Price List
☑ Multiple payment methods ☑ No up-front cost ☑ No termination fees
Version: 20240902v1.0
TWSC offers the most competitive and optimal pricing model for your growing business. You are more than welcome to contact us via sales@twsc.io to get the best offer from us.
Per-second Billing
⁕ Calculation example:
The time usage of a c.super instance is 11 hours 35 minutes 10 seconds = 41710 seconds = 11.5861 hours. The list credits for on-demand plan is 60.90 Credits/hour. Therefore, the cost is 11.5861 * 60.90 = 705.5934, a total of 705.5934 credits.
Notes
AI Foundry Service (AFS)
AFS ModelSpace
Public Mode
| |||
Model series⁕ | Type (Input / Output) | Credits⁕⁕ (Per thousand tokens) | |
Meta-Llama3.1 | AFS.MS.inp.Meta-Llama3.1-70B-32K | 0.0315 | |
AFS.MS.otp.Meta-Llama3.1-70B-32K | |||
AFS.MS.inp.Meta-Llama3.1-8B-32K | 0.00525 | ||
AFS.MS.otp.Meta-Llama3.1-8B-32K | |||
Llama3-FFM | AFS.MS.inp.Llama3-FFM-70B | 0.0315 | |
AFS.MS.otp.Llama3-FFM-70B | |||
AFS.MS.inp.Llama3-FFM-8B | 0.00525 | ||
AFS.MS.otp.Llama3-FFM-8B | |||
FFM-Mistral | AFS.MS.inp.FFM-Mixtral-8x7B | 0.0168 | |
AFS.MS.otp.FFM-Mixtral-8x7B | |||
AFS.MS.inp.FFM-Mistral-7B | 0.00525 | ||
AFS.MS.otp.FFM-Mistral-7B | |||
FFM-Llama2-v2 | AFS.MS.inp.FFM-Llama2-v2-70B | 0.0315 | |
AFS.MS.otp.FFM-Llama2-v2-70B | |||
AFS.MS.inp.FFM-Llama2-v2-13B | 0.0105 | ||
AFS.MS.otp.FFM-Llama2-v2-13B | |||
AFS.MS.inp.FFM-Llama2-v2-7B | 0.00525 | ||
AFS.MS.otp.FFM-Llama2-v2-7B | |||
FFM-embedding | AFS.MS.inp.FFM-embedding⁕⁕⁕ | 0.0021 | |
GTE | AFS.MS.inp.GTE-Base-zh⁕⁕⁕ | 0.0021 | |
BAAI-bge-reranker | AFS.MS.inp.bge-reranker-v2-m3⁕⁕⁕ | 0.0021 | |
DeepSeek-Coder | AFS.MS.inp.DeepSeek-Coder-6_7B-base | 0.00525 | |
AFS.MS.otp.DeepSeek-Coder-6_7B-base |
info
⁕Models with the prefix "FFM" indicate the traditional Chinese-enhanced new weights by TWSC.
⁕⁕Public mode is billed by tokens, per thousand tokens as a billing unit, and it will be calculated as 1,000 tokens when the usage is less than 1,000 tokens.
⁕⁕⁕Both the input and output token usage of the model will be incorporated into the input product measurement billing.
⁕⁕Public mode is billed by tokens, per thousand tokens as a billing unit, and it will be calculated as 1,000 tokens when the usage is less than 1,000 tokens.
⁕⁕⁕Both the input and output token usage of the model will be incorporated into the input product measurement billing.
