Custom Search
Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
hey, quiet often, when i read papers from the VLSI field, the power is one of the big factors taken into account to evaluate the performance of a design. I have never seen a technical report taking into account this factor from the FPGA world designers. They just concentrate on speed and area why that? In article <tBox5.16556$Lu3.301835@east7.usenetserver.com>, "Gary Watson" <gary2@nexsan.com> wrote: > I know it's difficult to predict the power requirements of Xilinx parts, but > what's a safe 2.5V regulator to use for the internal supply of a XC2S150? > The data sheet is most unhelpful in figuring this out. Since I plan to roll > out this product in phases over the next year, I can't say what all my > internal logic might be doing down the road, so I'm happy to over- spec the > regulator to a reasonable degree. > > By the way, I'm getting quoted over 10 UK pounds ($14) for the config prom > for this puppy (XC18V01S20C). Is there a cheaper way to do this? This prom > increases the cost of using a Spartan II by 50%! > > -- > > Gary Watson > gary2@nexsan.com > Nexsan Technologies Ltd. > Derby DE21 7BF ENGLAND > http://www.nexsan.com > > Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25751
If you use the following axioms: 1) An interrupt must not destroy the CPU context. 2) The processor must not hang due to bus error signals (unless double fault) 3) The processor must not ignore bus errors and continue execution. 4) MMU page faults must not hang the CPU nor be ignored. To that you add the following architecture features (which are not patented by ARM) * The interrupt handling in performed in S/W instead of hardware * Use registers to save the return address/PSR The ARM implementation becomes obvious in my opinion. ------- Assume that the interrupt vector gets corrupted and points out in non existing memory. At the interrupt time, the ARM will save some register in internal registers, jump to the bad location and get a bus error. This happens before the CPU has saved the internal registers on the interrupt stack. It is obvious that the CPU cannot save the new pc into the return address register because then axiom 1 is violated (context destroyed). It must take the bus error trap, or violating axiom (2 ,3). The trap needs to store the current PC/PSR somewhere so additional resources are needed. They have basically two alternatives, implement a H/W stack for registers or dedicate registers for the bus error trap, interrupts or whatever. I doubt that selecting one of them should be patentable. -- Best regards, ulf at atmel dot com The contents of this message is intended to be my private opinion and may or may not be shared by my employer Atmel SwedenArticle: 25752
Rick, Instead of specifying "c:\Program Files",try "c:\progra~1". Since Win95/98 etc, is still just a hack on top of DOS, it still conforms to the 8.3 filename system. if you open a DOS shell and type DIR, you'll see that the "expanded" name is "Program Files", but the real name is "Progra~1", or something similar. Microsoft could of made all of lives a h*ll of a lot easier by choosing a default pathname without a space. Almost every command line tool known to Man (and Woman) has problems with spaces in filenames. -kent >rickman wrote: >> The first problem I had was the fact that the install did not seem to >> handle properly the path in the shortcut as I put it under the "Program >> files" directory and it needed quotes around it to be able to >> incorporate the space in the path.Article: 25753
Of course, anyone can look at the printouts and analyze the differences. But what would be nice, as the thread has suggested/concluded, is to have everything in one easy location, i.e., an app note. Why search every nook and cranny of the Xilinx web site looking for every difference when it can be found in one document? -Simon Ramirez, Consultant Synchronous Design, Inc. This is my opinion, necessarily my employer. I also hate spam. "Hal Murray" <murray@pa.dec.com> wrote in message news:8q5v5p$dv8@src-news.pa.dec.com... > > [Slightly sorry for beating a dead horse.] > > If the pintouts were available in raw text, it would be > simple to run diff on the two canidates and see the differences. > > -- > These are my opinions, not necessarily my employers. I hate spam. >Article: 25754
