f



Popular Inexpensive Vector Processing or Parallel Add-in Board?

Some time ago, a number of papers were published on using the vertex
shader chips in modern video cards to perform specialized massive
floating-point computations. In general, the means of programming these
directly is only made available under NDAs to game companies, and, now,
what with DirectX, even that isn't needed, as the manufacturers just
need to write their own drivers. One company, 3dfx, made the details of
its cards public, but they're no longer in business, having been bought
out by another video card maker.

However, such chips tended to work either with integer arithmetic, or
with low-precision special floating-point types that were 16 or perhaps
24 bits long. Recently, however, partly due to changes in the DirectX
spec, there has been a trend to 32 bit floating-point.

The CELL Broadband Engine is a new chip with very impressive
performance. At least with 32-bit floating-point numbers, but it just
uses its internal single-precision ALUs for double-precision work,
leading to raw performance numbers that *still* exceed those of
ordinary microprocessors, but which are not quite as impressive: 15
Gflop/s as against over 200 Gflop/s. While CELL BE blades and add-in
cards are quite expensive, units which may have one bad SPU are used in
the rather inexpensive PlayStation 3 console.

The Japanese GRAPE project has led to the creation of chips with an
immense amount of floating-point power on the chip, but designed for
specialized use only.

All this, though, tends to suggest that the technology is *there* for
an add-in board for the PC platform that would increase its raw
floating-point performance perhaps a hundredfold. Such a piece of
equipment would seem to me to be worthwhile, and would produce more
versatile setups than simply going directly to a Beowulf cluster with
ordinary, unmodified PC hardware.

There seem to have been *some* attempts at this.

There was the ClearSpeed CSX600 chip. Whatever became of it?

Microway seems to advertise a number of products aimed at this market,
but as far as I can see, they are based on relatively conventional
microprocessors.

Am I just looking in the wrong place, or is this market not being
served?

John Savard

0
jsavard (654)
11/26/2006 8:20:06 PM
comp.sys.super 442 articles. 0 followers. Post Follow

5 Replies
519 Views

Similar Articles

[PageSpeed] 57

I wrote:
> There was the ClearSpeed CSX600 chip. Whatever became of it?
>
> Microway seems to advertise a number of products aimed at this market,
> but as far as I can see, they are based on relatively conventional
> microprocessors.

Also, I see that Texas Memory Systems' TM-44 chip, used in products
that at first glance seem to be what I'm looking for, also specialize
in 32-bit floating-point; and Bittware, another company that turned up
in some further searching, also, like TMS, addresses the signal
processing market.

John Savard

0
jsavard (654)
11/26/2006 8:27:19 PM
I wrote:
> I wrote:
> > There was the ClearSpeed CSX600 chip. Whatever became of it?
> >
> > Microway seems to advertise a number of products aimed at this market,
> > but as far as I can see, they are based on relatively conventional
> > microprocessors.
>
> Also, I see that Texas Memory Systems' TM-44 chip, used in products
> that at first glance seem to be what I'm looking for, also specialize
> in 32-bit floating-point; and Bittware, another company that turned up
> in some further searching, also, like TMS, addresses the signal
> processing market.

On the other hand, there are systems from Orion and Tyan, and these
also use conventional microprocessors, but they put a large number of
them into one box.

Then there's an add-in card which is said to be useful for
floating-point applications, but which is built using FPGA chips; this
doesn't seem right to me - this one is produced by DRC Computer, as the
RPU100 and the RPU110 coprocessor.

John Savard

John Savard

0
jsavard (654)
11/26/2006 9:43:29 PM
Majsavard@ecn.ab.ca wrote:
> I wrote:
>> I wrote:
> 
> Then there's an add-in card which is said to be useful for
> floating-point applications, but which is built using FPGA chips; this
> doesn't seem right to me - this one is produced by DRC Computer, as the
> RPU100 and the RPU110 coprocessor.
> 
> John Savard
> 
> John Savard
> 

