Differences

This shows you the differences between two versions of the page.

--- hpc:hpc_clusters [2026/07/20 12:14] – [AMD CPU] Yann Sagon
+++ hpc:hpc_clusters [2026/10/09 08:14] (current) – [GPUs models available] Yann Sagon
@@ Line 71: / Line 71: @@
 that will use only CPU or GPU nodes.
-===== Cost model =====
-In September 2024, we {{ :hpc:communication_facturation_hpc_2025_vfinale.pdf |announced}} via our mailing list (Some of the information is now obsolete. Please see below for the updated information.) that our service would become a paid offering. Users can find detailed information about this change on the page below and in our [[hpc:faq#cost|FAQ]]. We are committed to providing transparent communication and ensuring you have all the necessary details about our new pricing model.
-Our service remains free in specific cases:
-  * **Free usage** as part of an educational course.
-  * **Free usage** through the annual allocation of CPU hours.
-For additional needs, paid options are available:
+====== Purchase: Hardware Pricing ======
-  * **[[ hpc:hpc_clusters#price_per_hour|Pay-per-hour]]** based on the FNS rate table.
+**CPU and GPU server example pricing **
-  * **Purchase or rent** compute nodes for more intensive workloads.
-**Summary:**
+See below the current price of a compute node (without the extra 15% and without VAT)
-  * Starting this year, you receive a **CPU hours credit** based on the hardware you own (if any) in the cluster (private partition).
+===== AMD CPU=====
-  * You can find instructions on how to check your annual credit here: [[accounting#resources_available_for_research_group|Resources Available for Research Groups]]. If you know your research group has bought some compute nodes but your PI doesn't appear in the report, please contact us.
-  * The credit calculation in the provided script assumes a **5-year hardware ownership period**. However, **if** this policy was introduced after your compute nodes were purchased, we have extended the production duration by two years.
-  * To ensure **flexibility and simplicity**, we have standardized resource usage by converting CPU Memory, and GPU hours into CPU hours, using different conversion ratios depending on the GPU type. More details can be found here: [[accounting#resource_accounting_uniformization|Resource Accounting Uniformization]].
-  * You can use your credit across all three clusters (**Baobab, Yggdrasil, and Bamboo**), not just on your private compute nodes. However, when using your own compute nodes, you will receive a **higher priority**.
-  * To check your group's current resource usage, visit: [[accounting#report_and_statistics_with_sreport|Report and Statistics with sreport]].
-==== Price per hour ====
-<WRAP center round important 60%>
-Since we have [[https://doc.eresearch.unige.ch/hpc/accounting#resource_accounting_uniformization|unified our resources]], we refer to **CPUhours** in a generic way, which may include CPU hours, GPU hours, or memory usage. Please refer to the conversion table for details. This means that any previous references specifically mentioning GPUs should no longer be considered. We'll update the table.
-</WRAP>
-Overview:
-{{:hpc:pasted:20240404-092421.png}}
-You can find the whole table that you can send to the FNS {{:hpc:hpc:acrobat_2024-04-09_15-58-28.png?linkonly|here}}.
-University of Genevea members will be charged the cost indicated by line "U1". The line U2 and U3 are for external users such as company that would use the cluster.
-=== Free CPU Hour Allocation ===
-Each PI (Principal Investigator) is entitled to 100,000 CPU hours per year free of charge. This allocation applies per PI, not per individual user. See [[hpc:accounting|how to check PI and user past usage]]..
-=== Progressive Pricing for HPC Compute Hours ===
-A tiered pricing model applies to compute hour billing. Discounts increase as usage grows: once you reach 200K, 500K, and 1,000K compute hours, an additional 10% reduction is applied at each threshold. This ensures cost efficiency for large-scale workloads.
-^ Usage (Compute Hours) ^ Discount Applied ^
-| 0 – 199,999           | Base Rate       |
-| 200,000 – 499,999     | Base Rate -10%  |
-| 500,000 – 999,999     | Base Rate -20%  |
-| 1,000,000+            | Base Rate -30%  |
-===== Purchasing or Renting Private Compute Nodes =====
-Research groups have the option to purchase or rent "private" compute nodes to expand the resources available in our clusters. This arrangement provides the group with a **private partition**, granting higher priority access to the specified nodes (resulting in reduced wait times) and extended job runtimes of up to **7 days** (compared to 4 days for public compute nodes).