Private Mode
| |||
Model series⁕ | Type | Deployment spec | Credits⁕⁕ (Per hour) |
Llama3-FFM | AFS.MS.Llama3-FFM-70B | Standard | 283.50 |
AFS.MS.Llama3-FFM-8B | Standard | 60.90 | |
FFM-Mistral | AFS.MS.FFM-Mixtral-8x7B | Standard | 283.50 |
AFS.MS.FFM-Mistral-7B | Standard | 147.00 | |
AFS.MS.FFM-Mistral-7B-4K | Standard | 60.90 | |
FFM-Llama2-v2 | AFS.MS.FFM-Llama2-v2-70B | Standard | 278.25 |
AFS.MS.FFM-Llama2-v2-13B | Standard | 89.25 | |
AFS.MS.FFM-Llama2-v2-7B | Standard | 57.75 | |
FFM-Llama2 | AFS.MS.FFM-Llama2-70B | Standard | 273.00 |
AFS.MS.FFM-Llama2-13B | Standard | 84.00 | |
AFS.MS.FFM-Llama2-7B | Standard | 52.50 | |
FFM-embedding | AFS.MS.FFM-embedding | Standard | 52.50 |
GTE | AFS.MS.GTE-Base-zh | Standard | 52.50 |
TAIDE | AFS.MS.TAIDE-LX-7B | Standard | 57.75 |
Meta-Llama3.1 | AFS.MS.Meta-Llama3.1-70B-32K | Standard | 283.50 |
AFS.MS.Meta-Llama3.1-8B-32K | Standard | 60.90 | |
Meta-CodeLlama | AFS.MS.Meta-CodeLlama-34B-8K | Standard | 273.00 |
AFS.MS.Meta-CodeLlama-13B-8K | Standard | 168.00 | |
AFS.MS.Meta-CodeLlama-7B-8K | Standard | 105.00 | |
Meta-Llama2 | AFS.MS.Meta-Llama2-70B | Standard | 273.00 |
AFS.MS.Meta-Llama2-13B | Standard | 84.00 | |
AFS.MS.Meta-Llama2-7B | Standard | 52.50 | |
BAAI-bge-reranker | AFS.MS.bge-reranker-v2-m3 | Standard | 52.50 |
DeepSeek-Coder | AFS.MS.DeepSeek-Coder-6_7B-base | Standard | 52.50 |
info
⁕Models with the prefix "FFM" indicate the traditional Chinese-enhanced new weights by TWSC.
AFS Platform
| ||||
Fine-tuning Strategy | Model⁕ | Type | Performance⁕⁕⁕ (Million tokens/Per hour) | Credits (Per hour) |
Full parameter | Llama3-FFM | AFS.plt.Llama3-FFM-70B.22M | 22 | 21,000.00 |
AFS.plt.Llama3-FFM-8B.26M | 26 | 2,625.00 | ||
FFM-Mistral | AFS.plt.FFM-Mixtral-8x7B.17M | 17 | 10,500.00 | |
AFS.plt.FFM-Mistral-7B.26M | 26 | 2,625.00 | ||
FFM-Llama2-v2 | AFS.plt.FFM-Llama2-v2-70B.12M | 12 | 10,500.00 | |
AFS.plt.FFM-Llama2-v2-13B.21M | 21 | 5,250.00 | ||
AFS.plt.FFM-Llama2-v2-7B.24M | 24 | 2,625.00 | ||
FFM-Llama2⁕⁕ | AFS.plt.FFM-Llama2-70B.12M | 12 | 10,500.00 | |
AFS.plt.FFM-Llama2-13B.21M | 21 | 5,250.00 | ||
AFS.plt.FFM-Llama2-7B.24M | 24 | 2,625.00 | ||
Meta-Llama2⁕⁕ | AFS.plt.Meta-Llama2-70B.12M | 12 | 10,500.00 | |
AFS.plt.Meta-Llama2-13B.21M | 21 | 5,250.00 | ||
AFS.plt.Meta-Llama2-7B.24M | 24 | 2,625.00 | ||
LoRA(Low-Rank Adaptation) | Llama3-FFM | AFS.plt.Llama3-FFM-LoRA-70B.10M | 10 | 10,500.00 |
AFS.plt.Llama3-FFM-LoRA-8B.23M | 23 | 2,625.00 | ||
FFM-Mistral | AFS.plt.FFM-Mixtral-LoRA-8x7B.22M | 22 | 10,500.00 | |
AFS.plt.FFM-Mistral-LoRA-7B.35M | 35 | 2,625.00 | ||
FFM-Llama2-v2 | AFS.plt.FFM-Llama2-v2-LoRA-70B.16M | 16 | 10,500.00 | |
AFS.plt.FFM-Llama2-v2-LoRA-13B.29M | 29 | 5,250.00 | ||
AFS.plt.FFM-Llama2-v2-LoRA-7B.31M | 31 | 2,625.00 | ||
FFM-Llama2 | AFS.plt.FFM-Llama2-LoRA-70B.16M | 16 | 10,500.00 | |
AFS.plt.FFM-Llama2-LoRA-13B.29M | 29 | 5,250.00 | ||
AFS.plt.FFM-Llama2-LoRA-7B.31M | 31 | 2,625.00 | ||
Meta-Llama2 | AFS.plt.Meta-Llama2-LoRA-70B.16M | 16 | 10,500.00 | |
AFS.plt.Meta-Llama2-LoRA-13B.29M | 29 | 5,250.00 | ||
AFS.plt.Meta-Llama2-LoRA-7B.31M | 31 | 2,625.00 |
info
⁕Models with the prefix "FFM" represent the new weight versions of those models with enhanced support for Taiwan Traditional Chinese.
