计算资源

命名规则和优先级

每组计算节点均对应两个队列；标准队列和优先队列

• 标准队列（CPU和加速卡型号+网络类型）：所有用户都可以提交作业到这个队列进行计算；
• 优先队列（CPU和加速卡型号+网络类型+!）：
  • 公共设备的优先队列：仅用于测试用途，优先级高于标准队列，但运行时间、作业数量、核数都有限制，没有抢占。
  • 自购设备的优先队列：只允许队列购置者提交作业，优先!队列的作业如果因为资源不足而不能运行，则会抢占标准队列作业的资源，将标准队列的作业挂起（状态SSUSP）或杀掉，腾出资源来运行优先!队列的作业，当作业完成后被挂起的作业会自动恢复执行。被挂起的作业可以用 brequeue <JOBID> 命令杀掉并自动重新提交。
• 每组计算节点的所有队列共享同一个 fairshare 数据。如果一个队列中的计算机由不止一个组出资购买，那么每个组在优先!队列运行中作业所能使用的节点限制为出资购买的部分，其它队列则无此限制。
• 组管理员可以来信设定本组所有用户可提交的队列，默认为所有队列均可提交作业。
• 所有公共共享队列共享同一个 fairshare 数据，公共共享队列的作业不会被抢占。

计算队列

合计

• 节点数量：341
• CPU插槽数：713
• CPU核数：15768
• GPU卡数：191
• GPU CUDA核数：2375808
• 内存：83232 GB
• 显存：4680 GB
• CPU和数据中心GPU双精度浮点计算理论峰值：1049.0776 TFLOPS (CPU: 833.818, GPU: 215.26)
• 消费级GPU单精度浮点计算理论峰值：10170.912 TFLOPS

总览

• 双路CPU：e52692v2ib、e52680v3ib、e5v3ib、e52630v4、e52640v4、e52650v4、e52680v4ib、e52682v4opa、5218、7702ib、5220r、6230r、5318h、6330ib、7k83、9654、9754
• 四路CPU：e74820v4、6140ib、5218q、9242opa
• 高主频CPU：e52643tgb、6226rib
• 大内存CPU：e7v4ib、6338、6458q、97554090ib
• 数据中心GPU：e5v3k40ib、e5v4p100ib、6148v100ib、7552v100、62v100ib、83a100ib
• 消费级GPU：722080tiib、72rtxib、813080、723080、723090ib、814080、734090d、944090d、734090ib、75434090ib、97554090ib、945090ib、65305090ib
• CPU融合：cpu1

表格

查看详细表格

SSD：

• / 表示节点本地根文件系统采用SSD，用户可用 /tmp 和 /var/tmp
• /ssd 表示节点有额外的SSD用于临时存储，用户可用 /ssd 和 /tmp/ssd

详情

e52643tgb

• 双精度浮点计算理论峰值：2.112 TFLOPS
• 购置者：万贤纲100%
• 10*Dell PowerEdge 12G R620
• Hostname: r001-r010
• CPU: 2*Intel Xeon E5-2643 (4 Cores, 10MB Cache, 3.30 GHz, 8.00 GT/s QPI)
• RAM: 4 Channels, DDR3-1600 ECC RDIMM, Bandwidth: 102.4 GB/s
  • [r001-r003] 96 GB
  • [r004-r010] 128 GB，-R largemem 指定使用
• HDD: 600 GB 2.5in 10Krpm 6Gb/s SAS
• Network: 10 Gigabit Ethernet

e52692v2ib

• 双精度浮点计算理论峰值：16.896 TFLOPS
• 购置者：孙建75%，王强华25%
• 40*Inspur NX5440
• Hostname: i01n01-i01n20 i02n01-i02n20
• CPU: 2*Intel Xeon E5-2692 v2 (12 Cores, 30MB Cache, 2.20 GHz, 8 GT/s QPI)
• RAM: 64 GB (8x8GB), 4 Channels, DDR3-1866 ECC RDIMM, Bandwidth: 119.4 GB/s
• HDD: 600 GB 2.5in 10Krpm 6Gb/s SAS
• Network: Gigabit Ethernet, 40 Gbit/s QDR InfiniBand

e52680v3ib

• 双精度浮点计算理论峰值：24.96 TFLOPS
• 购置者：张海军100%
• 26*HP BladeSystem ProLiant BL460c Gen9
• Hostname: s03n01-s03n13 s04n01-s04n13
• CPU: 2*Intel Xeon E5-2680 v3 (12 Cores, 30MB Cache, 2.50 GHz, 9.6 GT/s QPI)
• RAM: 128 GB ([s03n01-s03n13 s04n01-s04n02] 8x16GB | [s04n03-s04n13] 16x8GB), 4 Channels, DDR4-2133 ECC RDIMM, Bandwidth: 136 GB/s
• HDD: 300 GB 2.5in 10Krpm 6Gb/s SAS
• Network: Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