May also be worth revisiting the 3D card as a source of cheap FP 
processing. ATI recently released a dev kit that goes below the 
directx/opengl interfaces. See 
http://www.reghardware.co.uk/2006/11/14/amd_ships_ctm_stream/
0
news9805 (4)
11/28/2006 10:20:24 PM
jsavard@ecn.ab.ca wrote:
> Some time ago, a number of papers were published on using the vertex
> shader chips in modern video cards to perform specialized massive
> floating-point computations. In general, the means of programming these
> directly is only made available under NDAs to game companies, and, now,
> what with DirectX, even that isn't needed, as the manufacturers just
> need to write their own drivers. One company, 3dfx, made the details of
> its cards public, but they're no longer in business, having been bought
> out by another video card maker.
> 
> However, such chips tended to work either with integer arithmetic, or
> with low-precision special floating-point types that were 16 or perhaps
> 24 bits long. Recently, however, partly due to changes in the DirectX
> spec, there has been a trend to 32 bit floating-point.
> 

May also be worth revisiting the 3D card as a source of cheap FP 
processing. ATI recently released a dev kit that goes below the 
directx/opengl interfaces. See 
http://www.reghardware.co.uk/2006/11/14/amd_ships_ctm_stream/
0
news9805 (4)
11/28/2006 10:32:20 PM
I'm not sure exactly what you are looking for but maybe if you told me
what your application is. Have you looked at nVidia 8800 cards? They
have a more general architecture than past cards and if you contact
them they might be willing to work with you.

Derek

James Wilson wrote:
> jsavard@ecn.ab.ca wrote:
> > Some time ago, a number of papers were published on using the vertex
> > shader chips in modern video cards to perform specialized massive
> > floating-point computations. In general, the means of programming these
> > directly is only made available under NDAs to game companies, and, now,
> > what with DirectX, even that isn't needed, as the manufacturers just
> > need to write their own drivers. One company, 3dfx, made the details of
> > its cards public, but they're no longer in business, having been bought
> > out by another video card maker.
> >
> > However, such chips tended to work either with integer arithmetic, or
> > with low-precision special floating-point types that were 16 or perhaps
> > 24 bits long. Recently, however, partly due to changes in the DirectX
> > spec, there has been a trend to 32 bit floating-point.
> >
>
> May also be worth revisiting the 3D card as a source of cheap FP
> processing. ATI recently released a dev kit that goes below the
> directx/opengl interfaces. See
> http://www.reghardware.co.uk/2006/11/14/amd_ships_ctm_stream/

0
dereks314 (31)
12/31/2006 1:27:26 PM
Reply:

Similar Artilces:

Board architecture choice? parallel or serial processing?
Hi all, I'm strugling making the right choice for our new board. I Need something like 2 GFLOPS in order to process audio. Audio is sampled with 44Ks/s externally and internally at 10xFS. I have been aiming at the ADSP-21161 or ADSP-21261 as a target. I know there are 2 ways of processing; Parallel and serial. I have no clue how to create a parallel system ( bus arbiter, software ). Does VDSP divides the code over the DSP's or do you have to program th parallel processes manually? I do know how to create a serial system, but it's not that flexible. Anyhow the board is less compl...

Add a short vector to a large vector
I have a vector "y" which is 100 million rows and a vector "win" that has a length of 50. The vector indicates the area of interest within vector "y". I was wondering if there is a better way of doing the addition other than... d = length(y) - win(end); S = [false(win(1)-1,1); true(length(win),1); false(d,1)]; y(S) = y(S) + 1; For a single operation this is not a problem but I am reading line by line from a text file that has 30 million data points. Any thoughts would be greatly appreciated. Ivan Did you mean to do something like this? (It makes it easier if...

[ANN] parallel 1.3: simple parallel processing.
Version 1.3 of ssh_parallel and parallel have been released. This release features much better error handling in ssh_parallel, so network errors are less likely to cause jobs to go missing. Additionally, it allows you to specify a nice level for processes. Same as the last time, I'm interested in any comments/critiques anyone may have. Original release notes (with updated URLs): If you ever write a shell script along the lines of: some | pipieline | bash then you may be interested in my program parallel. The above line would be written as: some | pipeline | parallel 4 to perform ...

can not dynamically add element to vector in going over the vector
Hi, I just found out that in looping the vector, you can not push_back() to the vector even in some "safe" cases. Here is the code snippet, vector<pair<UINT32, UINT32> >::iterator itr; for (itr = vec.begin(); itr< vec.end(); ++itr){ pair<UINT32, UINT32> p1, p2; ... assign value of p1 and p2 ... if (..in some condition...){ vec.push_back(p1); vec.push_back(p2); vec.erase(itr); return 0; } } I think this is a safe dynamic addition to the vector as you can see once something is added to t...