⁕⁕The number of tokens varies from models to models. Meta-Llama2 and FFM-Llama2 model series improve performance to 1 Chinese character with 2 tokens; while other models represent about 1 Chinese character with 1 token.
⁕⁕⁕The performance specifications are based on completing 2 epochs.
⁕⁕The number of tokens varies from models to models. Meta-Llama2 and FFM-Llama2 model series improve performance to 1 Chinese character with 2 tokens; while other models represent about 1 Chinese character with 1 token.
⁕⁕⁕The performance specifications are based on completing 2 epochs.
AFS Cloud
| |||
Model series⁕ | Type | Deployment spec | Credits (Per hour) |
Llama3-FFM | AFS.cld.Llama3-FFM-70B.ft | Standard | 283.50 |
AFS.cld.Llama3-FFM-8B.ft | Standard | 60.90 | |
FFM-Mistral | AFS.cld.FFM-Mixtral-8x7B.ft | Standard | 283.50 |
AFS.cld.FFM-Mistral-7B.ft | Standard | 147.00 | |
FFM-Llama2-v2 | AFS.cld.FFM-Llama2-v2-70B.ft | Standard | 278.25 |
AFS.cld.FFM-Llama2-v2-13B.ft | Standard | 89.25 | |
AFS.cld.FFM-Llama2-v2-7B.ft | Standard | 57.75 | |
FFM-Llama2 | AFS.cld.FFM-Llama2-70B.ft | Standard | 273.00 |
AFS.cld.FFM-Llama2-13B.ft | Standard | 84.00 | |
AFS.cld.FFM-Llama2-7B.ft | Standard | 52.50 | |
Meta-Llama2 | AFS.cld.Meta-Llama2-70B.ft | Standard | 273.00 |
AFS.cld.Meta-Llama2-13B.ft | Standard | 84.00 | |
AFS.cld.Meta-Llama2-7B.ft | Standard | 52.50 |
info
⁕Models with the prefix "FFM" indicate the Taiwan Traditional Chinese-enhanced new weight versions of those models by TWSC.
Enterprise Brain Computing Plan
| ||||
Plan | Contract duration | Price (NTD) | Annual payment (NTD) | Total usable computing power (NTD) |
Enterprise Brain 95 | One year | 950,000 | 950,000 | 1,090,000 |
Three year | 2,850,000 | 950,000 | 3,420,000 | |
Enterprise Brain 150 | One year | 1,500,000 | 1,500,000 | 1,800,000 |
Three year | 4,500,000 | 1,500,000 | 5,700,000 | |
Enterprise Brain 250 | One year | 2,500,000 | 2,500,000 | 3,300,000 |
Three year | 7,500,000 | 2,500,000 | 10,500,000 |
Artificial Intelligence: OneAI
| |||||||
Type | Model | Spec | Unit | Credits | |||
Subscription | oai.subscription | Cycle billing | Per month | ⁕1050.00 | |||
Storage | oai.storage | OneAI standard storage | Per GB-month | 1.26 | |||
Service | oai.annot | Standard annotation service | Per hour | 7.35 | |||
Service | oai.ibs.cstd | Standard image builder service-CPU | Per hour | 8.7675 | |||
Service | oai.ibs.gstd | Standard image builder service-GPU | Per hour | 21.00 | |||
Type | Model | GPU V100 (pcs) | CPU | Memory (GB) | Shared Memory (GB) | Credits (Per hour) | |
Training | oai.train1v | 1 | 4 | 90 | - | 60.90 | |
oai.train2v | 2 | 8 | 180 | - | 121.80 | ||
oai.train4v | 4 | 16 | 360 | - | 243.60 | ||
