ArthurChiao's Blog

GPU Prices Quick Reference (2023)

Published at 2023-06-17 | Last Update 2023-08-06

This post lists some GPU node prices collected from several public cloud vendors, intended primarily for personal reference. Note that these prices are subject to change over time, so if you are planning a serious budget, please consult each cloud vendor’s pricing page for the most up-to-date information.

1 AWS (as of 2023.06)

GPU instances types (flavors),

Note that some flavors may only be available in specific regions.

1.1 Region ap-southeast-1 (Singapore)

A100

Flavors: p4d.*

Not available in Singapore at the time of this writing (2023-06).

| Instance type | vCPUs | Architecture | Memory (GiB) | Network performance | GPUs     |
|---------------|-------|--------------|--------------|---------------------|----------|
| p4d.24xlarge  | 96    | x86_64       | 1152         | 4x 100 Gigabit      | 8        |

Reference prices & discount prices

TODO.

V100

Flavors: p3.*

| Instance type | vCPUs | Architecture | Memory (GiB) | Network performance | GPUs     |
|---------------|-------|--------------|--------------|---------------------|----------|
| p3.2xlarge    | 8     | x86_64       |  61          | Up to 10 Gigabit    | 1 * V100 |
| p3.8xlarge    | 32    | x86_64       | 244          | 10 Gigabit          | 4 * V100 |
| p3.16xlarge   | 64    | x86_64       | 488          | 25 Gigabit          | 8 * V100 |

Reference prices & discount prices

On-demand OS=Linux prices, in USD/hour:

| Instance type | vCPUs | Arch   | Memory (GiB) | Network performance | GPUs     | Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|---------------|-------|--------|--------------|---------------------|----------|-----------|---------|---------|---------|---------|---------|
| p3.2xlarge    | 8     | x86_64 |  61          | Up to 10 Gigabit    | 1 * V100 | 4.234     |   2.964 |   2.540 |   2.117 |   1.694 |   1.270 |
| p3.8xlarge    | 32    | x86_64 | 244          | 10 Gigabit          | 4 * V100 | 16.936    |  11.855 |  10.162 |   8.468 |   6.774 |   5.081 |
| p3.16xlarge   | 64    | x86_64 | 488          | 25 Gigabit          | 8 * V100 | 33.872    |  23.710 |  20.323 |  16.936 |  13.549 |  10.162 |

T4

Flavors: g4dn.*

| Instance type | vCPUs | Architecture | Memory (GiB) | Network performance | GPUs   |
|---------------|-------|--------------|--------------|---------------------|--------|
| g4dn.xlarge   |  4    | x86_64       |  16          | Up to 25 Gigabit    | 1 * T4 |
| g4dn.2xlarge  |  8    | x86_64       |  32          | Up to 25 Gigabit    | 1 * T4 |
| g4dn.4xlarge  | 16    | x86_64       |  64          | Up to 25 Gigabit    | 1 * T4 |
| g4dn.8xlarge  | 32    | x86_64       | 128          | 50 Gigabit          | 1 * T4 |
| g4dn.12xlarge | 48    | x86_64       | 192          | 50 Gigabit          | 4 * T4 |
| g4dn.16xlarge | 64    | x86_64       | 256          | 50 Gigabit          | 1 * T4 |
| g4dn.metal    | 96    | x86_64       | 384          | 100 Gigabit         | 8 * T4 |

Reference prices & discount prices

On-demand OS=Linux prices, in USD/hour:

| Instance type | vCPUs | Arch   | Memory (GiB) | Network performance | GPUs   | Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|---------------|-------|--------|--------------|---------------------|--------|-----------|---------|---------|---------|---------|---------|
| g4dn.xlarge   |  4    | x86_64 |  16          | Up to 25 Gigabit    | 1 * T4 |  0.736    |   0.515 |   0.442 |   0.368 |   0.294 |   0.221 |
| g4dn.2xlarge  |  8    | x86_64 |  32          | Up to 25 Gigabit    | 1 * T4 |  1.052    |   0.736 |   0.631 |   0.526 |   0.421 |   0.316 |
| g4dn.4xlarge  | 16    | x86_64 |  64          | Up to 25 Gigabit    | 1 * T4 |  1.685    |   1.179 |   1.011 |   0.843 |   0.674 |   0.505 |
| g4dn.8xlarge  | 32    | x86_64 | 128          | 50 Gigabit          | 1 * T4 |  3.045    |   2.131 |   1.827 |   1.522 |   1.218 |   0.913 |
| g4dn.12xlarge | 48    | x86_64 | 192          | 50 Gigabit          | 4 * T4 |  5.474    |   3.832 |   3.284 |   2.737 |   2.190 |   1.642 |
| g4dn.16xlarge | 64    | x86_64 | 256          | 50 Gigabit          | 1 * T4 |  6.089    |   4.262 |   3.653 |   3.045 |   2.436 |   1.827 |
| g4dn.metal    | 96    | x86_64 | 384          | 100 Gigabit         | 8 * T4 | 10.948    |   7.664 |   6.569 |   5.474 |   4.379 |   3.284 |