Thought I would comment on the PCB layout of things. I use PADS PCB for all my layout. I have a fancy package with many options (READ TIME SAVERS) and a 10 years worth of pre-made libraries with many of the Xilinx parts already made. What this means is I can turn a board VERY fast provided my client has his/her shit together and is not changing the design on me, has all there part data sheets so I don't have to hunt for them etc. So that said I can turn a decent size PBC rather fast, esp if I already have most of the parts in my library. Options like Cluster Placement, pre-made client specs etc also speed things up. While I don't have the specifics of this PCB Assuming a large Xilinx with a small CPU/DSP some RAM and a few DACs/ADCs and associated op-amps like the client mentioned I could probably turn this PCB (including schematic) in <20 hours. So from that point it is realistic. However not in Protell. Protell is a much slower program as it does not have all the fancies of PADS and it has data base integrity problems so one has to watch what they are doing more. As for Dan's comments about who eats mistakes that is simple. Here is my policy. I give you all net lists to proof.. (Ya'll know how to have a high-lighter party with the engineering team right?) that is your responsibly so a Netlist error or an error because you did not give me proper information you eat. An error that I made (like a library mistake, a CAD violation, manufacturability issue etc) I eat that. So basically whoever made the mistake eats the mistake. By the way I don't believe in the hack it out and get it right the third time approach, I try (and usually do) get it right the first time. I'm more expensive than some but I tend to get it right the first time. Enough promoting me.. I intended to make a point not look for work. Dan wrote: > Not to mention one simple mistake in the PCB schematic that would require a > second turn; not uncommon. Who eats it ? - New photo plot + new PCB + some > new parts + time. >Article: 25755
I believe that using LVDS would reduce system power (the LVPECL pulldown resistors burn a fair amount) and also reduce noise and EMI due to the reduced signal swing (0.35V v.s. ~1V for LVPECL). If the driver slew rate is fixed and the required signal swing is reduced, faster signaling rates (or better timing margins) should be possible. You would need to check Xilinx's driver models and suggested termination schemes to see how significant the differences might be. If it's just a couple of signals, LVPECL might be easiest, but a wide bus could be something else entirely. regards, tom Hal Murray wrote: > > LVDS needs 2.5V on Vcco. LVPECL uses 3.3. > > Suppose you want to get bits from a Virtex-E on one card > through a cable to a Virtex-E on another card and you > control both ends of the design. Is there any reason > to prefer LVDS or LVPECL signaling? > > Is there any reason not to use LVPECL and avioid the extra > power level if you don't have any other need for 2.5V? > > -- > These are my opinions, not necessarily my employers. I hate spam. Tom Burgess -- Digital Engineer Dominion Radio Astrophysical Observatory Penticton, B.C. Canada V2A 6K3Article: 25756
Hey, I just found this cool course for CPLD and FPGA design. It's online and self-paced. It provides an excellent foundation for experienced engineers who want to start their first CPLD or FPGA design. Also, engineers who have done several designs in the past can get valuable knowledge for their current design methodologies. The name of the course is "Introduction to CPLD and FPGA Design," and if it sounds interesting, you can find it on www.chalknet.com. The instructor is an IEEE award winning engineer and sounds like he knows what he is talking about. ShaiArticle: 25757
"Tobias F. Garde" wrote: > > rickman wrote: > > > > I am trying to get the Xilinx Web Pack up and running on my machine and > > am having trouble. > > > > The first problem I had was the fact that the install did not seem to > > handle properly the path in the shortcut as I put it under the "Program > > files" directory and it needed quotes around it to be able to > > incorporate the space in the path. > > > > The next problem I have is trying to access the help. Many of the items > > in the help window cause errors like "Can not find or run the program or > > file 'dkxilinx.hlp'". It also could not find a current version (or any > > for that matter) of "hhcntl.ocx". > > > > Rick, > > The problems may still be related to your - very unorthodox :) - choice > of install path for the Xilinx S/W. Even though the installer accepts a > path with spaces in it, the S/W does not necessarily support it. I had > that problem when installing the 3.1i S/W. > > I recommend that you check if the missing files are actually present on > your system. If this is the case, an uninstall followed by a new install > to the default path may do the trick. BTW, in my case, the uninstall was > a bit difficult, because the uninstall info could not be found due to > the spaces in the path... > > Tobias F. Garde > ASIC Development Engineer > Tellabs Denmark A/S I understand that. In fact I gave up and tried to uninstall the software, but the uninstaller has the same problem with the path! So I just deleted it and installed it in the default directory (which I had to enter since the new default