oai.train8v | 8 | 32 | 720 | - | 487.20 | ||
oai.mtrain1v | 1 | 4 | 60 | 30 | 60.90 | ||
oai.mtrain2v | 2 | 8 | 120 | 60 | 121.80 | ||
oai.mtrain4v | 4 | 16 | 240 | 120 | 243.60 | ||
oai.mtrain8v | 8 | 32 | 480 | 240 | 487.20 | ||
oai.m1train8v | 8 | 32 | 360 | 360 | 487.20 | ||
Type | Model | GPU T4 (pcs) | vCPU | Memory (GB) | Credits (Per hour) | ||
Compute | oai.comp.c2m4 | - | 2 | 4 | 2.1525 | ||
oai.comp.c4m8 | - | 4 | 8 | 4.305 | |||
oai.comp.c8m16 | - | 8 | 16 | 8.61 | |||
oai.comp.c4m16 | - | 4 | 16 | 5.25 | |||
oai.comp.c8m32 | - | 8 | 32 | 10.50 | |||
oai.comp.c16m64 | - | 16 | 64 | 21.00 | |||
oai.comp.c4m64gt1 | 1 | 4 | 64 | 16.80 | |||
oai.comp.c8m64gt1 | 1 | 8 | 64 | 22.05 | |||
oai.comp.c8m128gt2 | 2 | 8 | 128 | 31.50 |
INFO
⁕ The free trial of the subscription service ends on 12/31/2023.
T-Proof Service
| |||
Service | type | Spec | Credits (Per upload) |
Notarization service | tpf.notarize.std | Standard | 5.25 |
Verification service | tpf.verify.std | Standard | 0.00⁕ |
Retrieval service | tpf.retrieve.std | Standard | 0.00⁕ |
info
⁕ Verification and Retrieval services are offered a discount price until December 31, 2024.
Compute
Virtual Compute Service (VCS)
| |||||||||
CPU instance series | Instance type | vCPU | Memory (GB) | HDD (GB) | (Per hour) | ||||
Linux⁕ | Windows Server⁕⁕ | SQL Server Web⁕⁕⁕ | |||||||
v-series | v.super | 2 | 16 | 100 | 2.6985 | 4.20 | 6.5625 | ||
v.xsuper | 4 | 32 | 5.397 | 7.035 | 13.125 | ||||
v.2xsuper | 8 | 64 | 10.794 | 14.07 | 26.25 | ||||
v.4xsuper | 16 | 128 | 21.588 | 28.14 | 52.50 | ||||
v.8xsuper | 32 | 256 | 43.176 | 56.28 | 105.00 | ||||
v.12xsuper | - | 48 | 320 | 100 | 61.68 | ||||
v.ws.12xsuper⁕ | - | 48 | 320 | 100 | 68.26 | ||||
v1-series | v1.super | 2 | 2 | 100 | 1.68 | 2.73 | 4.83 | ||
v1.xsuper | 4 | 4 | 3.36 | 5.46 | 9.66 | ||||
v1.2xsuper | 8 | 8 | 6.72 | 10.92 | 19.32 | ||||
v1.4xsuper | 16 | 16 | 13.44 | 21.84 | 38.64 | ||||
v1.8xsuper | 32 | 32 | 26.88 | 43.68 | 77.28 | ||||
v2-series | v2.super | 2 | 4 | 100 | 1.827 | 2.8875 | 5.04 | ||
v2.xsuper | 4 | 8 | 3.654 | 5.67 | 10.08 | ||||
v2.2xsuper | 8 | 16 | 7.308 | 11.34 | 20.16 | ||||
v2.4xsuper | 16 | 32 | 14.616 | 22.68 | 40.32 | ||||
v2.8xsuper | 32 | 64 | 29.232 | 45.36 | 80.64 | ||||
v4-series | v4.super | 2 | 8 | 100 | 2.1525 | 3.15 | 5.4075 | ||
v4.xsuper | 4 | 16 | 4.305 | 6.30 | 10.815 | ||||
v4.2xsuper | 8 | 32 | 8.61 | 12.60 | 21.63 | ||||
v4.4xsuper | 16 | 64 | 17.22 | 25.20 | 43.26 | ||||
v4.8xsuper | 32 | 128 | 34.44 | 50.40 | 86.52 | ||||
GPU instance series | Instance type | GPU (pcs) | GPU type | vCPU | Memory (GB) | (Per hour) | |||
Linux⁕ | Windows Server⁕⁕ | ||||||||
vgv-series | vgv.xsuper | 1 | NVIDIA Tesla V100/32GB | 8 | 90 | 73.50 | 76.65 | ||