1.2 Region eu-central-1 (Frankfurt)

A100

Flavors: p4d.*

Not available in Frankfurt at the time of this writing (2023-06).

| Instance type | vCPUs | Architecture | Memory (GiB) | Network performance | GPUs     |
|---------------|-------|--------------|--------------|---------------------|----------|
| p4d.24xlarge  | 96    | x86_64       | 1152         | 4x 100 Gigabit      | 8        |

Reference prices & discount prices

TODO.

V100

Flavors: p3.*

Not available in Frankfurt at the time of this writing (2023-06).

| Instance type | vCPUs | Architecture | Memory (GiB) | Network performance | GPUs     |
|---------------|-------|--------------|--------------|---------------------|----------|
| p3.2xlarge    | 8     | x86_64       |  61          | Up to 10 Gigabit    | 1 * V100 |
| p3.8xlarge    | 32    | x86_64       | 244          | 10 Gigabit          | 4 * V100 |
| p3.16xlarge   | 64    | x86_64       | 488          | 25 Gigabit          | 8 * V100 |

Reference prices & discount prices

On-demand OS=Linux prices, in USD/hour:

| Instance type | vCPUs | Arch   | Memory (GiB) | Network performance | GPUs     | Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|---------------|-------|--------|--------------|---------------------|----------|-----------|---------|---------|---------|---------|---------|
| p3.2xlarge    | 8     | x86_64 |  61          | Up to 10 Gigabit    | 1 * V100 | 3.823     |   2.676 |   2.294 |   1.911 |   1.529 |   1.147 |
| p3.8xlarge    | 32    | x86_64 | 244          | 10 Gigabit          | 4 * V100 | 15.292    |  10.704 |   9.175 |   7.646 |   6.117 |   4.588 |
| p3.16xlarge   | 64    | x86_64 | 488          | 25 Gigabit          | 8 * V100 | 30.584    |  21.409 |  18.350 |  15.292 |  12.234 |   9.175 |

T4

Flavors: g4dn.*

| Instance type | vCPUs | Architecture | Memory (GiB) | Network performance | GPUs   |
|---------------|-------|--------------|--------------|---------------------|--------|
| g4dn.xlarge   |  4    | x86_64       |  16          | Up to 25 Gigabit    | 1 * T4 |
| g4dn.2xlarge  |  8    | x86_64       |  32          | Up to 25 Gigabit    | 1 * T4 |
| g4dn.4xlarge  | 16    | x86_64       |  64          | Up to 25 Gigabit    | 1 * T4 |
| g4dn.8xlarge  | 32    | x86_64       | 128          | 50 Gigabit          | 1 * T4 |
| g4dn.12xlarge | 48    | x86_64       | 192          | 50 Gigabit          | 4 * T4 |
| g4dn.16xlarge | 64    | x86_64       | 256          | 50 Gigabit          | 1 * T4 |
| g4dn.metal    | 96    | x86_64       | 384          | 100 Gigabit         | 8 * T4 |

Reference prices & discount prices

On-demand OS=Linux prices, in USD/hour:

| Instance type | vCPUs | Arch   | Memory (GiB) | Network performance | GPUs   | Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|---------------|-------|--------|--------------|---------------------|--------|-----------|---------|---------|---------|---------|---------|
| g4dn.xlarge   |  4    | x86_64 |  16          | Up to 25 Gigabit    | 1 * T4 | 0.658     |   0.461 |   0.395 |   0.329 |   0.263 |   0.197 |
| g4dn.2xlarge  |  8    | x86_64 |  32          | Up to 25 Gigabit    | 1 * T4 | 1.308     |   0.916 |   0.785 |   0.654 |   0.523 |   0.392 |
| g4dn.4xlarge  | 16    | x86_64 |  64          | Up to 25 Gigabit    | 1 * T4 | 1.505     |   1.053 |   0.903 |   0.752 |   0.602 |   0.451 |
| g4dn.8xlarge  | 32    | x86_64 | 128          | 50 Gigabit          | 1 * T4 | 4.192     |   2.934 |   2.515 |   2.096 |   1.677 |   1.258 |
| g4dn.12xlarge | 48    | x86_64 | 192          | 50 Gigabit          | 4 * T4 | 5.015     |   3.510 |   3.009 |   2.507 |   2.006 |   1.504 |
| g4dn.16xlarge | 64    | x86_64 | 256          | 50 Gigabit          | 1 * T4 | 5.440     |   3.808 |   3.264 |   2.720 |   2.176 |   1.632 |
| g4dn.metal    | 96    | x86_64 | 384          | 100 Gigabit         | 8 * T4 | 9.780     |   6.846 |   5.868 |   4.890 |   3.912 |   2.934 |

2 AlibabaCloud (as of 2023.06)

2.1 Region cn-shanghai (Shanghai)

Data from [2].

2.1.1 A100

Flavors: ecs.gn7*

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model)   | GPU Memory| Price (按量/pay-as-you-go) | Price (包月/subscription) |
|:------------------------|:------|:---------|:--------------------|:----------|:---------------------------|:--------------------------|
| ecs.gn7e-c16g1.4xlarge  |  16   |  125 GiB | 1 * NVIDIA A100 80G | 1 * 80GB  |  ￥34.74/Hour              | ￥16676.0/Month |
| ecs.gn7-c12g1.3xlarge   |  12   |   94 GiB | 1 * NVIDIA A100     | 1 * 40GB  |  ￥31.58/Hour              | ￥15160.0/Month |
| ecs.gn7e-c16g1.16xlarge |  64   |  500 GiB | 4 * NVIDIA A100 80G | 4 * 80GB  | ￥138.96/Hour              | ￥66704.0/Month |
| ecs.gn7e-c16g1.32xlarge | 128   | 1000 GiB | 8 * NVIDIA A100 80G | 8 * 80GB  | ￥277.93/Hour              |￥133408.0/Month |
| ecs.gn7-c13g1.13xlarge  |  52   |  378 GiB | 4 * NVIDIA A100     | 4 * 40GB  | ￥126.33/Hour              | ￥60640.0/Month |
| ecs.gn7-c13g1.26xlarge  | 104   |  756 GiB | 8 * NVIDIA A100     | 8 * 40GB  | ￥252.66/Hour              |￥121280.0/Month |
| ecs.ebmgn7e.32xlarge    | 128   | 1024 GiB | 8 * NVIDIA A100 80G | 8 * 80GB  | ￥277.93/Hour              |￥133408.0/Month |

Reference prices & discount prices

Pay-as-you-go prices (按量计费), RMB/hour:

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model)   | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:------------------------|:------|:---------|:--------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| ecs.gn7e-c16g1.4xlarge  |  16   |  125 GiB | 1 * NVIDIA A100 80G | 1 * 80GB  |  34.7     |  24.3   |  20.8   |  17.4   |  13.9   |  10.4   |
| ecs.gn7-c12g1.3xlarge   |  12   |   94 GiB | 1 * NVIDIA A100     | 1 * 40GB  |  31.5     |  22.0   |  18.9   |  15.8   |  12.6   |   9.4   |
| ecs.gn7e-c16g1.16xlarge |  64   |  500 GiB | 4 * NVIDIA A100 80G | 4 * 80GB  | 138.9     |  97.2   |  83.3   |  69.5   |  55.6   |  41.7   |
| ecs.gn7e-c16g1.32xlarge | 128   | 1000 GiB | 8 * NVIDIA A100 80G | 8 * 80GB  | 277.9     | 194.5   | 166.7   | 138.9   | 111.2   |  83.4   |
| ecs.gn7-c13g1.13xlarge  |  52   |  378 GiB | 4 * NVIDIA A100     | 4 * 40GB  | 126.3     |  88.4   |  75.8   |  63.1   |  50.5   |  37.9   |
| ecs.gn7-c13g1.26xlarge  | 104   |  756 GiB | 8 * NVIDIA A100     | 8 * 40GB  | 252.6     | 176.8   | 151.6   | 126.3   | 101.0   |  75.8   |
| ecs.ebmgn7e.32xlarge    | 128   | 1024 GiB | 8 * NVIDIA A100 80G | 8 * 80GB  | 277.9     | 194.5   | 166.7   | 138.9   | 111.2   |  83.4   |