Parallel processing
Hello, I'm trying to implement a parallel process, and I'm not sure how to set it up and I was wondering if someone could help me please? My code resembles, for i=1:N % N approx = 10^6 x_new = my_func(x_old) A(:,i) = x_new; x_old = x_new; end where x_old and x_new are Mx1 vectors and A is a MxN matrix. From one iteration to the next, the only dependence is x_new(i) on x_old(i). I'd like to parallelise this by splitting the elements of x_new and x_old across more than one core. Is there a way I can, say, matlabpool open 5 <use core's ...

parallel processing
Hi, I'm not quite sure what 'feature' i'm looking for ... any input appreciated. I want to parallelize a particular task. #/usr/bin/perl -w use strict; my @target; our @result; for (my $i = 0; $i < @target; $i++) { $result[$i] = &do_some_work($target[$i]); } &report_results; .... &do_some_work requires a minute or so to complete. @target contains several hundred elements. Therefore, total execution time runs in the hundreds of minutes. Also, @target is not ordered ... e.g. there are no dependencies within @target ... if &do_some_work finishes proce...

Parallel processing
Hello, I am running version 7.01 on a quad core CPU. I am doing some heavy calculations but was too lazy to use any special commands for parallelization. However, to my surprise, when I check the performance I see that all 4 kernels a busy calculating. Does anybody have any info about this automatic parallelization of Mathematica? Daniel As far as I know, Mathematica does not paralellize unless you explicitedly tell it to. Did you see 4 Mathematica kernels running or just 4 cores being busy with unknown processes? Cheers -- SjoerdOn May 28, 12:50 pm, dh <...

Parallel processes
Is there any good reason to have two almost similar processes, with almost similar sensitivity list? I have a problem with a size of the design. I'm using Lattice LC4128V and currently the design is using 130/128 logic functions. In on vhdl block there is two processes almost similar, like shown below. I got the code from the other designer and I'm just starting with VHDL, so I ask for your help. Can I save in logic elements by combining these two processes and does it affect the functionality or timing in some way? architecture ltr of dio_write is signal IO_WRITE_tmp : std_logic; ...

parallel process
How to make parallel calculations on SKILL ? one way i can see is to launch several external processes using interprocess communication (ipc) "Andrey Orlenko" <eagle@ukr.net> wrote in message news:c4gldj$2mf2$1@news.ntu-kpi.kiev.ua... > How to make parallel calculations on SKILL ? S. Badel wrote: > one way i can see is to launch several external processes > using interprocess communication (ipc) > >>How to make parallel calculations on SKILL ? My situation: I've got program GUI on SKILL, it starts external program / sh("./program &&qu...

Vector boards!
I found this site that still sells boards that will work with the C64. Is this usefull information? It's new to me. www.vectorelect.com On Mon, 11 Dec 2006 13:03:32 -0800, Sky King wrote: > I found this site that still sells boards that will work with the C64. > Is this usefull information? It's new to me. > > www.vectorelect.com You got my attention. I think I'll e-mail them later on. -- -----> http://members.dodo.com.au/~izabellion1/tristan/index.html <----- ===== It's not pretty, it's not great, but it is mine. ===== ...

parallel processing
HI, I have a tricky problem about parallel processing using JavaScript. a script makes use of classes. 2 objects A and B are created at intialisation. The two objects make use of the same function foo(). at runtime, the action: A.foo(); B.foo(); is executed. I noticed that in any browser, the result will be that the browser executes the function foo() related to A. Then stop and execute the function foo() related to B. This leave A unfinished. Is there a possiblity to have the function called by A running at the same time than B's one? I mean having two instance of the function running ...