e5v3ib

• 双精度浮点计算理论峰值：72 TFLOPS
• 购置者：公共共享100%
• 70*Lenovo Flex System x240 M5 5*Lenovo System x3650 M5
• Hostname: f01n01-f01n14 f02n01-f02n14 f03n01-f03n14 f04n01-f04n14 f05n01-f05n14 x013-x017 x013-x017
• CPU: 2*Intel Xeon E5-2680 v3 (12 Cores, 30MB Cache, 2.50 GHz, 9.6 GT/s QPI)
• RAM: 4 Channels, [f01n01-f05n14] DDR4-2133 ECC RDIMM, Bandwidth: 136 GB/s | [x013-x017] DDR4-1866 ECC RDIMM, Bandwidth: 119.4 GB/s
  • [f01n01-f05n04] 128 GB (8x16GB)
  • [f05n05-f05n14] 256 GB (8x32GB), -R mem256g 或 -R largemem 指定使用
  • [x013-x017] 384 GB (24x32GB), -R mem384g 或 -R largemem 指定使用
• SSD: 2.5in SATA 6Gb/s
  • [f01n01-f05n14] Samsung Enterprise SSD PM863, 32-layer 3D TLC, 120GB 170TBW
  • [x013-x017] Samsung SSD 850 PRO, 32-layer 3D TLC, 512GB
• Network: Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

e5v3k40ib

• 双精度浮点计算理论峰值：3.82 TFLOPS
• 购置者：公共共享100%
• 1*Lenovo System x3650 M5
• Hostname: x001
• CPU: 2*Intel Xeon E5-2680 v3 (12 Cores, 30MB Cache, 2.50 GHz, 9.6 GT/s QPI)
• RAM: 128 GB (8x16GB), 4 Channels, DDR4-2133 ECC RDIMM, Bandwidth: 136 GB/s
• GPU: 2*NVIDIA Tesla K40 (2880 Cores 875MHz, 12GB 384-bit GDDR5 3.0GHz 288GB/s, PCIe3.0 x16)
• HDD: 300 GB 2.5in 10Krpm 6Gb/s SAS
• Network: 10 Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

e52630v4

• 双精度浮点计算理论峰值：1.408 TFLOPS
• 购置者：邵陆兵100%（邢定钰组）
• 2*Inspur NX5460M4
• Hostname: i04n01-i04n02
• CPU: 2*Intel Xeon E5-2630 v4 (10 Cores, 25MB Cache, 2.20 GHz, 8 GT/s QPI)
• RAM: 64 GB (4x16GB), 2 Channels, DDR4-2133 ECC RDIMM, Bandwidth: 68.3 GB/s
• HDD: 2*300 GB 2.5in 10Krpm 12Gb/s SAS RAID1
• Network: 10 Gigabit Ethernet

e52640v4

• 双精度浮点计算理论峰值：0.768 TFLOPS
• 购置者：李建新100%
• 1*Inspur NF5280M4
• Hostname: n001
• CPU: 2*Intel Xeon E5-2640 v4 (10 Cores, 25MB Cache, 2.40 GHz, 8 GT/s QPI)
• RAM: 128 GB (8x16GB), 4 Channels, DDR4-2133 ECC RDIMM, Bandwidth: 136.6 GB/s
• HDD: 2*600 GB 2.5in 10Krpm 12Gb/s SAS RAID1
• Network: Gigabit Ethernet

e52650v4

• 双精度浮点计算理论峰值：2.5344 TFLOPS
• 购置者：盛利100%
• 3*Inspur NF5270M4
• Hostname: n003-n005
• CPU: 2*Intel Xeon E5-2650 v4 (12 Cores, 30MB Cache, 2.20 GHz, 9.6 GT/s QPI)
• RAM: 128 GB (8x16GB), 4 Channels, DDR4-2400 ECC RDIMM, Bandwidth: 153.6 GB/s
• HDD: [n003-n004] 4*300 GB 2.5in 10Krpm 12Gb/s SAS RAID5 | [n005] 3*300 GB 2.5in 10Krpm 12Gb/s SAS RAID5
• Network: Gigabit Ethernet

e52680v4ib

• 双精度浮点计算理论峰值：20.4288 TFLOPS
• 购置者：王炜84.21%，法伟15.79%
• 19*HPE BladeSystem ProLiant BL460c Gen9
• Hostname: s04n14-s04n16 s05n01-s05n16
• CPU: 2*Intel Xeon E5-2680 v4 (14 Cores, 35MB Cache, 2.40 GHz, 9.6 GT/s QPI)
• RAM: 128 GB (8x16GB), 4 Channels, DDR4-2400 ECC RDIMM, Bandwidth: 153.6 GB/s
• SSD: Samsung Enterprise SSD PM863, 32-layer 3D TLC, 120GB 170TBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

