博客
关于我
强烈建议你试试无所不能的chatGPT,快点击我
NVIDIA Tesla C2075 vs Tesla K10 theoretical performance
阅读量:2444 次
发布时间:2019-05-10

本文共 4207 字,大约阅读时间需要 14 分钟。

Each graphics unit has several vital theoretical parameters that affect real-world game, 3D graphics and compute performance. These are texture fillrate, pixel fillrate, memory bandwidth, along with single- and double-precision performance. Below you will find why they are important, and which card has better characteristics.

Pixel fill rate (gigapixels/s)

60
48
36
24
12
0
 
27.6
 
47.7
 
 
Higher is better
The Tesla K10 card has more Raster Operations Pipelines (ROPs) than the NVIDIA Tesla C2075. On top of that, its graphics clock rate is higher, therefore its pixel fillrate is substantially higher. Better maximum pixel fill rate means that the GPU can draw more pixels on and off screen each second, increasing overall performance, unless the card is limited by something else, such as texture fillrate, memory bandwidth or CPU speed.
  - NVIDIA Tesla C2075       - NVIDIA Tesla K10

Texture fill rate (gigatexels/s)

300
240
180
120
60
0
 
32.2
 
191
 
 
Higher is better
Because the NVIDIA Tesla K10 graphics unit has many more TMUs (Texture Mapping Units) and higher graphics frequency, its texture fillrate is considerably higher. Better texture fill rate means that the GPU can use more complex 3D effects and/or apply more textures to each textured picture element, which improves games visual appearance.

Single Precision performance(GFLOPS)

6000
4800
3600
2400
1200
0
 
1030
 
4577
 
 
Higher is better
Maximum Single Precision performance indicates how fast the graphics card is at executing programs, that process primarily single-precision floating point data. The performance is expressed in GFLOPS or billions of Floating Point Operations Per Second. Generally, the more stream processors or CUDA cores the graphics card has, and the the faster they run at, the higher Single Precision performance will be. The NVIDIA Tesla K10 GPU has an upper hand here. Higher single-precision performance number means the graphics card will perform better in general computing applications. Since CUDA cores or stream processors are also used as vertex and geometry shaders for 3D image generation, higher performance is also beneficial to games.

Double Precision performance(GFLOPS)

600
480
360
240
120
0
 
515
 
191
 
 
Higher is better
Maximum Double Precision performance is similar to the Single Precision performance, except that it applies to double-precision (64-bit) floating point operations. Since games do not use double-precision arithmetics, this characteristic is unimportant to games performance. The Tesla C2075 is faster when processing 64-bit floating-point numbers.
  - NVIDIA Tesla C2075       - NVIDIA Tesla K10

Memory bandwidth (GB/s)

400
320
240
160
80
0
 
144
 
320
 
 
Higher is better
To speed up processing, the GPUs store 3D scene data, textures and intermediate data, used for image generation, in on-board memory. The video memory usually has much higher bandwidth than system RAM, and more bandwidth allows the GPU to run at higher display resolutions, use larger and more detailed textures, and apply more complex 3D effects and filters. The bandwidth depends on a few components, such as memory type, speed, and memory interface width. Specifically, higher memory bandwidth of the NVIDIA Tesla K10 is due to higher memory clock.

NVIDIA Tesla C2075 vs Tesla K10 specs comparison

All rows with different specifications or features are highlighted.

     

General information

Market segment HPC / Server
Manufacturer NVIDIA
Model

Architecture / Interface

Die name GF100 2 x GK104
Architecture Fermi Kepler
Fabrication process 40nm 28nm
Bus interface PCI-E 2.0 x 16 PCI-E 3.0 x 16

Cores / shaders

CUDA cores 448 3072
ROPs 48 64
Pixel fill rate 27.6 gigapixels/s 47.68 gigapixels/s
Texture units 56 256
Texture fill rate 32.2 gigatexels/s 190.72 gigatexels/s
Single Precision performance 1030.4 GFLOPS 4577.28 GFLOPS
Double Precision performance 515.2 GFLOPS 190.72 GFLOPS

Clocks / Memory

Base clock   745 MHz
Graphics clock 575 MHz  
Processor clock 1150 MHz  
Memory size 6144 MB 8192 MB
Memory type GDDR5
Memory clock 750 MHz 1250 MHz
Memory interface width 384 256
Memory bandwidth 144 GB/s 320 GB/s

Other features

Maximum power 247 Watt 250 Watt

Better values / features are marked with green color, and worse values are in red color.

转载地址:http://jtiqb.baihongyu.com/

你可能感兴趣的文章
Chrome 27的新功能
查看>>
浏览器趋势(2013年5月):IE8降至10%以下
查看>>
谁偷了我的CPU?
查看>>
Microsoft将IE10更新推送到Windows 7
查看>>
验证码放缓存里_浏览器趋势2013年8月:夏季放缓?
查看>>
liferay_云中的Liferay
查看>>
SQL或NoSQL:Google App Engine-第1部分
查看>>
SitePoint Podcast#178:Web设计过程和创造力
查看>>
移动端获取视频第一帧移动端_后端即服务-第1部分
查看>>
畅谈理想未来为主题的铅笔画_与专家畅谈Node.js
查看>>
SitePoint Podcast#173:释放混乱的猴子
查看>>
unity 暴风雨天气效果_浏览器趋势2012年10月:暴风雨前的平静?
查看>>
php 查询成绩_与专家讨论PHP: 成绩单
查看>>
一年新的一年_一年的云创新
查看>>
使用PHP从Access数据库中提取对象,第2部分
查看>>
openbiz_Openbiz Cubi:健壮PHP应用程序框架,第1部分
查看>>
使用PHP从Access数据库中提取对象,第1部分
查看>>
使用云waf的案例_9种流行的云使用案例
查看>>
类集合转换类集合_PHP中的集合类
查看>>
使用SimplePie消费Feed
查看>>