was to the "Program Files" directory! It is simply amazing how little QC there is in this stuff. But the bottom line is that I still could not get the device types to show up. So I gave up and reinstalled Foundation 1.5i from an old job I had done a couple years back. That runs, but it won't let me synthesize my VHDL. So I am back to square one. At least I can syntax check the code. I expect my customer to provide a copy of the current tools at which point I will be able to get Xilinx involved (assuming they even talk to you on the phone anymore). I guess I could try to talk the customer into using the Lucent parts. At least I can get the tools to work in less time than it takes to get the design to work. Don't mind me, I am just frustrated. I'll be ok tomorrow. :-( -- Rick Collins rick.collins@XYarius.com Ignore the reply address. To email me use the above address with the XY removed. Arius - A Signal Processing Solutions Company Specializing in DSP and FPGA design Arius 4 King Ave Frederick, MD 21701-3110 301-682-7772 Voice 301-682-7666 FAX Internet URL http://www.arius.comArticle: 25758
That includes the Windows tools themselves. One of the problems I found was that the install script entered the shortcut command line with the spaces. Because they did not enclose the path in quotes, the parser quit at the space and could not find the program. "K. Orthner" wrote: > > Rick, > > Instead of specifying "c:\Program Files",try "c:\progra~1". > > Since Win95/98 etc, is still just a hack on top of DOS, it still conforms > to the 8.3 filename system. if you open a DOS shell and type DIR, you'll > see that the "expanded" name is "Program Files", but the real name is > "Progra~1", or something similar. > > Microsoft could of made all of lives a h*ll of a lot easier by choosing a > default pathname without a space. Almost every command line tool known to > Man (and Woman) has problems with spaces in filenames. > > -kent > > >rickman wrote: > >> The first problem I had was the fact that the install did not seem to > >> handle properly the path in the shortcut as I put it under the "Program > >> files" directory and it needed quotes around it to be able to > >> incorporate the space in the path. -- Rick Collins rick.collins@XYarius.com Ignore the reply address. To email me use the above address with the XY removed. Arius - A Signal Processing Solutions Company Specializing in DSP and FPGA design Arius 4 King Ave Frederick, MD 21701-3110 301-682-7772 Voice 301-682-7666 FAX Internet URL http://www.arius.comArticle: 25759
I suspect that is somewhat a cyclic dependancy. No one can get any real data on power consumption until the circuit is built (if then). So there are not many designers who use that as a metric. Since it does not show up as a metric, no one writes software to evaluate it. Of course there are some crude tools that will let you enter your circuit parameters to estimate the power consumption, but none tie into the design directly. erika_uk@my-deja.com wrote: > > hey, > > quiet often, when i read papers from the VLSI field, the power is one > of the big factors taken into account to evaluate the performance of a > design. > > I have never seen a technical report taking into account this factor > from the FPGA world designers. They just concentrate on speed and area > > why that? > > In article <tBox5.16556$Lu3.301835@east7.usenetserver.com>, > "Gary Watson" <gary2@nexsan.com> wrote: > > I know it's difficult to predict the power requirements of Xilinx > parts, but > > what's a safe 2.5V regulator to use for the internal supply of a > XC2S150? > > The data sheet is most unhelpful in figuring this out. Since I plan > to roll > > out this product in phases over the next year, I can't say what all my > > internal logic might be doing down the road, so I'm happy to over- > spec the > > regulator to a reasonable degree. > > > > By the way, I'm getting quoted over 10 UK pounds ($14) for the config > prom > > for this puppy (XC18V01S20C). Is there a cheaper way to do this? > This prom > > increases the cost of using a Spartan II by 50%! > > > > -- > > > > Gary Watson > > gary2@nexsan.com > > Nexsan Technologies Ltd. > > Derby DE21 7BF ENGLAND > > http://www.nexsan.com > > > > > > Sent via Deja.com http://www.deja.com/ > Before you buy. -- Rick Collins rick.collins@XYarius.com Ignore the reply address. To email me use the above address with the XY removed. Arius - A Signal Processing Solutions Company Specializing in DSP and FPGA design Arius 4 King Ave Frederick, MD 21701-3110 301-682-7772 Voice 301-682-7666 FAX Internet URL http://www.arius.comArticle: 25760