e52682v4opa

• 双精度浮点计算理论峰值：25.6 TFLOPS
• 购置者：万贤纲100%
• 20*Inspur NX5440M4
• Hostname: i03n01-i03n20
• CPU: 2*Intel Xeon E5-2682 v4 (16 Cores, 40MB Cache, 2.50 GHz, 9.6 GT/s QPI)
• RAM: 64 GB (8x8GB), 4 Channels, DDR4-2400 ECC RDIMM, Bandwidth: 153.6 GB/s
• SSD: Intel SSD DC S3500, 20nm MLC, 240GB 140TBW
• SSD: 2.5in SATA 6Gb/s, [i03n01-i03n10] Samsung Enterprise SSD PM863, 32-layer 3D TLC, 960GB 1400TBW | [i03n11-i03n20] Intel SSD DC S3610, 20nm MLC, 800GB 5.3PBW
  • SSD mounted at /ssd & /tmp/ssd
• Network: Gigabit Ethernet, 100 Gbit/s Omni-Path Architecture

e7v4ib

• 双精度浮点计算理论峰值：2.1504 TFLOPS
• 购置者：公共共享100%
• 1*Dell PowerEdge R930
• Hostname: r011
• CPU: 4*Intel Xeon E7-4850 v4 (16 Cores, 40MB Cache, 2.10 GHz, 8 GT/s QPI)
• RAM: 1024 GB (32x32GB), 4 Channels, DDR4-1333 ECC RDIMM
• SSD: 2*Intel SSD DC S3610, 20nm MLC, 200GB 1.1PBW, 2.5in SATA 6Gb/s RAID1
• Network: 10 Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

e5v4p100ib

• 双精度浮点计算理论峰值：10.296 TFLOPS
• 购置者：公共共享100%
• 1*Lenovo System x3650 M5
• Hostname: x002
• CPU: 2*Intel Xeon E5-2660 v4 (14 Cores, 35MB Cache, 2.00 GHz, 9.6 GT/s QPI)
• RAM: 128 GB (8x16GB), 4 Channels, DDR4-2400 ECC RDIMM, Bandwidth: 153.6 GB/s
• GPU: 2*NVIDIA Tesla P100 PCIe 16GB (3584 Cores 1328MHz, 16GB 715MHz 4096-bit HBM2 732 GB/s, PCIe3.0 x16)
• HDD: 300 GB 2.5in 10Krpm 6Gb/s SAS
• Network: Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

6140ib

• 双精度浮点计算理论峰值：84.7872 TFLOPS
• 购置者：公共共享100%
• 16*H3C B7800 G3
• Hostname: b01n01-b01n08 b02n01-b02n08
• CPU: 4*Intel Xeon Gold 6140 (18 Cores, 24.75MB Cache, 2.30 GHz)
• RAM: 384 GB (24x16GB), 6 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 511.2 GB/s
• SSD: Micron 5100, TLC, 240GB, M.2 2280, SATA 6Gb/s
• Network: 10 Gigabit Ethernet, 100 Gbit/s EDR InfiniBand

6148v100ib

• 双精度浮点计算理论峰值：10.072 TFLOPS
• 购置者：赵宇心100%
• 1*Lenovo ThinkSystem SR650
• Hostname: x003
• CPU: 2*Intel Xeon Gold 6148 (20 Cores, 27.5MB Cache, 2.40 GHz)
• RAM: 192 GB (12x16GB), 6 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 255.6 GB/s
• GPU: NVIDIA Tesla V100 PCIe 16GB (5120 CUDA Cores 1380MHz, 16GB HBM2 877MHz 4096-bit 900GB/s, PCIe3.0 x16)
• SSD: LITEON, 128GB, M.2 2280, SATA 6Gb/s
• Network: Gigabit Ethernet, 56 Gbit/s FDR InfiniBand

5218

• 双精度浮点计算理论峰值：1.1776 TFLOPS
• 提供者：CPU/准系统 Intel、内存 本中心
• 1*Intel S2600WFT
• Hostname: o003
• CPU: 2*Intel Xeon Gold 5218 (16 Cores, 22MB Cache, 2.30 GHz)
• RAM: 192 GB (12x16GB), 6 Channels, DDR4-2666 ECC RDIMM
• SSD: Intel SSD DC S3500, 20nm MLC, 275TBW 480GB, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet

5218q

• 双精度浮点计算理论峰值：2.3552 TFLOPS
• 购置者：李建新100%
• 1*Inspur NF8480M5
• Hostname: n032
• CPU: 4*Intel Xeon Gold 5218 (16 Cores, 22MB Cache, 2.30 GHz)
• RAM: 256 GB (8x32GB), 4 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 170.7 GB/s
• HDD: 2*600 GB 2.5in 10Krpm 12Gb/s SAS RAID1
• Network: Gigabit Ethernet

7552v100

• 双精度浮点计算理论峰值：59.3792 TFLOPS
• 购置者：卢毅100%
• 1*Tongtaiyi TG659V2
• Hostname: o004
• CPU: 2*AMD EPYC 7552 (48 Cores, 192MB Cache, 2.2 GHz)
• RAM: 512 GB (16x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• GPU: 8*NVIDIA Tesla V100 PCIe 32GB (5120 CUDA Cores 1380MHz, 32GB HBM2 877MHz 4096-bit 900GB/s, PCIe3.0 x16)
• SSD: Samsung PM883, TLC 3D NAND, 960GB 1366TBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet

62v100ib

• 双精度浮点计算理论峰值：65.6 TFLOPS
• 购置者：公共共享100%
• 1*Inspur NF5468M5-S
• Hostname: n002
• CPU: 2*Intel Xeon Gold 6248 (20 Cores, 27.5MB Cache, 2.50 GHz)
• RAM: 768 GB (24x32GB), 6 Channels, DDR4-2933 ECC RDIMM, Bandwidth: 281.6 GB/s
• GPU: 8*NVIDIA Tesla V100 SXM2 32GB (5120 CUDA Cores 1530MHz, 32GB HBM2 877MHz 4096-bit 900GB/s, NVLink 300GB/s, PCIe3.0 x16)
• SSD: 2*Samsung Enterprise SSD PM883, TLC 3D NAND, 480GB 683TBW, 2.5in SATA 6Gb/s, RAID1
• SSD: 2*Intel SSD DC P4510, 64-Layer TLC 3D NAND, 2TB 2.61PBW, U.2 15mm, PCIe3.1 x4 NVMe, RAID0
  • SSD mounted at /ssd & /tmp/ssd
• Network: 10 Gigabit Ethernet, 100 Gbit/s EDR InfiniBand

9242opa

• 双精度浮点计算理论峰值：141.312 TFLOPS
• 购置者：孙建100%
• 20*Intel S9248WK2HAC
• Hostname: s001-s020
• CPU: 2*Intel Xeon Platinum 9242 (48 Cores, 71.5MB Cache, 2.30 GHz)
• RAM: 384 GB (24x16GB), 12 Channels, DDR4-2933 ECC RDIMM, Bandwidth: 563.2 GB/s
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 480GB 1.2PBW, M.2 2280, SATA 6Gb/s
• Network: Gigabit Ethernet, 100 Gbit/s Omni-Path Architecture

7702ib

• 双精度浮点计算理论峰值：57.344 TFLOPS
• 购置者：公共共享100%
• 14*Asus RS720A-E9-RS24V2
• Hostname: a001-a014
• CPU: 2*AMD EPYC 7702 (64 Cores, 256MB Cache, 2.0 GHz)
• RAM: 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
  • [a001-a010] 256 GB (16x16GB)
  • [a011-a014] 512 GB (32x16GB), -R largemem 指定使用
• SSD: Samsung SSD 970 EVO Plus, V-NAND 3-bit MLC, 250GB 150TBW, M.2 2280, PCIe3.0 x4 NVMe 1.3
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

722080tiib

• CPU双精度浮点计算理论峰值：3.072 TFLOPS
• GPU单精度浮点计算理论峰值：215.168 TFLOPS
• 购置者：公共共享100%
• 4*Gigabyte G242-Z10
• Hostname: g001-g004
• CPU: 1*AMD EPYC 7302 (16 Cores, 128MB Cache, 3.0 GHz)
• GPU: 4*NVIDIA GeForce RTX 2080 Ti 11GB (4352 CUDA Cores 1545MHz, 11GB GDDR6 7000MHz 352-bit 616 GB/s, PCIe3.0 x16)
• RAM: 256 GB (8x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 204.8 GB/s
• SSD: Intel SSD DC S4600, 3D NAND TLC, 240GB 1.40PBW, 2.5in SATA 6Gb/s
• SSD: Intel SSD DC P4510, 64-Layer TLC 3D NAND, 2TB 2.61PBW, U.2 15mm, PCIe3.1 x4 NVMe
  • SSD mounted at /ssd & /tmp/ssd
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