Vector of Vectors.
I'm having a bit of trouble manipulating an element of a vector of vectors. I'm using GNAT GPL 2008. Here is a small program to illustrate the issue: with Ada.Containers.Vectors; procedure Check is package Integer_Vectors is new Ada.Containers.Vectors (Index_Type => Natural, Element_Type => Integer); package Nested_Vectors is new Ada.Containers.Vectors (Index_Type => Natural, Element_Type => Integer_Vectors.Vector, "=" => Integer_Vectors."="); IV : Integer_Vectors.Vector; NV : Nested_Vectors.Vector; begin IV.Append(42); NV.Append(IV); NV.Element(0).Append(43); -- Compile error here. end Check; The error message is, "Actual for Container must be a variable." I understand what this means. The problem is that I'm trying to use an expression as an argument for an 'in out' parameter. What I'm having trouble figuring out is how to work around this. I tried using 'renames' to give a simple name to NV.Element(0) but that didn't fool the compiler (no surprise). I also tried playing around with anonymous access types but I couldn't figure out how to declare the Element_Type of package Nested_Vectors to be aliased. I'm not sure if it would have worked in any case. Surely there must be a straight forward way to do this! Peter Peter C. Chapin wrote: > > Surely there must be a straight forward ...

Running a process requiring super user privilege from non super user
Hi All I want to run a process, that can only be run by a super user, by a normal user. I heard that this can be done by using some uid etc. Please help me if anyone knows anything about this or the uid solution. "chets" <gautamchetana1982@gmail.com> wrote: >Hi All > I want to run a process, that can only be run by a super user, >by a normal user. I heard that this can be done by using some uid etc. >Please help me if anyone knows anything about this or the uid solution. You need to talk to your Sysadmin about sudo. Regards, Ted. =================...

Parallel Computing and "The process cannot access the file because it is being used by another process."
while using parfor loop, I get "The process cannot access the file because it is being used by another process." error generated from an executable function embedded in the parfor loop. The function opens a file "tmp.key" and writes an image file "tmp.pgm" into "tmp.key" file. However, it seems that when this parfor loop runs on parallel processors, the processors are unable to access this "tmp.key" at the same time, and hence, the error is generated. First of all, is my assesment correct? secondly, how to resolve this issue and use parfor loop successfully. Any help is greatly appreciated, please. Irteza "Syed " <irtezaa@gatech.edu> writes: > while using parfor loop, I get "The process cannot access the file > because it is being used by another process." error generated from an > executable function embedded in the parfor loop. The function opens a > file "tmp.key" and writes an image file "tmp.pgm" into "tmp.key" > file. However, it seems that when this parfor loop runs on parallel > processors, the processors are unable to access this "tmp.key" at the > same time, and hence, the error is generated. First of all, is my > assesment correct? secondly, how to resolve this issue and use parfor > loop successfully. > > Any help is greatly appreciated, please. It sounds like your assessment is correct. By ...

Running a process requiring super user privilege from non super user #2
Hi All It is a serious help required. I want to write a time synchronization service in BEA TUXEDO. This service will synchronize the time of a client machine according to the server machine. For this I wrote one tuxedo service and after compiling I gave it super user priviledge to run settimeofday function of C . I did it by making root its owner and set the s uid bit by chmod u+s. But when I give tmboot the service does not get started giving following error :- error while loading shared libraries: libtux.so: cannot open shared object file: No such file or directory if I remove the set uid bit it gets booted but the time of the client machine will not be changed. Anyone kindly help me. I am struggling for a long time and this is a very critical problem. I did not find any TUXEDO specific group on internet. Thanks in advance For security reasons, shared library searching and resolution is different for setuid programs. Run "man 5 dld.sl" and look for the section titled " Running Setuid Programs" on HP-UX 11I version 2 or later. It says in part : Running Setuid Programs For looking up shared libraries for setuid applications, the dynamic loader uses only the paths listed in /etc/dld.sl.conf. If LD_LIBRARY_PATH and SHLIB_PATH are set, they are validated against the list of paths in /etc/dld.sl.conf. (This allows the individual applications to appropriately order the list from the conf file). You ...

