|
v****e
发帖数: 10715 | 2 【 以下文字转载自 PDA 讨论区 】
发信人: veggie (走位飘忽), 信区: PDA
标 题: 贴个智能手机CPU GPU对比
发信站: BBS 未名空间站 (Thu Jan 21 17:28:01 2010, 美东)
都来自网上
CPU：
SNAPDRAGON @ 1GHZ = 2100 MIPS
Samsung Omnia @ 624 MHZ = 800 MIPS
Iphone 3GS @ 600 MHZ = 1200+ MIPS
INTEL ATOM @ 1.6 GHZ = 3800+ MIPS
Snapdrgaon约等于 Intel Atom @ 900 Mhz
GPU：
Current Mobile GPUs: Triangles/sec, Fill Rate
PowerVR MBX-Lite (iPhone 3G): 1 M triangles/s, 100 M pixels/s
Samsung S3C6410 (Omnia II): 4 M triangles/s, 125.6 M pixels/s
PowerVR SGX 520 (Palm Pre): 14 |
|
T******T
发帖数: 3066 | 3 Cooling wise it should be fine, they are all dual slot GPUs, with heat
pipe and air vent pump directly outside of the case. |
|
q********g
发帖数: 10694 | 4 有点私活儿，有人赞助，要上GPU了。考虑了很久，到底上Tesla呢还是3个GTX480呢。
目前个人想到的后者的好处是，后者以后处理二手容易。欢迎给点意见。
Form Factor 9.75" PCIe x16 form factor
# of Tesla GPUs 1
# of CUDA Core 448
Frequency of CUDA Cores 1.15 GHz
Double Precision floating point performance (peak) 515 Gflops
Single Precision floating point performance (peak) 1.03 Tflops
Total Dedicated Memory*
Tesla C2050 3GB GDDR5
Tesla C2070 6GB GDDR5
Memory Speed 1.5 GHz
Memory Interface 384-bit
Memory Bandwidth 144 GB/sec
Power Consumption 247W TDP
System Interface |
|
|
|
|
s***q
发帖数: 10585 | 8 XFX 5970 Black 4GB card - 6 monitor support $1,199.99 Newegg
Currently the fastest dual-GPU-on-a-PCB card on the planet, the XFX HD-
597X-ENFN Radeon HD 5970 Black Edition comes with 6 native mini
DisplayPort and 4GB GDDR5 memory, purposely designed for up to six-
display output with ATI Eyefinity technology. Powered by two brutally
fast GPUs and 3200 stream processors, the Radeon HD 5970 delivers
shockingly fast frame rates and is supported by other most robust
advanced features including ATI S |
|
T****n
发帖数: 6187 | 9 China's Tianhe-1A is world's fastest supercomputer, plans to usurp the West
now complete
It happened. China just passed the US and the world with the reveal of the
world's fastest supercomputer. The fully operational Tianhe-1A, located at
the National Supercomputer Center in Tianjin, scored 2.507 petaflops as
measured by the LINPACK benchmark. That moves it past Cray's 2.3 petaflops
Jaguar located at Oak Ridge National Lab in Tennessee. Tianhe-1A achieved
the record using 7,168 NVIDIA Tesla M205... 阅读全帖 |
|
|
k*****y
发帖数: 744 | 11 Codec: CoreAVC Video Decoder 3.0.1
Bench Software: DXVAChecker
Video Clip: Ducks Take Off 1080p - 108Mbps
GPU: GMA4500m, HD3000, GTX460
Result:
GMA4500m
Frame Rate(fps): Avg 38, Min 32, Max 45
CPU Usage(%): Avg 07, Min 04, Max 13
HD3000
Frame Rate(fps): Avg 134, Min 118, Max 144
CPU Usage(%): Avg 05, Min 03, Max 06
GTX460
Frame Rate(fps): Avg 21, Min 16, Max 26
CPU Usage(%): Avg 01, Min 00, Max 03
GTX460 PV4 DXVA硬解结果基本跟
http://www.anandtech.com/show/4380/discrete-htpc-gpus-shootout/
这里一致。I... 阅读全帖 |
|
|
|
p********n
发帖数: 1707 | 14 "The same encoding using only software took 172 seconds. The same encoding
took 83 or 86 seconds GPU-assisted, using a Nvidia GeForce GTX 570 and a AMD
Radeon HD 6870 respectively, both of which are contemporary high end GPUs." |
|
c********l
发帖数: 8138 | 15 List of cryptocoins: （怕被无脑钻风删贴，故不加link了，自己找）
MAJOR:
-BTC Bitcoin - first, strongest, most accepted, most mined, high volume
market, a true currency
-LTC Litecoin - second only to BTC , faster than BTC, Small efficiency gap
between GPUs and CPUs , ASIC - proof
-NMC Namecoin merged mined with BTC, used for alternative p2p domain system
MINOR:
-PPC PPcoin Proof Of Stake [very innovative, low energy ] , compatibile
with BTC miners
-TRC Terracoin based on BTC, fast difficulty adjustment - miner-j... 阅读全帖 |
|
t**t
发帖数: 27760 | 16 MANNHEIM, Germany; BERKELEY, Calif.; and KNOXVILLE, Tenn.--Tianhe-2, a super
computer developed by China’s National University of Defense Technology, is
the world’s new No. 1 system with a performance of 33.86 petaflop/s on the
Linpack benchmark, according to the 41stedition of the twice-yearlyTOP500 l
ist of the world’s most powerful supercomputers. The list was announced Jun
e 17 during the opening session of the 2013 International Supercomputing Con
ference in Leipzig, Germany.