72rtxib

• CPU双精度浮点计算理论峰值：2.304 TFLOPS
• GPU单精度浮点计算理论峰值：195.744 TFLOPS
• 购置者：公共共享100%
• 3*Gigabyte G242-Z10
• Hostname: g005-g007
• CPU: 1*AMD EPYC 7302 (16 Cores, 128MB Cache, 3.0 GHz)
• GPU: 4*NVIDIA TITAN RTX 24GB (4608 CUDA Cores 1770MHz, 24GB GDDR6 7000MHz 384-bit 672 GB/s, PCIe3.0 x16)
• RAM: 256 GB (8x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 204.8 GB/s
• SSD: Intel SSD DC S4600, 3D NAND TLC, 240GB 1.40PBW, 2.5in SATA 6Gb/s
• SSD: Intel SSD DC P4510, 64-Layer TLC 3D NAND, 2TB 2.61PBW, U.2 15mm, PCIe3.1 x4 NVMe
  • SSD mounted at /ssd & /tmp/ssd
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

5220r

• 双精度浮点计算理论峰值：5.0688 TFLOPS
• 购置者：盛利100%
• 3*Inspur NF5280M5
• Hostname: n006-n008
• CPU: 2*Intel Xeon Gold 5220R (24 Cores, 35.75MB Cache, 2.20 GHz)
• RAM: 192 GB (6x32GB), 3 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 127.8 GB/s
• SSD: 2*Intel SSD DC S4500, 3D NAND TLC, 480GB 0.90PBW, 2.5in SATA 6Gb/s, RAID1
• Network: Gigabit Ethernet

6226rib

• 双精度浮点计算理论峰值：65.3312 TFLOPS
• 购置者：万贤纲100%
• 22*Huawei FusionServer Pro 1288H V5
• Hostname: f001-f022
• CPU: 2*Intel Xeon Gold 6226R (16 Cores, 22MB Cache, 2.90 GHz)
• RAM: 192 GB (12x16GB), 6 Channels, DDR4-2933 ECC RDIMM, Bandwidth: 281.6 GB/s
• SSD: Huawei ES3610C V5, 3D TLC, 1.6TB 8.76PBW, HHHL, PCIe3.0 x4 NVMe 1.3
• Network: Gigabit Ethernet, 100 Gbit/s EDR InfiniBand

6230r

• 双精度浮点计算理论峰值：10.4832 TFLOPS
• 购置者：盛利100%
• 3*Inspur NF5280M5
• Hostname: n009-n011
• CPU: 2*Intel Xeon Gold 6230R (26 Cores, 35.75MB Cache, 2.10 GHz)
• RAM: 192 GB (6x32GB), 3 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 127.8 GB/s
• SSD: 2*Samsung PM883, TLC 3D NAND, 480GB 683TBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet

5318h

• 双精度浮点计算理论峰值：5.76 TFLOPS
• 购置者：李建新100%
• 1*Inspur NF8480M6
• Hostname: n035
• CPU: 4*Intel Xeon Gold 5318H (18 Cores, 24.75MB Cache, 2.50 GHz)
• RAM: 512 GB (16x32GB), 4 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 170.7 GB/s
• HDD: 2*Samsung PM893, V-NAND TLC, 960GB 1752TBW, 2.5in SATA 6Gb/s, RAID1
• Network: Gigabit Ethernet

6330ib

• 双精度浮点计算理论峰值：57.344 TFLOPS
• 购置者：公共共享100%
• 16*Inspur NF5180M6
• Hostname: n016-n031
• CPU: 2*Intel Xeon Gold 6330 (28 Cores, 42MB Cache, 2.00 GHz)
• RAM: 256 GB (16x16GB), 8 Channels, DDR4-2933 ECC RDIMM, Bandwidth: 375.5 GB/s
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 240GB 0.9PBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

6338

• 双精度浮点计算理论峰值：16.384 TFLOPS
• 购置者：卢毅100%
• 4*Inspur NF5180M6
• Hostname: n012-n015
• CPU: 2*Intel Xeon Gold 6338 (32 Cores, 48MB Cache, 2.00 GHz)
• RAM: 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
  • [n012 n013] 1024 GB (32x32GB)
  • [n014 n015] 2048 GB (32x64GB), -R largemem 指定使用
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 240GB 0.9PBW, 2.5in SATA 6Gb/s
• Network: 10 Gigabit Ethernet

7k83

• 双精度浮点计算理论峰值： 15.6672 TFLOPS
• 购置者：雷群利100%
• 2*Gigabyte R182-Z90 1*Asus RS700A-E11-RS4U
• Hostname: g008-g009 a016
• CPU: 2*AMD EPYC 7K83 (64 Cores, 256MB Cache, 2.55 GHz)
• RAM: 256 GB (16x16GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 480GB 1.2PBW, 2.5in SATA 6Gb/s
• Network: 10 Gigabit Ethernet