CALL FOR PAPERS -- Special Session on Massively Parallel Processing at the 9th Workshop on Advances in Parallel and Distributed Computational Models -- IPDPS
*** CALL FOR PAPERS *** 2007 International Parallel & Distributed Processing Symposium Workshop on Advances in Parallel and Distributed Computational Models Special Session on Massively Parallel Processing *** Submission Deadline December 4th 2006 *** As part of the Workshop on Advances in Parallel and Distributed Computing Models (APDCM), the aim of this special session is to focus on computer systems that can scale to many thousands of processing elements and are used to solve a single problem. The focus is on identifying new and novel ideas rather than proving incremental advances. By concurrently exploring architecture, programming models, algorithms and applications, the session seeks to advance the state-of-the-art of Massively Parallel Processing (MPP) systems. Following the usual IPDPS practice, all MPP papers will be published in the Proceedings of the IEEE/ACM International Parallel and Distributed Processing Symposium (IPDPS). Topics of Interest: The topics of interest to this special session are: (+) Architectures and Experimental Systems: the use of increased parallelism on a single chip (MPP on a chip), the coordination and communication between massive numbers of processing elements, the use of processors in memory (PIMS), and parallelism in emerging technologies. (+) Large-scale Computing: the utilization of MPPs for large-scale computing, achieving peta-scale levels of processing, use of heterogeneous processing capabilities. (+) Paralleli...

CALL FOR PAPERS -- Special Session on Massively Parallel Processing at the 9th Workshop on Advances in Parallel and Distributed Computational Models -- IPDPS
*** CALL FOR PAPERS *** 2007 International Parallel & Distributed Processing Symposium Workshop on Advances in Parallel and Distributed Computational Models Special Session on Massively Parallel Processing *** Submission Deadline December 4th 2006 *** As part of the Workshop on Advances in Parallel and Distributed Computing Models (APDCM), the aim of this special session is to focus on computer systems that can scale to many thousands of processing elements and are used to solve a single problem. The focus is on identifying new and novel ideas rather than proving incremental advances...

Parallel Processing in SAS
I've heard that there are opportunties to do some parallel processing if your server has multiple CPUs. We have 4 CPUs running on windows 2003 SP1. Anyone have any insight or articles about Parallel processing? I heard only a few procs can actually utilize and take advantage of parallel processing, but wasn't sure. Any insight is appreciated! Jer wrote: > I've heard that there are opportunties to do some parallel processing if > your server has multiple CPUs. I think there are also opportunities for parallel processing with single CPU machines that support threading; on...

Running processes in parallel
Currently, there is a for loop in my program that takes anywhere from a couple of seconds to a couple of minutes. (The procedure runs a step by step animation using the wait function). It's called from an event handler whenever a button in the widget is pushed. I want to be able to stop anytime in the middle of the animation when another button is pressed. (As opposed to the ctrl break method). However, the event handler is unable to process the stop button getting pushed until the for loop procedure is complete, thus defeating the purpose. Is there any way to run two procedures in paralle...

Parallel Processing in IDL
What's an easy way to use multiple processors in IDL? I have a large program but I want to start with a simple program first. Let's say I have an array of a million integers and I have four processors. I want to add all the elements in this array and output it. I could split the array into four parts and give it to each processor right? Is this possible? How would I do this? Thanks! On 10/11/10 10:35 AM, Ammar Yusuf wrote: > What's an easy way to use multiple processors in IDL? I have a large > program but I want to start with a simple program first. Let's say I &g...

parallel processing in forth
http://www.ultratechnology.com/4thpar.html I know jeff fox told me fort can easily do this stuff I forget if there are any tutorials for it? does gforth have some words for this kind fo thing? gavino wrote: > http://www.ultratechnology.com/4thpar.html > > I know jeff fox told me fort can easily do this stuff > > I forget if there are any tutorials for it? > > does gforth have some words for this kind fo thing? There are some articles in Forth Dimensions written by a Russian gentleman, complete with code, that extend Forth with parallel processing techniques and para...

parallel processing #2
Dear all, I have some doubts. Please clarify, 1. Is parallel processing possible in DSP processors and other high level languages like C etc., If so, How?? 2. I am doing image processing on FPGA. I have to write a bitmap file from FPGA output. How can it be done?? 3. How a image can be processed parallely on FPGA. Waiting for replies.. Thanks prash "PrAsHaNtH@IIT" <prashaenator@gmail.com> wrote > > 2. I am doing image processing on FPGA. I have to write a bitmap file > from FPGA output. How can it be done?? > Carefully open up your FPGA. Take a picture ...