Tianhe-2, or M... 阅读全帖 |
|
|
|
|
|
G****A
发帖数: 4160 | 21 貌似AMD官方回应了
"As you know, we continuously tune our GPUs in order to maximize their
performance within their given power envelopes and the speed of the memory
interface, which in this case is an unprecedented 8 Gbps for GDDR5. Recently
, we identified select scenarios where the tuning of some RX 480 boards was
not optimal. Fortunately, we can adjust the GPU's tuning via software in
order to resolve this issue. We are already testing a driver that implements
a fix, and we will provide an update to ... 阅读全帖 |
|
F***Q
发帖数: 6599 | 22
GPUs expire quite rapidly these days ... 1080 Ti is expected to be launched
in Jan at CES (or even by the end of this year), and Volta will be on the
horizon by GTC 2017 ...
the mini-atx mobo kills this deal. there is only one PCIe x 16 slot on this
motherboard. it is not going to work for serious GPU computing tasks.
you need a motherboard with at least 2 PCI-e x 16 slots, and install two
graphics cards - one dedicated for display (a low-end one), the other
dedicated for computing. Otherwise, ... 阅读全帖 |
|
F********r
发帖数: 1748 | 23 Intel Says Iris and Iris Pro Graphics Can Outperform 80% of Discrete GPUs –
Casual and Mainstream Users Don’t Need dGPUs
http://wccftech.com/intel-iris-pro-graphics-gamers/
跑分肯定不会比GTX950M这档差。但是我觉得来自硬件厂商（笔记本制造商以及显卡制
造商）的阻力可能非常大，导致现在有价无货的局面。没人用的情况下游戏厂商当然不
会为它优化了。 |
|
|
|
F***Q
发帖数: 6599 | 26 really depends on what kind of tasks you want to run - does it require
double precision or single precision? is it memory-bound or compute bound?
how much memory do you need? can your kernel use multiple GPUs or just one?
in my case, I care more about number of cores than the size of memory, so
the best consumer card (Geforce) is almost always my most cost-effective
choice (as oppose to a Tesla or a Quadro of the same cores). |
|
|
b**a
发帖数: 1118 | 28 I got a ryzen 1700x and a few gpus. I heard it is possible to mine using
both cpu and gpu at the same time, but i can't find any details on how to do
it. Anyone can point me a direction? Thanks in advance. |
|
|
t*****y
发帖数: 17 | 30 I am not expert on video cards. From what I can tell, RX 480 is better than
GT 730 on performance:
http://gpuboss.com/gpus/Radeon-RX-480-vs-GeForce-GT-730
You might need config displays in OS to match the physical location of
monitors. This might be the issue in your first trial with mouse movement.
