NVIDIA Tesla M60 : Specifications, Architecture, Working, Differences & Its Applications

The NVIDIA Tesla M series professional-grade GPUs were designed for data centers and high-performance computing based on the Maxwell and Fermi architectures. These accelerators were primarily designed for data center applications such as deep learning inference, virtual desktops, and high-performance computing (HPC). Its key models mainly include the M2070, M2050, M2090 (Fermi), M60 (Maxwell) and the M40. Thus, these models provide significant performance gains within data analytics, professional visualizations, and scientific simulations. These cards can be characterized by various features, such as ECC memory, single precision, and high double precision, as well as particular cooling requirements, mainly for server environments. This article provides an in-depth examination of the NVIDIA Tesla M60 GPU, its operation, and its applications.

What is the NVIDIA Tesla M60 GPU?

The Tesla M60 GPU was a professional graphics card, launched by NVIDIA on August 30th, 2015. It is based on Maxwell architecture with a 28 nm process based on the GM204 graphics processor. Therefore, the GM204 graphics processor is a large chip with 398 mm² of die area & 5,200 million transistors.

Tesla M60 features shading units – 2048, texture mapping units – 128, and ROPs – 64. NVIDIA, with the Tesla M60, has connected GDDR5 memory – 16 GB using a 256-bit memory interface for each GPU. The GPU can be operated at 557 MHz to 1178 MHz frequency. The NVIDIA Tesla M60 features a dual-slot card that draws power from an 8-pin power connector with a power draw rated at 300 W maximum.

Specifications

The NVIDIA Tesla M60 specifications include the following.

• GPU Architecture is NVIDIA Maxwell™
• It features two GPUs for each Board.
• Max users for each Board – 16 or 32 per GPU.
• NVIDIA CUDA® Cores – 4096.
• GPU memory – 16 GB of GDDR5 memory/8 GB for each GPU.
• H.264 1080p30 streams – 36
• Maximum power consumption is 300 Watts.
• Thermal solution is active or passive.
• Form factor – PCIe 3.0 Dual Slo
• Memory Bandwidth is 320 GB/s total or 160.4 GB/s for each GPU.
• Memory Interface is 512-bit or 256-bit for each GPU.
• Its boost Clock is 1178 MHz
• Single-Precision Performance is 9.6 TFLOPS
• Double-Precision Performance is 440 GFLOPS
• It features an 8-pin power connector.
• It supports up to 32 users for each board for workstations and virtual desktops.
• Its ECC Protection improves reliability
• This GPU supports NVIDIA GRID vGPU for virtualized graphics
• It features H.265 and H.264 encoding capabilities.
• It adjusts CLK speeds dynamically for the highest performance within the power limit.

NVIDIA Tesla M60 Architecture

The NVIDIA Tesla M60 is a dual-GPU-based accelerator board that uses the Maxwell architecture. It is designed for high-performance computing and virtualized workstations. In addition, this GPU features dual GM204 GPUs with CUDA cores – 4096 & GDDR5 memory – 16GB. It connects through PCI-E 3.0 x16, which can be optimized for virtual environments through features like ECC memory and NVIDIA GRID support.

NVIDIA Tesla M60 Architecture Components

This Architecture can be built with different components like CUDA cores, memory, memory interface, shading units, texture mapping units, ROP Units, Host Interface, NVENC hardware encoder, cooling, ECC protection, etc. So these components can be explained below.

GPUs

The NVIDIA Tesla M60 GPU features two GM204 GPUs, based on the NVIDIA Maxwell 2.0 architecture. These GPUs on the card contain 4096 CUDA cores, where each GPU has 2048 CUDA cores and GDDR5 memory – 16 GB, where 8 GB is for each GPU. It is a dual-GPU accelerator that is designed for professional virtual workstations, high-performance computing, and data center virtualization.

CUDA cores

This architecture contains parallel processing units like 4096 CUDA cores in total, where 2048 cores are on each of its two GPUs. These processing units perform CUDA code by allowing the GPU to execute various computations concurrently for deep learning & scientific computing applications.

Memory

The NVIDIA Tesla M60 GPU features GDDR5 memory – 16 GB, where 8 GB of memory is allocated to each GPU. Thus, this memory can be protected by Error Correction Code (ECC) for better reliability, with a total of 320 GB/s or 160.4 GB/s of memory bandwidth for each GPU.

Memory Interface

The memory interface of this architecture is GDDR5 with a 512-bit total width & 320 GB/s memory bandwidth. Thus, this interface connects the two Maxwell architecture GPUs on the same board to their particular memory pools, which allows them to execute calculations through reading & writing data.

Shading Units

The shading units in this processor are shading units or universal processing units, which manage a wide range of tasks like pixel, geometry shading, vertex & general-purpose compute tasks. In addition, this architecture features shading units – 2048 for each GPU, which represents a unified design where a single type of processor can perform different shader programs. The main function of shading units is to have specific processors that manage specific parts of the parallel computing and graphics pipeline tasks.