COMTI
Hi, does somebody knows is COMTI capable of parallel processing of messages between Windows and MCP. If we assure that programs on MCP side would return messages in asynchronius matter, could I use COMTI for communication from Windows site. Or should I use some other tool from Unisys (or anybody else?) A little bit more detailed description..... WIN MCP request.1 work.1 (will take 10 sec) request.2 work.2 (will take 2 sec) answer.2 answer.1 web.page_return.1 web.page_return.2 problem is also that WIN clienc would most likely be web page. In comp.sys.unisys, (RM) wrote in <ebbtpq$otf$1@ss408.t-com.hr>:: >Hi, >does somebody knows is COMTI capable of parallel processing of messages >between Windows and MCP. >If we assure that programs on MCP side would return messages in asynchronius >matter, could I use COMTI for communication from Windows site. > >Or should I use some other tool from Unisys (or anybody else?) > >A little bit more detailed description..... > >WIN MCP >request.1 > work.1 (will take 10 sec) >request.2 > work.2 (will take...

Vector processing with Lisp
Inspired by the likes of K and with an eye of processing multiple densely packed arrays that form the base of MonetDB I decided to apply Lisp to vector processing. Papers like http://www-db.cs.wisc.edu/cidr/cidr2005/papers/P19.pdf talk at length about optimizing cache usage and processing multiple vectors in the same function. I would love to have optimized Lisp primitives (such as aggregate) to process a list of arrays and return a list of results. I thought I would start small, though. I quickly wrote up the following simple functions and thought I would look at the disassemble dump. I'...

Web resources about - Popular Inexpensive Vector Processing or Parallel Add-in Board? - comp.sys.super

‘Inexpensive’
Al Jigong Billings, regarding my short piece earlier on Woz’s 1977 description of the Apple II: I think @gruber misunderstands “inexpensive” ...

Safe and inexpensive: Government spruiks relocation from Nauru to Cambodia in fact sheet to asylum seekers ...
An Abbott government fact sheet, handed to asylum seekers in the Nauru detention centre and refugees living on Nauru by immigration officials, ...

The fiscal illusion of inexpensive government
The hostile reactions against the government's budget partly reflect existing fiscal arrangements concealing the true costs of public sector ...

Prince William helicopter trackable by inexpensive app, posing terror risk
A British newspaper said it has uncovered a major royal security breach by using an inexpensive app to track Prince William's air rescue ambulance ...

Cheapest Healthy Food: 10 Inexpensive Ways To Eat Well
You may hear people say they can't afford to eat healthy, but as it turns out, some of the healthiest foods in grocery stores won't make a dent ...

Four Inexpensive Ways for Canadian Entrepreneurs to Find Customers Online
Looking to start a business online? More Canadian entrepreneurs are looking for ways to supplement their income or even build a completely new ...

CES 2014: New era of inexpensive 4K monitors from Asus, Lenovo and Seiki ushered in
... , Seiki Continue reading more about Apple , Tech Industry , and Mac Pro at 9to5Mac . What do you think? Discuss "CES 2014: New era of inexpensive ...

Alabama tornadoes: Helmets provide inexpensive storm protection, UAB researchers say
Researchers call on CDC to include helmets in its tornado preparedness advice.

Bringing Inexpensive Mobile Access to Researchers in Antarctica
While cellphone networks have managed to cover large swaths of the six most-populated continents, making a call from Antarctica isn’t so easy. ...

BitFenix Outlaw: When Inexpensive Doesn't Mean Cheap
Introducing the BitFenix Outlaw A lot of our case reviews lately have been for enclosures that hover around the $99 price point and then range ...

Resources last updated: 3/22/2016 6:15:44 PM