SPECIFICATIONS

Brand Name: lotorasia

Chipset Manufacturer: NVIDIA

Choice: yes

Cooler Type: No Fan

GPU Model: Tesla K80

High-concerned chemical: None

Interface Type: PCI Express 3.0x16

Memory Interface: 384 BIT

Model Number: K80

Origin: Mainland China

Output Interface Type: VGA (D-Sub)

Overclocked: No

Power Connector: 8pin

RGB: No

Video Memory Capacity: 24GB

Video Memory Type: GDDR5

semi_Choice: yes

NVIDIA Tesla K80 is a dual-GPU computing card based on the Kepler architecture, whose main parameters include: Dual GK210 cores, 4992 CUDA cores, 24GB GDDR5 memory (12GB per GPU), 384-bit bandwidth, 480GB/s bandwidth, 824MHz core frequency, 300W power consumption, passive heat dissipation design ‌. ‌ ‌

1 ‌ ‌

2

Youdaoplaceholder0 Core parameters ‌

Youdaoplaceholder0 Architecture and core ‌

It adopts the Kepler architecture and is equipped with two GK210 GPU cores. Physically, it integrates two independent Gpus with 12GB of video memory within one card. ‌ ‌

1 ‌ ‌

3

Each GPU contains 2,496 CUDA cores (totaling 4,992), with a core frequency of 824MHz. ‌ ‌

1 ‌ ‌

4

Youdaoplaceholder0 Memory configuration ‌

The video memory type is GDDR5, with a total capacity of 24GB (12GB per GPU), a bit width of 384 bits, and a bandwidth of 480GB/s. ‌ ‌

1 ‌ ‌

2

Supports ECC error correction and Unified Virtual Address Space (UVA). ‌ ‌

4

Youdaoplaceholder0 Performance & power ‌

The performance of double-precision floating-point is 2.91 TFLOPs, and that of single-precision is 8.74 TFLOPs. ‌ ‌

2

The maximum power consumption is 300W. It requires an 8-pin power supply interface and PCIe 3.0 x16 bus. ‌ ‌

1 ‌ ‌

3

Youdaoplaceholder0 Design features ‌

Youdaoplaceholder0 Cooling method ‌ : Passive cooling design, relying on server air ducts for cooling, not suitable for ordinary workstations. ‌ ‌

3

Youdaoplaceholder0 Dual-GPU logic ‌ : Must be identified as two separate Gpus during programming and interconnected via a PCIe bridge chip (PLX). ‌ ‌

3

Youdaoplaceholder0 Application scenarios ‌

Optimized for high-density parallel computing such as machine learning and scientific computing, but due to its outdated architecture (released in 2014), the performance of modern AI tasks is relatively low