-==== Key Rules and Details ====
-  * **Shared Integration**: The compute node is added to the corresponding shared partition. Other users may utilize it when the owning group is not using it. For details, refer to the [[hpc/slurm#partitions|partitions]] section.
-  * **Usage Limit**: Each research group may consume up to **60% of the theoretical usage credit associated with the compute node**. This policy ensures fair access to shared cluster resources. . See  the [[hpc:hpc_clusters#usage_limits|Usage limits]] policy for more details
-  * **Cost**: In addition to the base cost of the compute node, a **15% surcharge** is applied to cover operational expenses such as cables, racks, switches, and storage (not yet valid).
-  * **Ownership Period**: The compute node remains the property of the research group for **5 years**. After this period, the node may remain in production but will only be accessible via public and shared partitions.
-  * **Warranty and Repairs**: Nodes come with a **3-year warranty**. If the node fails after this period, the research group is responsible for **100% of repair costs**. Repairing the node involves sending it to the vendor for diagnostics and a quote, with a maximum diagnostic fee of **420 CHF**, even if the node is irreparable.
-  * **Administrative Access**: The research group does not have administrative rights over the node.
-  * **Maintenance**: The HPC team handles the installation and maintenance of the compute node, ensuring it operates consistently with other nodes in the cluster.
-  * **Decommissioning**: The HPC team may decommission the node if it becomes obsolete, but it will remain in production for at least **5 years**.
-==== Cost of Renting a Compute Node ====
-The cost of renting a compute node is calculated based on the vendor price of the node, adjusted to account for operational and infrastructure expenses. Specifically, we add 15% to the vendor price to cover additional costs, such as maintenance and administrative overhead. The total cost is then amortized over an estimated 5-year lifespan of the compute node to determine the monthly rental rate.
-For example, consider a CPU compute node with a vendor price of **14,361 CHF**. Adding **15% for extra costs** brings the total to **16,515.15 CHF**. Dividing this by 60 months (5 years) results in a monthly rental cost of approximately **275.25 CHF**.
-Currently, we only rent standard AMD compute nodes, which have two 64-CPU cores and 512 GB of RAM. You will receive 1.34 million billing credits per year with this model.
-The minimum rental period for a compute node is six months. Any unused allocated resources will be lost at the end of the year.
-For more details or to request a specific quote, please contact the HPC support team.
-==== Usage Limits ====
-Users are entitled to utilize up to 60% of the computational resources they own or rent within the cluster. Example calculation if you rent a compute node with 128 CPU cores and 512GB RAM for one year:
-  * **CPU contribution:** 128 cores × 1.0 (factor)
-  * **Memory contribution:** 512 GB × 0.25 (factor)
-  * **Time period:** 24 hours × 365 days
-  * **Max usage rate:** 0.6
-Total:
-(128 × 1.0 + 512 × 0.25) × 24 × 365 × 0.6 = **1,342,848 core-hours**
-This credit can be used across any of our three clusters -- Bamboo, Baobab, and Yggdrasil -- regardless of where the compute node was rented or purchased.
-The main advantage is that you are not restricted to using your private nodes, but can access the three clusters and even the GPUs.
-We are developing scripts to allow to check the usage and the amount of hours you have the right to use regarding the hardware your group owns.
-The key distinction when using your own resources is that you benefit from a higher scheduling priority, ensuring quicker access to computational resources. As well, you are not restricted to using your private nodes, but can access the three clusters and even the GPUs.
-For more details, please contact the HPC support team.
