intel
- Architecture
- x86-64
- Clock speed
- 1.9 GHz (turbo 3 GHz)
- Burstable CPU
- No
storage & network
- Local storage
- 6 TB NVME (16 disks)
- Storage detail
- 16 x 375 GiB local SSD
- Network
- Up to 1600 Gbps
- Peak bandwidth
- 1,600 Gbps
accelerators
- Model
- NVIDIA H100
- Count
- 8
- Memory (total)
- 640 GiB
- Architecture
- Hopper
- Vendor
- NVIDIA
google cloud details
- API name
a3-megagpu-8g- Family
- A3
- Category
- GPU & accelerated
- Generation
- Current
- Regions
- 9 (Asia, Europe, North America, Oceania)
- Cheapest region
- Northern Virginia · $92.94/hr
- Max persistent disks
- 128
a3-megagpu-8g equivalents on other clouds
closest match by vCPU, memory, CPU architecture, accelerator and class · Linux on-demand in each provider's reference region
| Provider | Instance | vCPU | Memory | Processor | On-demand | vs a3-megagpu-8g |
|---|---|---|---|---|---|---|
| Google Cloud | a3-megagpu-8g | 208 | 1,872 GiB | 8 × NVIDIA H100 | $92.94 | – |
| AWS | p5.48xlarge192 vCPU (1.1× less) · 2048 GiB (1.09×) | 192 | 2 TiB | 8 × NVIDIA H100 | $55.04 | -41% |
| g7e.48xlarge192 vCPU (1.1× less) · 2048 GiB (1.09×) · NVIDIA RTX PRO 6000 Blackwell | 192 | 2 TiB | 8 × NVIDIA RTX PRO 6000 Blackwell | $33.14 | -64% | |
| Azure | ND96isr H100 v596 vCPU (2.2× less) | 96 | 1,900 GiB | 8 × NVIDIA H100 | $98.32 | +6% |
A3 family sizes
| Size | vCPU | Memory | Accelerator | Local storage | On-demand | Monthly | Spot | 1-yr |
|---|---|---|---|---|---|---|---|---|
| a3-highgpu-1g | 26 | 234 GiB | 1 × NVIDIA H100 | – | $11.06 | $8,075 | $6.588 | $7.673 |
| a3-highgpu-2g | 52 | 468 GiB | 2 × NVIDIA H100 | – | $22.12 | $16,149 | $13.18 | $15.35 |
| a3-highgpu-4g | 104 | 936 GiB | 4 × NVIDIA H100 | – | $44.25 | $32,299 | $26.35 | $30.69 |
| a3-highgpu-8g | 208 | 1,872 GiB | 8 × NVIDIA H100 | 6 TB NVME (16 disks) | $88.49 | $64,598 | $52.70 | $61.38 |
| a3-megagpu-8gus-east4 | 208 | 1,872 GiB | 8 × NVIDIA H100 | 6 TB NVME (16 disks) | $92.94 | $67,844 | $39.63 | $63.94 |
a3-megagpu-8g FAQ
How much does a3-megagpu-8g cost?
a3-megagpu-8g costs $92.94 per hour on demand with Linux in Northern Virginia (us-east4), about $67,844 per month. Across 9 regions the Linux on-demand price ranges from $92.94 in Northern Virginia to $120 in Singapore. Spot VM capacity costs $39.63 per hour (57% less), but can be reclaimed. A 1-yr Committed use commitment brings it to $63.94 per hour. With Windows it costs $103 per hour.
How many vCPUs and how much memory does a3-megagpu-8g have?
a3-megagpu-8g has 208 vCPUs and 1,872 GiB of memory (9 GiB per vCPU), running on Intel Sapphire Rapids. It is a gpu & accelerated machine type in the A3 family.
What GPU does a3-megagpu-8g have?
a3-megagpu-8g has 8 × NVIDIA H100 with 640 GiB of accelerator memory in total.
What are the equivalents of a3-megagpu-8g on other clouds?
The closest matches by vCPU, memory, CPU architecture, accelerator and instance class are on AWS, p5.48xlarge (192 vCPU, 2 TiB, $55.04/hour in US East (N. Virginia)); on Azure, ND96isr H100 v5 (96 vCPU, 1,900 GiB, $98.32/hour in East US).