vgv.2xsuper | 2 | 16 | 180 | 147.00 | 153.30 | ||||
vgv.4xsuper | 4 | 32 | 360 | 294.00 | 306.60 | ||||
vgv.8xsuper | 8 | 64 | 720 | 588.00 | 613.20 | ||||
vga1-series⁕⁕⁕⁕ | vga1.8xsuper | 8 | NVIDIA A100/40GB SXM | 240 | 960 | 1060.50 | N/A |
:::info⁕GPU 加速運算個體:提供專門處理圖形任務的 GPU 核心,能快速完成大量運算。
如何取得 GPU 資源? 請參考 FAQ-資源監控與配置-Q4 之說明。 :::
INFO
⁕Ubuntu, CentOS, and Rocky Linux are provided.
⁕⁕Include the license for Windows Server Standard edition.
⁕⁕⁕Include the licenses for Windows Server Standard edition and SQL Server Web edition.
⁕⁕⁕⁕The vga1-series equips with local 11TB SSD and is not available on demand. If you have any needs, please contact customer service.
⁕⁕Include the license for Windows Server Standard edition.
⁕⁕⁕Include the licenses for Windows Server Standard edition and SQL Server Web edition.
⁕⁕⁕⁕The vga1-series equips with local 11TB SSD and is not available on demand. If you have any needs, please contact customer service.
Virtual K8s Service (VKS)
| ||
Compute resources | Type | Credits (Per hour) |
Cluster service | VKS.base.std | 2.73 |
v-series | VKS.v series | Same as VCS v-series |
v1-series | VKS.v1 series | Same as VCS v1-series |
v2-series | VKS.v2 series | Same as VCS v2-series |
v4-series | VKS.v4 series | Same as VCS v4-series |
vgv-series | VKS.vgv series | Same as VCS vgv-series |
Network & security | Type | Credits |
Network Flow (NFL) | NetFlow_CHF_out | 3.15 /GB |
Elastic IP (EIP) | sip.renstatic | 105.00 /IP-month |
sip.entstatic | 105.00 /IP-month | |
Load balancing | Type | Credits (Per hour) |
Virtual K8s Service(VKS) | VKS.lbs.basic | 0.63 |
Storage | Type | Credits |
Virtual K8s Service(VKS) | VKS.hdd.std | 1.26 /GB-month |
Virtual K8s Service(VKS) | VKS.ssd.std | 2.73 /GB-month |
info
Compute nodes are provided with the Ubuntu OS.
Cloud PC (CPC)
| |||||
Instance series | Instance type | vCPU | Memory (GB) | Credits (Per hour) | |
cpc-series | cpc.c2m4 | 2 | 4 | 2.625 |
Container Compute Service (CCS)
| |||||||
Container Series | Container Type | GPU (pcs) | CPU (cores) | Memory (GB)⁕ | Shared Memory (GB)⁕ | Credits (Per hour) | |
Standard AI | c.super | 1 | 4 | 90 | - | 60.90 | |
c.xsuper | 2 | 8 | 180 | - | 121.80 | ||
c.2xsuper | 4 | 16 | 360 | - | 243.60 | ||
c.4xsuper | 8 | 32 | 720 | - | 487.20 | ||
c.8xsuper | 16 | 64 | 1440 | - | 974.40 | ||
Memory Optimized | cm.super | 1 | 4 | 60 | 30 | 60.90 | |
cm.xsuper | 2 | 8 | 120 | 60 | 121.80 | ||
cm.2xsuper | 4 | 16 | 240 | 120 | 243.60 | ||
cm.4xsuper | 8 | 32 | 480 | 240 | 487.20 | ||
cm1.4xsuper | 8 | 32 | 360 | 360 | 487.20 |
INFO
⁕GB in the price table denotes a decimal unit. In the computer system, 90GB is shown as 84 GiB. For more information, see Memory Capacity Conversion.