Exactly. Now add the effort to needed to get"his/her shit together", collecting the data sheets, and preparing the package for you, and yellow-lining the netlist when you are done. Now you are over the 30-40 hours and you haven't accounted for even 1 hour of design time. Throw in on top of that any board layout issues for noise, high speed clocks, decoupling etc, and I think for all but the simplest boards you are going to be pushing the 30-40 hours yourself. Don't forget the effort to generate all the gerbers, interfacing to a fab house etc. Hawker wrote: > > Thought I would comment on the PCB layout of things. > I use PADS PCB for all my layout. I have a fancy package with many options > (READ TIME SAVERS) and a 10 years worth of pre-made libraries with many of the > Xilinx parts already made. What this means is I can turn a board VERY fast > provided > my client has his/her shit together and is not changing the design on me, has > all there > part data sheets so I don't have to hunt for them etc. > > So that said I can turn a decent size PBC rather fast, esp if I already have > most > of the parts in my library. Options like Cluster Placement, pre-made client > specs etc also speed things up. While I don't have the specifics of this PCB > Assuming a large Xilinx with a small CPU/DSP some RAM and a few DACs/ADCs and > associated > op-amps like the client mentioned I could probably turn this PCB (including > schematic) in <20 hours. > So from that point it is realistic. > > -- -Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email ray@andraka.com http://www.andraka.com or http://www.fpga-guru.comArticle: 25761
In my experience it takes an incredibly well defined project to even consider a fixed price contract. And without some knowledge of the market and the company itself the revenue off of sales is a crap shoot. Doesn't seem to make a lot of sense with the money that can be made elsewhere in this job market but maybe there's more details that we don't know about. - Arthur Aaron In article <39C6C3B8.5F7F1360@andraka.com>, Ray Andraka <ray@andraka.com> wrote: > There ain't much info here as to what is in the FPGA, or even how big a device > it is. The FPGA design and verification alone could take as much as 10X or more > of his estimated 30-40hours. I think the board layout alone, given a complete > and verified netlist could be more than the 30-40 hours (check with a layout > house if you don't believe me). For the pricing, I am always leery of projected > volumes, especially if such a large percentage of the income is to be based on > it. If the company tanks, or he only manages to sell one or tow boards or if > the project gets scrapped, the consultant/contractor is working for about $50/hr > (assuming the hours are met), which is quite low (I don't think you can get warm > bodies to fill seats on your location for that anymore, and if you did you'd > spend more than the 40 hours just getting them up to speed) > > That said, the DDS function could easily be incorporated into the FPGA. > > "S. Ramirez" wrote: > > > > Does anyone in this newsgroup realistically think that they can do the > > work below in 30-40 hours? Of course I would have to inquire to find out > > the details in order to make an estimate of what is really required. But > > just reading what this company wants, I see four main tasks: 1) design and > > schematic capture of the board, 2) design and HDL/schematic capture of the > > FPGA, 3) simulation of maybe the board and certainly the FPGA, and 4) lab > > testing/debug. I personally do not think that I can do all of the above > > tasks in 30-40 hours. > > At least they have a detailed block diagram. > > How about yous guys/gals? Can anyone do this work based on what's > > written below? > > -Simon Ramirez, Consultant > > -Synchronous Design, Inc. > > > > "Walt" <walt_white@southwestsoftware.com> wrote in message > > news:v98x5.31686$I6.182833@news1.rdc1.az.home.com... > > > Hello, > > > > > > We are looking for a freelance designer who can help us develop a > > > daughterboard module. This work will involve the schematic capture and > > > layout of the PCB in (preferably in Protel), and the programming of the > > > SpartonII FPGA. The hardware design will be based on a detailed block > > > diagram of the board that we provide, along with some key part numbers and > > > other needed information. The board includes some A/Ds, D/As and a DDS. > > > The work should take between 30 to 40 hrs and can be done at your > > location. > > > > > > The project deliverables will be Protel design files, the FPGA design > > files, > > > and any documentation used to create the design. All rights and claims to > > > any work on this project at all times is unequivocally the property and > > > intellectual property in origin and by any extension with all rights and > > > privileges reserved to Southwest Software & Systems LLC (3S). > > > > > > Payment for the work will be a flat fee of $2000.00 plus a royalty of 5.0% > > > of the sale price for each board, and the boards will sell about $675ea > > and > > > volume