83a100ib

• 双精度浮点计算理论峰值：82.9248 TFLOPS
• 购置者：公共共享100%
• 1*Supermicro SYS-420GP-TNAR+ (Powerleader PR4908WV)
• Hostname: m001
• CPU: 2*Intel Xeon Platinum 8358 (32 Cores, 48MB Cache, 2.60 GHz)
• RAM: 512 GB (16x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• GPU: 8*NVIDIA A100 SXM4 40GB (6912 CUDA Cores 1410MHz, 40GB HBM2 1215MHz 5120-bit 1555GB/s, NVLink 600GB/s, PCIe4.0 x16)
• SSD: Samsung 983 DCT, V-NAND 3-bit MLC, 960GB 1.366PBW, M.2 22110, PCIe3.0 x4 NVMe 1.2b
• SSD: 2*Intel SSD D7-P5520, 144-Layer TLC 3D NAND, 3.84TB 7.0PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.4, RAID0
  • SSD mounted at /ssd & /tmp/ssd
• Network: 10 Gigabit Ethernet, 200 Gbit/s HDR InfiniBand

813080

• CPU双精度浮点计算理论峰值：7.9872 TFLOPS
• GPU单精度浮点计算理论峰值：476.32 TFLOPS
• 购置者：雷群利100%
• 1*Supermicro SYS-4029GP-TRT 1*Asus ESC8000 G4
• Hostname: m004 a018
• CPU: [m004] 2*Xeon Platinum 8163 (24 Cores, 33MB Cache, 2.50 GHz) | [a018] 2*Xeon Platinum 8168 (24 Cores, 33MB Cache, 2.70 GHz)
• RAM: [m004] 64 GB (4x16GB), 2 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 85.3 GB/s | [a018] 128 GB (8x16GB), 4 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 170.7 GB/s
• GPU: 8*NVIDIA GeForce RTX 3080 10GB (8704 CUDA Cores 2100MHz, 10GB GDDR6X 9501MHz 320-bit 760.3 GB/s, PCIe3.0 x16)
• SSD: [m004] Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 480GB 1.2PBW, 2.5in SATA 6Gb/s | [a018] Intel SSD D3-S4610, 64-Layer TLC 3D NAND, 240GB 1.4PBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet

723080

• CPU双精度浮点计算理论峰值：4.608 TFLOPS
• GPU单精度浮点计算理论峰值：238.16 TFLOPS
• 购置者：雷群利100%
• 1*Asus ESC8000A-E11
• Hostname: a015
• CPU: 2*AMD EPYC 7B12 (64 Cores, 256MB Cache, 2.25 GHz)
• RAM: 128 GB (16x8GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• GPU: 8*NVIDIA GeForce RTX 3080 10GB (8704 CUDA Cores 2100MHz, 10GB GDDR6X 9501MHz 320-bit 760.3 GB/s, PCIe4.0 x16)
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 240GB 0.9PBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet

723090ib

• CPU双精度浮点计算理论峰值：4.3008 TFLOPS
• GPU单精度浮点计算理论峰值：569.28 TFLOPS
• 购置者：公共共享100%
• 2*Supermicro AS -4124GS-TNR (Roycom)
• Hostname: m002-m003
• CPU: 2*AMD EPYC 7402 (24 Cores, 128MB Cache, 2.8 GHz)
• RAM: 512 GB (16x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• GPU: 8*NVIDIA GeForce RTX 3090 24GB (10496 CUDA Cores 2100MHz, 24GB GDDR6X 9751MHz 384-bit 936.2 GB/s, PCIe4.0 x16)
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 480GB 1.2PBW, 2.5in SATA 6Gb/s
• SSD: 2*Intel SSD D7-P5520, 144-Layer TLC 3D NAND, 3.84TB 7.0PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.4, RAID0
  • SSD mounted at /ssd & /tmp/ssd
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

814080

• CPU双精度浮点计算理论峰值：4.1472 TFLOPS
• GPU单精度浮点计算理论峰值：194.960 TFLOPS
• 购置者：雷群利100%
• 1*Supermicro X11DPG-QT
• Hostname: m009
• CPU: 2*Xeon Platinum 8168 (24 Cores, 33MB Cache, 2.70 GHz)
• RAM: 64 GB (4x16GB), 2 Channels, DDR4-2666 ECC RDIMM, Bandwidth: 85.3 GB/s
• GPU: 4*NVIDIA GeForce RTX 4080 16GB (9728 CUDA Cores 3105MHz, 16GB GDDR6X 11201MHz 256-bit 716.8 GB/s, PCIe4.0 x16)
• SSD: Samsung PM883, TLC 3D NAND, 240GB 341TBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet

734090d

• CPU双精度浮点计算理论峰值：5.7344 TFLOPS
• GPU单精度浮点计算理论峰值：588.32 TFLOPS
• 购置者：王炜100%
• 1*Inspur NF5468A5
• Hostname: n034
• CPU: 2*AMD EPYC 7543 (32 Cores, 256MB Cache, 2.8 GHz)
• RAM: 768 GB (24x32GB) DDR4-2933 ECC RDIMM
• GPU: 8*NVIDIA GeForce RTX 4090 D 24GB (14592 CUDA Cores 3105MHz, 24GB GDDR6X 10501MHz 384-bit 1008 GB/s, PCIe4.0 x16)
• SSD: Intel SSD D5-P5530, 128-Layer TLC NAND, 960GB 1.7PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.3c
• SSD: 2*SAMSUNG PM983a, 7.68TB 10.9PBW, U.2 7mm, PCIe3.0 x4, NVMe 1.3c
  • SSD mounted at /ssd1 & /ssd2
• Network: Gigabit Ethernet

944090d

• CPU双精度浮点计算理论峰值：7.3728 TFLOPS
• GPU单精度浮点计算理论峰值：441.24 TFLOPS
• 购置者：王炜100%
• 1*Inspur NF5468A7
• Hostname: n033
• CPU: 2*AMD EPYC 9654 (96 Cores, 384MB Cache, 2.4 GHz)
• RAM: 768 GB (24x32GB) 12 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 921.6 GB/s
• GPU: 6*NVIDIA GeForce RTX 4090 D 24GB (14592 CUDA Cores 3105MHz, 24GB GDDR6X 10501MHz 384-bit 1008 GB/s, PCIe4.0 x16)
• SSD: SSD 960GB
• SSD: 2*SAMSUNG PM983a, 7.68TB 10.9PBW, U.2 7mm, PCIe3.0 x4, NVMe 1.3c
  • SSD mounted at /ssd1 & /ssd2
• Network: Gigabit Ethernet

734090ib

• CPU双精度浮点计算理论峰值：5.7344 TFLOPS
• GPU单精度浮点计算理论峰值：1321.28 TFLOPS
• 购置者：公共共享100%
• 2*Supermicro AS -4124GS-TNR (Wuzhou S748E4)
• Hostname: m005-m006
• CPU: 2*AMD EPYC 7543 (32 Cores, 256MB Cache, 2.8 GHz)
• RAM: 512 GB (16x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• GPU: 8*NVIDIA GeForce RTX 4090 24GB (16384 CUDA Cores 3105MHz, 24GB GDDR6X 10501MHz 384-bit 1008 GB/s, PCIe4.0 x16)
• SSD: Samsung PM883, TLC 3D NAND, 480GB 683TBW, 2.5in SATA 6Gb/s
• SSD: 2*Intel SSD D7-P5520, 144-Layer TLC 3D NAND, 3.84TB 7.0PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.4, RAID0
  • SSD mounted at /ssd & /tmp/ssd
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

75434090ib

• CPU双精度浮点计算理论峰值：5.7344 TFLOPS
• GPU单精度浮点计算理论峰值：1321.28 TFLOPS
• 购置者：孙建100%
• 2*Supermicro AS -4124GS-TNR (Wuzhou S748E4 | Roycom)
• Hostname: m007-m008
• CPU: 2*AMD EPYC 7543 (32 Cores, 256MB Cache, 2.8 GHz)
• RAM: 512 GB (16x32GB), 8 Channels, DDR4-3200 ECC RDIMM, Bandwidth: 409.6 GB/s
• GPU: 8*NVIDIA GeForce RTX 4090 24GB (16384 CUDA Cores 3105MHz, 24GB GDDR6X 10501MHz 384-bit 1008 GB/s, PCIe4.0 x16)
• SSD: Samsung PM883, TLC 3D NAND, 480GB 683TBW, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

9654

• 双精度浮点计算理论峰值：7.3728 TFLOPS
• 购置者：雷群利100%
• 1*Asus RS720A-E12-RS12
• Hostname: a017
• CPU: 2*AMD EPYC 9654 (96 Cores, 384MB Cache, 2.4 GHz)
• RAM: 384 GB (24x16GB), 12 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 921.6 GB/s
• SSD: Intel SSD D3-S4510, 64-Layer TLC 3D NAND, 480GB 1.2PBW, 2.5in SATA 6Gb/s
• Network: 10 Gigabit Ethernet