-==== CPU and GPU server example pricing ====
-See below the current price of a compute node (without the extra 15% and without VAT)
-=== AMD CPU===
 ^ id  ^ nb cpu ^ cores per cpu ^ cpu model              ^ RAM in GB                   ^ disk in GB ^ GB / core  ^ approximate price HT ^
@@ Line 194: / Line 105: @@
   * - power consumption is 400W for 9754 vs 240W for 7763
   * - 9754 is more difficult to cool as the inlet temperature for air colling must be 22° max
-=== GPU H100 with AMD===
+===== GPU H100 with AMD=====
   * 2 x 64 Core AMD EPYC 9554 3.15GHz Processor
@@ Line 203: / Line 115: @@
   * ~ 28,5k CHF HT per extra GPU
-=== GPU RTX4090 with AMD ===
+===== GPU RTX4090 with AMD =====
   * 2 x 64 Core AMD EPYC 9554 3.10GHz Processor
@@ Line 271: / Line 183: @@
 | [[https://www.amd.com/en/support/downloads/drivers.html/processors/epyc/epyc-7001-series/amd-epyc-7601.html#amd_support_product_spec | EPYC-7601]] | V7 | Naples | 32 | 2.2GHz |
 | [[https://www.amd.com/en/products/processors/server/epyc/7002-series.html | EPYC-7302P]] | V8 | Rome | 16 | 3.0GHz |
-| EPYC-7742 | V8 | Rome | 64 | 2.25GHz |
+| [[https://www.amd.com/en/support/downloads/drivers.html/processors/epyc/epyc-7002-series/amd-epyc-7742.html | EPYC-7742]] | V8 | Rome | 64 | 2.25GHz |
 | [[https://ark.intel.com/content/www/us/en/ark/products/193954/intel-xeon-gold-6234-processor-24-75m-cache-3-30-ghz.html | GOLD-6234]] | V9 | Cascade Lake | 8 | 3.30GHz |
 | [[https://ark.intel.com/content/www/fr/fr/ark/products/192443/intel-xeon-gold-6240-processor-24-75m-cache-2-60-ghz.html | GOLD-6240]] | V9 | Cascade Lake | 18 | 2.60GHz |
@@ Line 277: / Line 189: @@
 | [[https://ark.intel.com/content/www/fr/fr/ark/products/193390/intel-xeon-silver-4208-processor-11m-cache-2-10-ghz.html | SILVER-4208]] | V9 | Cascade Lake | 8 | 2.10GHz |
 | [[https://www.intel.com/content/www/us/en/products/sku/197098/intel-xeon-silver-4210r-processor-13-75m-cache-2-40-ghz/specifications.html | SILVER-4210R]] | V9 | Cascade Lake | 10 | 2.6GHz |
+| [[https://amperecomputing.com/briefs/ampereone-family-product-brief | AMPEREONE-A192-26X]] | V10 | ARM AmperOne | 192 | 2.6GHz |
 | [[https://www.amd.com/en/products/processors/server/epyc/7003-series/amd-epyc-72f3.html | EPYC-72F3]] | V10 | Milan | 8 | 3.7GHz |
 | [[https://www.amd.com/fr/products/processors/server/epyc/7003-series/amd-epyc-7763.html | EPYC-7763]] | V10 | Milan | 64 | 2.45GHz |
@@ Line 282: / Line 195: @@
 | [[https://www.amd.com/en/products/processors/server/epyc/4th-generation-9004-and-8004-series/amd-epyc-9654.html | EPYC-9654]] | V12 | Genoa | 96 | 3.70GHz |
 | [[https://www.amd.com/en/products/processors/server/epyc/4th-generation-9004-and-8004-series/amd-epyc-9754.html | EPYC-9754]] | V13 | Genoa | 128 | 3.70GHz |
 ==== GPUs models available ====
 Several GPU models are available across the three clusters. The table below summarizes the available resources.