Subscription prices (包月), RMB/month:

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model)   | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:------------------------|:------|:---------|:--------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| ecs.gn7e-c16g1.4xlarge  |  16   |  125 GiB | 1 * NVIDIA A100 80G | 1 * 80GB  |  16676    | 11673.2 | 10005.6 |  8338.0 |  6670.4 |  5002.8 |
| ecs.gn7-c12g1.3xlarge   |  12   |   94 GiB | 1 * NVIDIA A100     | 1 * 40GB  |  15160    | 10612.0 |  9096.0 |  7580.0 |  6064.0 |  4548.0 |
| ecs.gn7e-c16g1.16xlarge |  64   |  500 GiB | 4 * NVIDIA A100 80G | 4 * 80GB  |  66704    | 46692.8 | 40022.4 | 33352.0 | 26681.6 | 20011.2 |
| ecs.gn7e-c16g1.32xlarge | 128   | 1000 GiB | 8 * NVIDIA A100 80G | 8 * 80GB  | 133408    | 93385.6 | 80044.8 | 66704.0 | 53363.2 | 40022.4 |
| ecs.gn7-c13g1.13xlarge  |  52   |  378 GiB | 4 * NVIDIA A100     | 4 * 40GB  |  60640    | 42448.0 | 36384.0 | 30320.0 | 24256.0 | 18192.0 |
| ecs.gn7-c13g1.26xlarge  | 104   |  756 GiB | 8 * NVIDIA A100     | 8 * 40GB  | 121280    | 84896.0 | 72768.0 | 60640.0 | 48512.0 | 36384.0 |
| ecs.ebmgn7e.32xlarge    | 128   | 1024 GiB | 8 * NVIDIA A100 80G | 8 * 80GB  | 133408    | 93385.6 | 80044.8 | 66704.0 | 53363.2 | 40022.4 |

2.1.2 V100

Flavors: ecs.gn6[v|e]*

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model)     | GPU Memory| Price (按量/pay-as-you-go) | Price (包月/subscription) |
|:------------------------|:------|:---------|:----------------------|:----------|:---------------------------|:--------------------------|
| ecs.gn6v-c8g1.2xlarge   |  8    |  32 GiB  | 1 * NVIDIA V100       | 1 * 16 GB |  ￥26.46/Hour              |  ￥7620.0/Month |
| ecs.gn6v-c8g1.8xlarge   | 32    | 128 GiB  | 4 * NVIDIA V100       | 4 * 16 GB | ￥105.84/Hour              | ￥30480.0/Month |
| ecs.gn6v-c8g1.16xlarge  | 64    | 256 GiB  | 8 * NVIDIA V100       | 8 * 16 GB | ￥211.68/Hour              | ￥60960.0/Month |
| ecs.gn6v-c10g1.20xlarge | 82    | 336 GiB  | 8 * NVIDIA V100       | 8 * 16 GB | ￥219.64/Hour              | ￥63255.0/Month |
| ecs.gn6e-c12g1.3xlarge  | 12    |  92 GiB  | 1 * NVIDIA V100       | 1 * 32 GB |  ￥19.73/Hour              |  ￥9475.0/Month |
| ecs.gn6e-c12g1.12xlarge | 48    | 368 GiB  | 4 * NVIDIA V100       | 4 * 32 GB |  ￥78.95/Hour              | ￥37900.0/Month |
| ecs.ebmgn6v.24xlarge    | 96    | 384 GiB  | 8 * Nvidia Tesla V100 | 8 * 16 GB | ￥237.12/Hour              | ￥68292.0/Month |
| ecs.gn6e-c12g1.24xlarge | 96    | 736 GiB  | 8 * NVIDIA V100       | 8 * 32 GB | ￥157.91/Hour              | ￥75800.0/Month |