9754

• 双精度浮点计算理论峰值：9.216 TFLOPS
• 购置者：雷群利100%
• 1*Inventec K885G6
• Hostname: v001
• CPU: 2*AMD EPYC 9754 (128 Cores, 256MB Cache, 2.25 GHz)
• RAM: 384 GB (24x16GB), 12 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 921.6 GB/s
• SSD: Intel SSD D5-P5530, 128-Layer TLC NAND, 960GB 1.7PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.3c
• Network: 1 Gigabit Ethernet

6458q

• 双精度浮点计算理论峰值：25.3952 TFLOPS
• 购置者：卢毅100%
• 4*Nettrix R620 G50
• Hostname: u001-u004
• CPU: 2*Intel Xeon Gold 6458Q (32 Cores, 60MB Cache, 3.10 GHz)
• RAM: 1024 GB (16x64GB), 8 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 614.4 GB/s
• SSD: Samsung PM883, TLC 3D NAND, 480GB 683TBW, 2.5in SATA 6Gb/s, RAID1
• Network: 10 Gigabit Ethernet

97554090ib

• CPU双精度浮点计算理论峰值：66.3552 TFLOPS
• GPU单精度浮点计算理论峰值：165.16 TFLOPS
• 购置者：王锐100%
• 3*Tongtaiyi TU629V3 (Sitonholy AW2245-2GR)
• Hostname: t002-t004
• CPU: 2*AMD EPYC 9755 (128 Cores, 512MB Cache, 2.7 GHz)
• RAM: 12 Channels, DDR5-6400 ECC RDIMM, Bandwidth: 1228 GB/s
  • [t002] 2304 GB (24x96GB), -R largemem 指定使用
  • [t003-t004] 1536 GB (24x64GB)
• GPU: [t003-t004] 1*NVIDIA GeForce RTX 4090 24GB (16384 CUDA Cores 3105MHz, 24GB GDDR6X 10501MHz 384-bit 1008 GB/s, PCIe4.0 x16)
• SSD: GSAITECH, 960GB, 2.5in SATA 6Gb/s
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

945090ib

• CPU双精度浮点计算理论峰值： 12.672 TFLOPS
• GPU单精度浮点计算理论峰值： 2515.2 TFLOPS
• 购置者：公共共享100%
• 3*Enginetech EG8621A4
• Hostname: e001-e003
• CPU: 2*AMD EPYC 9454 (48 Cores, 256MB Cache, 2.75 GHz)
• RAM: 768 GB (24x32GB), 12 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 921.6 GB/s
• GPU: 8*NVIDIA GeForce RTX 5090 32GB (21760 CUDA Cores 3090MHz, 32GB GDDR7 14001MHz 512-bit 1792 GB/s, PCIe5.0 x16)
• SSD: Intel SSD D7-P5520, 144-Layer TLC 3D NAND, 7.68TB 14.0PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.4
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

65305090ib

• CPU双精度浮点计算理论峰值： 8.6016 TFLOPS
• GPU单精度浮点计算理论峰值： 1676.8 TFLOPS
• 购置者：孙建100%
• 2*Enginetech EG8621G4
• Hostname: e004-e005
• CPU: 2*Intel Xeon Gold 6530 (32 Cores, 160MB Cache, 2.1 GHz)
• RAM: 512 GB (16x32GB), 8 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 614.4 GB/s
• GPU: 8*NVIDIA GeForce RTX 5090 32GB (21760 CUDA Cores 3090MHz, 32GB GDDR7 14001MHz 512-bit 1792 GB/s, PCIe5.0 x16)
• SSD: Intel SSD D7-P5520, 144-Layer TLC 3D NAND, 7.68TB 14.0PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.4
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

9654p6000ib

• CPU双精度浮点计算理论峰值： 7.3728 TFLOPS
• GPU单精度浮点计算理论峰值： 252 TFLOPS
• 购置者：王炜100%
• 1*Enginetech EG8621A4
• Hostname: e006
• CPU: 2*AMD EPYC 9654 (96 Cores, 384MB Cache, 2.4 GHz)
• RAM: 256 GB (4x64GB), 2 Channels, DDR5-4800 ECC RDIMM, Bandwidth: 153.6 GB/s
• GPU: 2*NVIDIA GeForce RTX PRO 6000 96GB (24064 CUDA Cores 2430MHz, 96GB GDDR7-ECC 12481MHz 512-bit 1597 GB/s, PCIe5.0 x16)
• SSD: Intel SSD D7-P5520, 144-Layer TLC 3D NAND, 7.68TB 14.0PBW, U.2 15mm, PCIe4.0 x4, NVMe 1.4
• Network: Gigabit Ethernet, 100 Gbit/s HDR100 InfiniBand

cpu1

• 作业≤24 cores 单节点运行于公共共享的CPU队列 e5v3ib 6140ib 7702ib 6330ib 中的任意节点，但优先级小于这些队列