is expected to be a few hundred (For example, assuming a 200 board > > > volume, the total payment for the hardware design would be: $2000.00 + > > 0.05 > > > x $675.00/Board x 200Boards = $8,750). Additionally, there could be > > similar > > > follow-up designs. > > > > > > If you are interested and qualified to do this work, please contact me as > > > soon as possible, we are ready to begin immediately. > > > > > > Regards, > > > > > > Walt White > > > Southwest Software & Systems LLC > > > www.southwestsoftware.com > > > > > > > > > > > > > > -- > -Ray Andraka, P.E. > President, the Andraka Consulting Group, Inc. > 401/884-7930 Fax 401/884-7950 > email ray@andraka.com > http://www.andraka.com or http://www.fpga-guru.com > Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25762
hi, can F2.1i generate the schematic from vhdl (or edif) entry. If no, is there any tool which does so Regards --Erika Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25763
sorry, i wanted to say edif . In article <8q8lqb$iu3$1@nnrp1.deja.com>, erika_uk@my-deja.com wrote: > hi, > > can F2.1i generate the schematic from vhdl (or edif) entry. > If no, is there any tool which does so > > Regards > > --Erika > > Sent via Deja.com http://www.deja.com/ > Before you buy. > Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25764
I'm doing GPS design using xilinx board. After modulating C/A code with navigation data using BPSK technique, I tried to mix the signal with L1 carrier signal. But in the spectrum analyzer, I observed a big spike in the center of L1 frequency. Does anybody has some idea how can I get rid of it? Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25765
This is a multi-part message in MIME format. --------------82C4CA1B62309F0FBF295C33 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Andy, I just wanted to update you on this FPGA Express bug. We already filed a CR (Change Request) and Synopsys is looking into this issue. If a Spartan/ 4K design is using both rising and falling edge clocks, the clock must be inverted using CLB clock muxes CLKX or CLKY. In IOBs, it is supposed to use OKMUX. Unfortunately FPGA Express isn't using these muxes, its using a LUT to invert the clock. A workaround would be to instantiate negative edge flops like FDC_1, FDCE_1 etc. This will be fixed in the future versions of the software. Vikram Pasham Xilinx Applications Andy Peters wrote: > File this under "bug in v3.4." Or file it under "FPGA Express is too > stupid to live." Or file it under "Why haven't they figured out the > goddamn architecture yet?" > > Here's the scenario. I'm doing a little experiment. Target > architecture is Spartan and 4KXLA; the problem is the same for both. > Most of my flops are clocked on the rising edge. One of the things I > did was to have a handful of flops clocked on the falling edge. > Normally, placing a period constraint on the clock is all that's > required; map and p&r should go, "Ah! two-phase clock" and do the right > thing. > > p&r runs, and the report at the end complains that there's skew in the > design due to something not using a global clock net, and the skew > wasn't analyzed. OK, run trace with the skew option. Ooops -- there's > a bunch of flops with like 4.8 ns clock skew (100 MHz period constraint) > and clocked from N1964_BUFGed. Weird. > > Poking around with FPGA Editor (which is really useful when you're > trying to figure out just what the hell the tools did), turns out that > N1964_BUFGed is the clock net for the falling-edge clocked flops. Right > then, I knew exactly what the problem was: > > FPGA Express 3.4 is too stupid to realize that the CLB HAS A > CLOCK-POLARITY MUX RIGHT IN FRONT OF THE FLOP SO IT DOESN'T NEED TO USE > A CLB TO INVERT THE CLOCK! At least it knew enough to route the output > of the CLB inverter to a BUFGLS, but it's still WRONG. > > This is a bug that needs to be fixed PRONTO. > > I am going to open a support case at Xilinx just as soon as their web > site decides it wants come back to life. > > -- a > ---------------------------- > Andy Peters > Sr. Electrical Engineer > National Optical Astronomy Observatory > 950 N Cherry Ave > Tucson, AZ 85719 > apeters (at) n o a o [dot] e d u --------------82C4CA1B62309F0FBF295C33 Content-Type: text/x-vcard; charset=us-ascii; name="Vikram.Pasham.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for Vikram Pasham Content-Disposition: attachment; filename="Vikram.Pasham.vcf" begin:vcard n:Pasham;Vikram x-mozilla-html:FALSE org:Xilinx adr:;;2100 Logic Drive;San Jose;CA;95124;USA version:2.1 email;internet:Vikram.Pasham@xilinx.com title:Applications Engineer fn:Vikram Pasham end:vcard --------------82C4CA1B62309F0FBF295C33--Article: 25766