Reference prices & discount prices

Pay-as-you-go prices (按量计费), RMB/hour:

| Instance Type (flavor)       | vCPUs | Memory   | GPU (Count*Model)    | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:-----------------------------|:------|:---------|:---------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| gn6v ecs.gn6v-c8g1.2xlarge   |  8    |  32 GiB  | 1 * NVIDIA V100      | 1 * 16 GB |  26.46    |    18.5 |    15.9 |    13.2 |    10.6 |     7.9 |
| gn6v ecs.gn6v-c8g1.8xlarge   | 32    | 128 GiB  | 4 * NVIDIA V100      | 4 * 16 GB | 105.84    |    74.1 |    63.5 |    52.9 |    42.3 |    31.8 |
| gn6v ecs.gn6v-c8g1.16xlarge  | 64    | 256 GiB  | 8 * NVIDIA V100      | 8 * 16 GB | 211.68    |   148.2 |   127.0 |   105.8 |    84.7 |    63.5 |
| gn6v ecs.gn6v-c10g1.20xlarge | 82    | 336 GiB  | 8 * NVIDIA V100      | 8 * 16 GB | 219.64    |   153.7 |   131.8 |   109.8 |    87.9 |    65.9 |
| gn6e ecs.gn6e-c12g1.3xlarge  | 12    |  92 GiB  | 1 * NVIDIA V100      | 1 * 32 GB |  19.73    |    13.8 |    11.8 |     9.9 |     7.9 |     5.9 |
| gn6e ecs.gn6e-c12g1.12xlarge | 48    | 368 GiB  | 4 * NVIDIA V100      | 4 * 32 GB |  78.95    |    55.3 |    47.4 |    39.5 |    31.6 |    23.7 |
| ecs.ebmgn6v.24xlarge         | 96    | 384 GiB  | 8 * Nvidia Tesla V100| 8 * 16 GB | 237.12    |   166.0 |   142.3 |   118.6 |    94.8 |    71.1 |
| ecs.gn6e-c12g1.24xlarge      | 96    | 736 GiB  | 8 * NVIDIA V100      | 8 * 32 GB | 157.91    |   110.5 |    94.7 |    79.0 |    63.2 |    47.4 |

Subscription prices (包月), RMB/month:

| Instance Type (flavor)       | vCPUs | Memory   | GPU (Count*Model)    | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:-----------------------------|:------|:---------|:---------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| gn6v ecs.gn6v-c8g1.2xlarge   |  8    |  32 GiB  | 1 * NVIDIA V100      | 1 * 16 GB |  7620.0   |  5334.0 |  4572.0 |  3810.0 |  3048.0 |  2286.0 |
| gn6v ecs.gn6v-c8g1.8xlarge   | 32    | 128 GiB  | 4 * NVIDIA V100      | 4 * 16 GB | 30480.0   | 21336.0 | 18288.0 | 15240.0 | 12192.0 |  9144.0 |
| gn6v ecs.gn6v-c8g1.16xlarge  | 64    | 256 GiB  | 8 * NVIDIA V100      | 8 * 16 GB | 60960.0   | 42672.0 | 36576.0 | 30480.0 | 24384.0 | 18288.0 |
| gn6v ecs.gn6v-c10g1.20xlarge | 82    | 336 GiB  | 8 * NVIDIA V100      | 8 * 16 GB | 63255.0   | 44278.5 | 37953.0 | 31627.5 | 25302.0 | 18976.5 |
| gn6e ecs.gn6e-c12g1.3xlarge  | 12    |  92 GiB  | 1 * NVIDIA V100      | 1 * 32 GB |  9475.0   |  6632.5 |  5685.0 |  4737.5 |  3790.0 |  2842.5 |
| gn6e ecs.gn6e-c12g1.12xlarge | 48    | 368 GiB  | 4 * NVIDIA V100      | 4 * 32 GB | 37900.0   | 26530.0 | 22740.0 | 18950.0 | 15160.0 | 11370.0 |
| ecs.ebmgn6v.24xlarge         | 96    | 384 GiB  | 8 * Nvidia Tesla V100| 8 * 16 GB | 68292.0   | 47804.4 | 40975.2 | 34146.0 | 27316.8 | 20487.6 |
| ecs.gn6e-c12g1.24xlarge      | 96    | 736 GiB  | 8 * NVIDIA V100      | 8 * 32 GB | 75800.0   | 53060.0 | 45480.0 | 37900.0 | 30320.0 | 22740.0 |