Texture Mapping Units

The NVIDIA Tesla M60 architecture features texture mapping units -128 per GPU. These are responsible for sampling & filtering textures to apply them to 3D models. In addition, these TMUs can perform different operations like revolving & resizing images to make the texels. These can be displayed on a 3D model with texture coordinates to discover the accurate color from a texture image.

ROP Units

The ROP units, or Render Output Units, are the final hardware stage of the rendering pipeline in the NVIDIA Tesla M60 architecture. Each of the two GM204 graphics processors in this card features 64 ROPs, resulting in 128 ROPs in total for the whole Tesla M60 accelerator. It processes and writes the complete pixel data into the frame buffer. These units perform the last pixel-level operations by taking the processed pixel & texel information and changing it into an ultimate depth or pixel value. The ROP count with clock speed is a main factor that determines the fill rate of the GPU, which measures how many pixels the graphics card can make per second.

Host Interface

The host interface of the NVIDIA Tesla M60 GPU is PCI Express 3.0 x16. Thus, it allows the graphics card to connect to the motherboard for communicating with the CPU & other system components. This interface is the physical connection that installs the graphics card within a PCIe slot of a computer. Thus, it enables high-speed data transmission between the GPU and the remaining system.

NVENC Hardware Encoder

The NVIDIA Tesla M60 GPU performs video encoding with its dedicated NVENC hardware encoder. It offloads the task to a particular part of the GPU from the CPU for accelerated performance. This ability can be supported through the NVIDIA Video Codec SDK through APIs, which is accessible on Kepler-generation GPUs and above. This performance mainly depends on different factors like the particular encoder used, the accessible memory bandwidth, and the video quality settings.

Cooling

The NVIDIA Tesla M60 GPU is designed for server environments that use passive cooling, which depends on the high-speed internal fans of a server to push air throughout its heatsink. Once this server is used outside of a server, then it needs an active or a custom cooling solution, like connecting a blower fan to force air using the heatsink or executing a water-cooling loop.

Software System

The NVIDIA GRID is the software system of the NVIDIA Tesla M60 GPU that allows virtualized high-performance workstations. The NVIDIA vGPU software enables a GPU to be shared between various virtual machines. In addition, its key components mainly include the guest drivers and a necessary host for the supported OS, such as Windows Server/ Linux. An NVIDIA license server handles the vGPU software, and also utilities, potentially including firmware for the particular server hardware.

Difference between GeForce RTX 4090 and Tesla M60

The difference between the GeForce RTX 4090 and Tesla M60 GPUs includes the following.

Advantages

The advantages of NVIDIA Tesla M60 include the following.

• The NVIDIA M60 on a single board features two high-end Maxwell GPUs.
• It supports up to 32 users for each board to make it very efficient for virtualized environments.
• It speeds up complex data tasks & professional applications with 4096 CUDA cores.
• This graphics card features GDDR5 memory- 16 GB, where 8 GB is allocated to every GPU to handle large datasets.
• The ECC (Error-Correcting Code) memory protection is included for better data integrity and reliability.
• It delivers high efficiency and throughput for server-based workloads.
• It provides accelerated graphics & performance for workstations and virtual desktops.
• It is suitable for scientific simulations & other complex computational tasks.
• This GPU can deliver better graphics performance for a variety of professional applications.

Disadvantages

The disadvantages of NVIDIA Tesla M60 include the following.

• The M60 is not optimized for recent AI workloads.
• This GPU lacks the particular tensor cores that are standard within the latest GPUs.
• Its memory bandwidth is low.
• The Tesla M60 GPU has 300W highest power consumption.
• The M60 needs proper and significant server cooling. Eventually, inefficient performance for each watt changes into higher operating costs, mainly for data centers.
• The single-precision floating-point performance of this card is low by present standards.

Applications

The applications of NVIDIA Tesla M60 include the following.

• It trains deep neural networks for image recognition tasks.
• It allows more accurate and quicker simulations in the astrophysics field.
• It speeds up a broad range of data center workloads.
• This GPU can create high-performance virtual workstations that run professional graphics applications, which can be accessed from any device.
• It powers servers to run & stream the games to low-power devices.
• This GPU centralizes data and applications within the data center to give accelerated virtual desktops to many users.
• It performs Monte Carlo simulations to represent financial asset values and know potential losses and gains.
• This GPU speeds up the machine learning models used to predict the chance of loan defaults.
• This GPU supports demanding graphics-intensive design & visualization tasks.
• It is used in cloud computing environments for a variety of acceleration tasks.

Thus, this is an overview of the NVIDIA Tesla M60 GPU. It is a specialized & dual-GPU accelerator, designed for use in data centers with NVIDIA GRID software for high-performance graphics virtualization. In addition, it can also deliver high-performance virtual workstations & desktops due to its dual-GPU design & ECC memory features. Thus, it provides a 2X performance increase within graphics-accelerated applications as compared to earlier generations to support a wide range of specialized use cases. Here is a question for you: What is GeForce RTX 4090?