High-performance Computing (HPC)
| |||||||
Service | Job Type | GPU (pcs) | CPU (cores) | Memory (GB) | Credits (Per hour) | ||
HPC Job | h.super | 1 | 4 | 90 | 60.90 | ||
h.xsuper | 2 | 8 | 180 | 121.80 | |||
h.2xsuper | 4 | 16 | 360 | 243.60 | |||
h.4xsuper | 8 | 32 | 720 | 487.20 | |||
Taiwania2 (HPC CLI) | hpc_cli | - | - | - | 60.90 (Per GPU hour) |
INFO
⁕You can freely specify the number of GPUs, and other resources will be allocated with the ratio of 1 GPU: 4 CPU: 90 GB memory for you.
Example: When you specify 18 GPUs, you allocate 18 GPUs: 72 CPUs: 1620 GB of memory. You will be charged for NTD 1,044(58.00*18) per hour.
⁕If the memory available is calculated in GiB (1 GiB = 230 bytes), taking h.super as an example, the memory available is 84 GiB. For more information, see Memory Capacity ConversionStorage
Cloud Object Storage (COS)
| ||||||||
Spec | Credits (GB-month) | |||||||
Standard object storage | 0.63 |
Virtual Disk Service (VDS)
| ||||||||
Disk type | Spec | Credits (GB-month) | ||||||
Solid State Drives (SSD) | Standard | 2.73 | ||||||
Hard Disk Drives (HDD) | Standard | 1.26 | ||||||
Image static storage⁕⁕ | Basic | 0.63 |
info
⁕ We offer free storage of the first 100 GB for system disks.
⁕⁕ For VCS image static storage. For VCS image in the workspace, please refer to Hard Disk Drives (HDD).
⁕⁕ For VCS image static storage. For VCS image in the workspace, please refer to Hard Disk Drives (HDD).
Hyper File System (HFS)
| |||||||
Storage | Credits (GB) | ||||||
Standard | 4.20 |
Cloud File Service (CFS)
| |||||
Type | Service | Model | Spec | Unit | Credits |
Storage | Hot storage | cfs.store.hot | Hot storage | GB-month | 0.63 |
Cold storage | cfs.store.cold | Cold storage | 0.105 | ||
Data Transfer | Cold to Hot | cfs.data.2hot | Cold to Hot | GB | 0.21 |
Hot to Cold | cfs.data.2cold | Hot to Cold | 0.00 |
Networking & Security
Load Balancing Service (LBS)
| ||||
Spec | Credits (Per hour) | |||
Classic | 0.63 (Billing start date: 7/7/2022) |
Elastic IP (EIP)
| |||||
Type | Model | Spec | Unit | Price | |
Subscription | sip.entstatic | Usage fee of Static IP - Commercial Network | per IP-month | 105.00 | |
sip.renstatic | Maintainence fee of Static IP - Research and Education Network | 105.00 |
INFO
⁕ Example: Subscribe at 11/26 15:00, and cancel subscription at 11/30 21:00. Billing period: 11/26 15:00:00 - 11/26 21:00:00.
⁕⁕ Example: Subscribe at 11/26 15:00, and cancel subscription at 11/30 21:00. Billing period: 11/26 15:00:00 - 11/26 21:00:00.
⁕⁕ Example: Subscribe at 11/26 15:00, and cancel subscription at 11/30 21:00. Billing period: 11/26 15:00:00 - 11/26 21:00:00.
Data Transfer: Virtual Compute Service (NetFlow)
| ||||||||
Transfer direction | Credits (GB) | |||||||
Data transfers in to TWSC VCS from Internet | Free | |||||||
Data transfers out from TWSC VCS to Internet | 3.15 | |||||||
Data transfers between TWSC VCS | Free |
Advanced Virtual Firewall (VFW)
| ||
Model | Spec | Credits (Per hour) |
pa_byol | Standard | 8.61 |