2.1.3 T4

Flavors: ecs.gn6i*

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model) | GPU Memory| Price (按量/pay-as-you-go) | Price (包月/subscription) |
|:------------------------|:------|:---------|:------------------|:----------|:---------------------------|:--------------------------|
| ecs.gn6i-c4g1.xlarge    |  4    |  15 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | ￥11.63/Hour               |  ￥3348.0/Month |
| ecs.gn6i-c8g1.2xlarge   |  8    |  31 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | ￥14.00/Hour               |  ￥4032.0/Month |
| ecs.gn6i-c16g1.4xlarge  | 16    |  62 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | ￥16.41/Hour               |  ￥4725.0/Month |
| ecs.gn6i-c24g1.6xlarge  | 24    |  93 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | ￥17.19/Hour               |  ￥4950.0/Month |
| ecs.gn6i-c40g1.10xlarge | 40    | 155 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | ￥14.81/Hour               |  ￥7112.9/Month |
| ecs.gn6i-c24g1.12xlarge | 48    | 186 GiB  | 2 * NVIDIA T4     | 2 * 16 GB | ￥34.38/Hour               |  ￥9900.0/Month |
| ecs.gn6i-c24g1.24xlarge | 96    | 372 GiB  | 4 * NVIDIA T4     | 4 * 16 GB | ￥68.75/Hour               | ￥19800.0/Month |
| ecs.ebmgn6i.24xlarge    | 96    | 384 GiB  | 4 * NVIDIA T4     | 4 * 16 GB | ￥68.75/Hour               | ￥19800.0/Month |

Reference prices & discount prices

Pay-as-you-go prices (按量计费), RMB/hour:

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model) | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:------------------------|:------|:---------|:------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| ecs.gn6i-c4g1.xlarge    |  4    |  15 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | 11.63     |     8.1 |     7.0 |     5.8 |     4.7 |     3.5 |
| ecs.gn6i-c8g1.2xlarge   |  8    |  31 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | 14.00     |     9.8 |     8.4 |     7.0 |     5.6 |     4.2 |
| ecs.gn6i-c16g1.4xlarge  | 16    |  62 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | 16.41     |    11.5 |     9.8 |     8.2 |     6.6 |     4.9 |
| ecs.gn6i-c24g1.6xlarge  | 24    |  93 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | 17.19     |    12.0 |    10.3 |     8.6 |     6.9 |     5.2 |
| ecs.gn6i-c40g1.10xlarge | 40    | 155 GiB  | 1 * NVIDIA T4     | 1 * 16 GB | 14.81     |    10.4 |     8.9 |     7.4 |     5.9 |     4.4 |
| ecs.gn6i-c24g1.12xlarge | 48    | 186 GiB  | 2 * NVIDIA T4     | 2 * 16 GB | 34.38     |    24.1 |    20.6 |    17.2 |    13.8 |    10.3 |
| ecs.gn6i-c24g1.24xlarge | 96    | 372 GiB  | 4 * NVIDIA T4     | 4 * 16 GB | 68.75     |    48.1 |    41.2 |    34.4 |    27.5 |    20.6 |
| ecs.ebmgn6i.24xlarge    | 96    | 384 GiB  | 4 * NVIDIA T4     | 4 * 16 GB | 68.75     |    48.1 |    41.2 |    34.4 |    27.5 |    20.6 |

Subscription prices (包月), RMB/month:

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model) | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:------------------------|:------|:---------|:------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| ecs.gn6i-c4g1.xlarge    |  4    |  15 GiB  | 1 * NVIDIA T4     | 1 * 16 GB |   3348.0  |  2343.6 |  2008.8 |  1674.0 |  1339.2 |  1004.4 |
| ecs.gn6i-c8g1.2xlarge   |  8    |  31 GiB  | 1 * NVIDIA T4     | 1 * 16 GB |   4032.0  |  2822.4 |  2419.2 |  2016.0 |  1612.8 |  1209.6 |
| ecs.gn6i-c16g1.4xlarge  | 16    |  62 GiB  | 1 * NVIDIA T4     | 1 * 16 GB |   4725.0  |  3307.5 |  2835.0 |  2362.5 |  1890.0 |  1417.5 |
| ecs.gn6i-c24g1.6xlarge  | 24    |  93 GiB  | 1 * NVIDIA T4     | 1 * 16 GB |   4950.0  |  3465.0 |  2970.0 |  2475.0 |  1980.0 |  1485.0 |
| ecs.gn6i-c40g1.10xlarge | 40    | 155 GiB  | 1 * NVIDIA T4     | 1 * 16 GB |   7112.9  |  4979.0 |  4267.7 |  3556.4 |  2845.2 |  2133.9 |
| ecs.gn6i-c24g1.12xlarge | 48    | 186 GiB  | 2 * NVIDIA T4     | 2 * 16 GB |   9900.0  |  6930.0 |  5940.0 |  4950.0 |  3960.0 |  2970.0 |
| ecs.gn6i-c24g1.24xlarge | 96    | 372 GiB  | 4 * NVIDIA T4     | 4 * 16 GB |  19800.0  | 13860.0 | 11880.0 |  9900.0 |  7920.0 |  5940.0 |
| ecs.ebmgn6i.24xlarge    | 96    | 384 GiB  | 4 * NVIDIA T4     | 4 * 16 GB |  19800.0  | 13860.0 | 11880.0 |  9900.0 |  7920.0 |  5940.0 |