@@ Line 292: / Line 203: @@
 | [[https://www.nvidia.com/en-in/data-center/tesla-p100/ | P100]] | 12GB | tesla_p100-pcie-12gb | COMPUTE_TYPE_PASCAL | COMPUTE_CAPABILITY_6_0 | 8.0 → 12.9 | COMPUTE_MODEL_TESLA_P100_PCIE_12GB | 1 |
 | [[https://www.nvidia.com/en-us/geforce/20-series/ | RTX 2080 Ti]] | 11GB | nvidia_geforce_rtx_2080_ti | COMPUTE_TYPE_TURING | COMPUTE_CAPABILITY_7_5 | 10.0 → 13.0 | COMPUTE_MODEL_NVIDIA_GEFORCE_RTX_2080_TI | 2 |
-| [[https://www.nvidia.com/fr-fr/geforce/graphics-cards/30-series/rtx-3080-3080ti/ | RTX 3080]] | 10GB | nvidia_geforce_rtx_3080 | COMPUTE_TYPE_AMPERE | COMPUTE_CAPABILITY_7_0 | 11.0 → 13.0 | COMPUTE_MODEL_NVIDIA_GEFORCE_RTX_3080 | 3 |
+| [[https://www.nvidia.com/fr-fr/geforce/graphics-cards/30-series/rtx-3080-3080ti/ | RTX 3080]] | 10GB | nvidia_geforce_rtx_3080 | COMPUTE_TYPE_AMPERE | COMPUTE_CAPABILITY_8_6 | 11.0 → 13.0 | COMPUTE_MODEL_NVIDIA_GEFORCE_RTX_3080 | 3 |
 | [[https://images.nvidia.com/content/technologies/volta/pdf/volta-v100-datasheet-update-us-1165301-r5.pdf | V100]] | 32GB | tesla_v100-pcie-32gb | COMPUTE_TYPE_VOLTA | COMPUTE_CAPABILITY_7_0 | 9.0 → 12.9 | COMPUTE_MODEL_TESLA_V100_PCIE_32GB | 3 |
 | [[https://www.nvidia.com/en-us/data-center/a100/ | A100 40GB]] | 40GB | nvidia_a100-pcie-40gb | COMPUTE_TYPE_AMPERE | COMPUTE_CAPABILITY_8_0 | 11.0 → 13.0 | COMPUTE_MODEL_NVIDIA_A100_PCIE_40GB | 5 |
@@ Line 303: / Line 214: @@
 | [[https://www.nvidia.com/en-us/products/workstations/rtx-5000/ | RTX 5000]] | 32GB | nvidia_rtx_5000 | COMPUTE_TYPE_ADA | COMPUTE_CAPABILITY_8_9 | 11.8 → 13.0 | COMPUTE_MODEL_NVIDIA_RTX_5000 | 9 |
 | [[https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/ | RTX 5090]] | 32GB | nvidia_geforce_rtx_5090 | COMPUTE_TYPE_BLACKWELL | COMPUTE_CAPABILITY_12_0 | 12.8 → 13.0 | COMPUTE_MODEL_NVIDIA_GEFORCE_RTX_5090 | 10 |
+| [[https://www.nvidia.com/en-us/data-center/rtx-pro-4500-blackwell-server-edition/ | RTX PRO 4500]] | 32GB | nvidia_rtx_pro_4500_blackwell | COMPUTE_TYPE_BLACKWELL | COMPUTE_CAPABILITY_12_0 | 12.8 → 13.0 | COMPUTE_MODEL_NVIDIA_RTX_PRO_4500_BLACKWELL | 10 |
 | [[https://www.nvidia.com/en-us/data-center/h100/ | H100]] | 94GB | nvidia_h100_nvl | COMPUTE_TYPE_HOPPER | COMPUTE_CAPABILITY_9_0 | 11.8 → 13.0 | COMPUTE_MODEL_NVIDIA_H100_NVL | 14 |
 | [[https://www.nvidia.com/en-us/data-center/rtx-pro-6000-blackwell-server-edition/ | RTX Pro 6000]] | 96GB | nvidia_rtx_pro_6000_blackwell | COMPUTE_TYPE_BLACKWELL | COMPUTE_CAPABILITY_9_0 | 12.8 → 13.0 | COMPUTE_MODEL_NVIDIA_RTX_PRO_6000_BLACKWELL | 16 |
@@ Line 309: / Line 221: @@
+<note tip>We don't have mixed GPUs models on the same node (except gpu027.baobab). Every GPU node has only one GPU model.</note>
-<note tip>We don't have mixed GPUs models on the same node. Every GPU node has only one GPU model.</note>
 See [[hpc::slurm#gpgpu_jobs|here]] how to request GPU for your jobs.