1TB |
|
f*******t
发帖数: 7549 | 31 不是单机跟李世石打吧， 1920 CPUs and 280 GPUs |
|
l*******m
发帖数: 1096 | 32 不明白你的model parallel定义。大公司标配是single box with 8 gpus |
|
l*******m
发帖数: 1096 | 33 cuda, nvcc都是binary. 如果cpu速度快些，轮子马上就多了。nvidia的优势是gamers
gpus价格低，一般人都能入手 |
|
l******n
发帖数: 9344 | 34 这个配置太豪华了
Four TITAN X GPUs with 12GB of memory per GPU64GB DDR4Asus X99-E WS
workstation class motherboard with 4-way PCI-E Gen3 x16 supportCore i7-5930K
6 Core 3.5GHz desktop processorThree 3TB SATA 6Gb 3.5” Enterprise Hard
Drive in RAID5512GB PCI-E M.2 SSD cache for RAID250GB SATA 6Gb Internal
SSD1600W Power Supply Unit from premium suppliers including EVGA |
|
l*******m
发帖数: 1096 | 35 CPU:
An L1 cache can be accessed by one core only.
GPU:
An L1 cache can be shared by a block of threads (64 or more)
Besides L1 caches, GPU has shared memory, which can be regarded as an
explicitly managed (in your code) cache. Shared memory can be shared by a
block of threads as well.
Therefore, GPUs are able to provide more efficient memory IO given specific
operations at least. |
|
|
l*******m
发帖数: 1096 | 37 而且是大力出奇迹，他们的训练是8个TITAN X GPUS转个3-4天 |
|
|
L****8
发帖数: 3938 | 39 I am going to buy workstation with 4 Titan V GPUs |
|
c*******h
发帖数: 1096 | 40 if you have a single core cpu, there is no way to do real parallelism.
(well, you can do some tricks on gpu and achieve somewhat parallelism,
but i don't think you can use matlab on gpus.)
parallel computing is for a multi-core cpu, or a cluster of cpus on either
one or many machines.
sing |
|
x*****u
发帖数: 3419 | 41 July/August 2009 Copublished by the IEEE CS and the AIP 1521-9615/09/$25.00
2009 IEEE
GettinG Started with GPU ProGramminG
By Michael A. Gray
This tutorial describes a step-by-step procedure for programming a Macintosh
Nvidia GPU. General scientific
programmers with some C knowledge can get started in parallel processing
application development with relative ease.
A recent surge in articles describes the use of graphics processing units (
GPUs) for scientific computation.[1,2] This isn’t surpris |
|
c*******3
发帖数: 21 | 42 Job Title: CUDA Compute Developer Technology Engineer
Location: Beijing
Job Description/Qualifications:
To spearhead its breakthroughs in High Performance Computing and Visual
Computing, NVIDIA is building a team with a unique and critical position
within the company and is looking for passionate world-class software
engineers.
In this role, you will open up new domains of application for GPUs by
researching and developing GPU computing algorithms, driving their adoption
with key application dev... 阅读全帖 |
|
w******t
发帖数: 369 | 43 【 以下文字转载自 JobHunting 讨论区 】
发信人: wishlist (wishlist), 信区: JobHunting
标 题: Signal Integrity Characterization Engineer needed.
发信站: BBS 未名空间站 (Wed Mar 21 14:19:32 2012, 美东)
SIGNAL INTEGRITY CHARACTERIZATION ENGINEER
In this position, you will work in the centralized physical design
engineering group. You will be responsible for power delivery
characterization and correlation. You will work closely with circuit
designers and signal integrity engineers to characterize interfaces
connected to Nvidi... 阅读全帖 |
|
b*****e
发帖数: 41 | 44 Dr. Wei Zhang, a professor in the Electrical and Computer Engineering
Department at Virginia Commonwealth University, is currently looking for
highly-motivated graduate students to perform research in the fields of real
-time and embedded systems, compiler, computer architecture, and graphics
processing units (GPU) computing.
Applicants should have a M.S or B.S degree in Computer Science, Computer
Engineering or a closely related discipline. Excellent programming skill in
C/C++ (or Java) is a mu... 阅读全帖 |
|
C*****5
发帖数: 8812 | 45 我是没有研究那么深啦，但是上次看某大牛的视频好像提到过。反正实践上deep net
training现在就是搞尽量多的数据，用GPUs+hadoop，开始总归有不收敛或者收敛太慢
的问题，慢慢调参数总能调出来的。
deep |
|
f****y
发帖数: 243 | 46 。。。。。。蛋疼
：我是没有研究那么深啦，但是上次看某大牛的视频好像提到过。反正实践上deep net
：training现在就是搞尽量多的数据，用GPUs+hadoop，开始总归有不收敛或者收敛太慢 |
|