2.1.4 A10

Flavors: ecs.gn7i*

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model) | GPU Memory| Price (按量/pay-as-you-go) | Price (包月/subscription) |
|:------------------------|:------|:---------|:------------------|:----------|:---------------------------|:--------------------------|
| ecs.gn7i-c8g1.2xlarge   |   8   |  30 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |  ￥9.53/Hour               |  ￥4575.66/Month |
| ecs.gn7i-c16g1.4xlarge  |  16   |  60 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | ￥10.09/Hour               |  ￥4844.81/Month |
| ecs.gn7i-c32g1.8xlarge  |  32   | 188 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | ￥13.30/Hour               |  ￥6387.98/Month |
| ecs.gn7i-c48g1.12xlarge |  48   | 310 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | ￥17.94/Hour               |  ￥8613.00/Month |
| ecs.gn7i-c56g1.14xlarge |  56   | 346 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | ￥21.53/Hour               | ￥10335.60/Month |
| ecs.gn7i-c32g1.16xlarge |  64   | 376 GiB  | 2 * NVIDIA A10    | 2 * 24 GB | ￥26.61/Hour               | ￥12775.95/Month |
| ecs.gn7i-c32g1.32xlarge | 128   | 752 GiB  | 4 * NVIDIA A10    | 4 * 24 GB | ￥53.23/Hour               | ￥25551.90/Month |

Reference prices & discount prices

Pay-as-you-go prices (按量计费), RMB/hour:

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model) | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:------------------------|:------|:---------|:------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| ecs.gn7i-c8g1.2xlarge   |   8   |  30 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |  9.53     |   6.671 |   5.718 |   4.765 |   3.812 |   2.859 |
| ecs.gn7i-c16g1.4xlarge  |  16   |  60 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | 10.09     |   7.063 |   6.054 |   5.045 |   4.036 |   3.027 |
| ecs.gn7i-c32g1.8xlarge  |  32   | 188 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | 13.30     |   9.310 |   7.980 |   6.650 |   5.320 |   3.990 |
| ecs.gn7i-c48g1.12xlarge |  48   | 310 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | 17.94     |  12.558 |  10.764 |   8.970 |   7.176 |   5.382 |
| ecs.gn7i-c56g1.14xlarge |  56   | 346 GiB  | 1 * NVIDIA A10    | 1 * 24 GB | 21.53     |  15.071 |  12.918 |  10.765 |   8.612 |   6.459 |
| ecs.gn7i-c32g1.16xlarge |  64   | 376 GiB  | 2 * NVIDIA A10    | 2 * 24 GB | 26.61     |  18.627 |  15.966 |  13.305 |  10.644 |   7.983 |
| ecs.gn7i-c32g1.32xlarge | 128   | 752 GiB  | 4 * NVIDIA A10    | 4 * 24 GB | 53.23     |  37.261 |  31.938 |  26.615 |  21.292 |  15.969 |

Subscription prices (包月), RMB/month:

| Instance Type (flavor)  | vCPUs | Memory   | GPU (Count*Model) | GPU Memory| Ref Price | 30% off | 40% off | 50% off | 60% off | 70% off |
|:------------------------|:------|:---------|:------------------|:----------|:----------|:--------|:--------|:--------|:--------|:--------|
| ecs.gn7i-c8g1.2xlarge   |   8   |  30 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |   4575.66 |  3202.9 |  2745.3 |  2287.8 |  1830.2 | 1372.69 |
| ecs.gn7i-c16g1.4xlarge  |  16   |  60 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |   4844.81 |  3391.3 |  2906.8 |  2422.4 |  1937.9 | 1453.44 |
| ecs.gn7i-c32g1.8xlarge  |  32   | 188 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |   6387.98 |  4471.5 |  3832.7 |  3193.9 |  2555.1 | 1916.39 |
| ecs.gn7i-c48g1.12xlarge |  48   | 310 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |   8613.00 |  6029.1 |  5167.8 |  4306.5 |  3445.2 | 2583.90 |
| ecs.gn7i-c56g1.14xlarge |  56   | 346 GiB  | 1 * NVIDIA A10    | 1 * 24 GB |  10335.60 |  7234.9 |  6201.3 |  5167.8 |  4134.2 | 3100.68 |
| ecs.gn7i-c32g1.16xlarge |  64   | 376 GiB  | 2 * NVIDIA A10    | 2 * 24 GB |  12775.95 |  8943.1 |  7665.5 |  6387.9 |  5110.3 | 3832.78 |
| ecs.gn7i-c32g1.32xlarge | 128   | 752 GiB  | 4 * NVIDIA A10    | 4 * 24 GB |  25551.90 | 17886.3 | 15331.1 | 12775.9 | 10220.7 | 7665.57 |

Appendix

The first letter in GPU model names denote their GPU architectures, with:

  1. T for Turing;
  2. A for Ampere;
  3. V for Volta;

  4. H for Hopper;

    Introduced in 2022, support hardware implementation of Transformer architecture in AI training;

A. Quick comparison of T4/A10/A10G/V100

  T4 A10 A10G A30 V100 PCIe/SMX2
Designed for Data center workloads (Desktop) Graphics-intensive workloads Desktop Desktop Data center
Year 2018 2020     2017
Manufacturing 12nm 12nm 12nm    
Architecture Turing Ampere Ampere Ampere Volta
Max Power 70 watts 150 watts   165 watts 250/300watts
GPU Mem 16GB GDDR6 24GB GDDR6 48GB GDDR6 24GB HBM2 16/32GB HBM2
GPU Mem BW 400 GB/s 600 GB/s   933GB/s 900 GB/s
Interconnect PCIe Gen3 32GB/s PCIe Gen4 66 GB/s   PCIe Gen4 64GB/s, NVLINK 200GB/s PCIe Gen3 32GB/s, NVLINK 300GB/s
FP32 8.1 TFLOPS 31.2 TFLOPS   10.3TFLOPS 14/15.7 TFLOPS
BFLOAT16 TensorCore   125 TFLOPS   165 TFLOPS  
FP16 TensorCore   125 TFLOPS   165 TFLOPS  
INT8 TensorCore   250 TFLOPS   330 TOPS  
INT4 TensorCore       661 TOPS  

Datasheets:

  1. T4
  2. A10
  3. A30
  4. V100-PCIe/V100-SXM2/V100S-PCIe

B. Quick comparison of A100/A800/H100/H800

  A800 (PCIe/SXM) A100 (PCIe/SXM) H800 (PCIe/SXM) H100 (PCIe/SXM)
Year 2022 2020 2022 2022
Manufacturing 7nm 7nm 4nm 4nm
Architecture Ampere Ampere Hopper Hopper
Max Power 300/400 watt 300/400 watt   350/700 watt
GPU Mem 80G HBM2e 80G HBM2e 80G HBM3 80G HBM3
GPU Mem BW   1935/2039 GB/s   2/3.35 TB/s
Interconnect NVLINK 400GB/s PCIe Gen4 64GB/s, NVLINK 600GB/s   PCIe Gen5 128GB/s, NVLINK 900GB/s
FP32   19.5 TFLOPS   51/67 TFLOPS
TF32 (TensorFloat)   156/312 TFLOPS   756/989 TFLOPS
BFLOAT16 TensorCore   156/312 TFLOPS    
FP16 TensorCore   312/624 TFLOPS   1513/1979 TFLOPS
FP8 TensorCore NO support NO support   3026/3958 TFLOPS
INT8 TensorCore   624/1248 TFLOPS   3026/3958 TFLOPS

H100 vs. A100 in one word: 3x performance, 2x price.

Datasheets:

  1. A100
  2. H100

References

  1. AWS EC2 price estimation page
  2. AlibabaCloud ECS price estimation page

Tables in this post are partially generated with this script, and vim.

© 2016-2023 Arthur Chiao, Powered by Jekyll , Theme originated from Long Haul. Site visits: , powered by busuanzi