I'm doing GPS design using xilinx board. After modulating C/A code with navigation data using BPSK technique, I tried to mix the signal with L1 carrier signal. But in the spectrum analyzer, I observed a big spike in the center of L1 frequency. Does anybody has some idea how I can get rid of it? Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25767
This is a multi-part message in MIME format. --------------9B630FF35197B534808B72AD Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Jamil, Check out the Xilinx "Bluetooth" webpage at http://support.xilinx.com/products/virtex/wireless/bluetooth.htm Generally FPGAs are used for implementing some of the Bluetooth components like Memory controllers, DSP Filters, Peripherals ( PCI, Co-processor, Protocol processor) etc. You can also find detailed Bluetooth specs at www.bluetooth.com Hope this helps. Vikram Xilinx Applications khatib@ieee.org wrote: > Hi, > Me and some young engineers (students and non-students) like get more > practice in digital design by doing a real design. > We found that the bluetooth technology is a new technology and may have > good future even in a home made systems. > > Could you please help us on the following questions: > 1. Is there any avialable cores for this technology that we can learn > from? > 2. how large is this core going to be? > 3. where can I find information about implementing it? > 4. can a group of 3 young engineers work on it? > 5. Is ther any possibility to implement a device at home with this core? > 6. can we implement it on FPGA? > 7. Does it worth the time or should we look for another technology to > learn from? > > Thanks in advance > Jamil Khatib > > Sent via Deja.com http://www.deja.com/ > Before you buy. --------------9B630FF35197B534808B72AD Content-Type: text/x-vcard; charset=us-ascii; name="Vikram.Pasham.vcf" Content-Transfer-Encoding: 7bit Content-Description: Card for Vikram Pasham Content-Disposition: attachment; filename="Vikram.Pasham.vcf" begin:vcard n:Pasham;Vikram x-mozilla-html:FALSE org:Xilinx adr:;;2100 Logic Drive;San Jose;CA;95124;USA version:2.1 email;internet:Vikram.Pasham@xilinx.com title:Applications Engineer fn:Vikram Pasham end:vcard --------------9B630FF35197B534808B72AD--Article: 25768
There was some speculation about the Virtex shape. Here are the facts: Physically, most Xilinx chips are close to square, a shape that is desirable for fitting them into a package. Physically, the Virtex CLB has a rectangular shape, it is roughly 50% higher than it is wide. That's why there are 50% more columns than rows. ( Assuming the conventional view of vertical columns and horizontal rows. ) Each BlockRAM has a height of four CLBs and a width of about 2.5 CLBs. So the BlockRAM area is equivalent to the area of about 10 CLBs. If you do the math, the chip size ends up almost square, but XCV405E and XCV812E with their huge number of BlockRAMs are oblong. Remember also that a Virtex CLB has four LUTs and flip-flops, while Spartan, XC4000 and XC3000 have only two LUTs in their CLB. Just to satisfy the curiosity... Peter Alfke, Xilinx ApplicationsArticle: 25769
> What excactly do the offset = in and offset = out mean. Do they mean > that a signal must be valid X ns BEFORE the clock signal arives, and > that a signal is Y seconds AFTER the clock has arived stable? This could be useful: http://support.xilinx.com/support/techsup/journals/timing/presentation/timin g/noframe/index.htm Regards, LCArticle: 25770
It is very difficult to beat the performance of the dedicated fast carry chains in Altera and Xilinx FPGAs using any of the conventional fast carry schemes. The fast carry chains have dedicated logic and connections that are roughly an order of magnitude faster than the general interconnect and logic resources. Additionally, by using the fast carry chains in xilinx you are getting the carry function without using an additional LUT, so the adder is half the size of an otherwise identical ripple carry adder using the LUTs instead of the fast carry. More area also generally means slower because the connections have to be made over longer distances. When you get into very wide adders (more than 30 bits or so), then you can start improving on the performance by using the fast carry schemes to combine segments made up of smaller ripple carry adders. So for Altera and Xilinx FPGAs the built in ripple carry function is the smallest and pretty much the fastest. In devices such as the Atmel parts, the story is different. I was part of a group that looked at adders in the Atmel6K and NSC Clay31 several years back. The conclusion there was that for small adders, the ripple carry was still better because of the smaller area and tighter (read faster) routing. As the number of bits increased, If memroy serves me correctly, 8 bits was about where ultimate speed made looking at other carry schemes attractive. Charle' Rupp had a scheme he called arithmetic progression that was pretty compact. It consisted of progressively longer ripple adders tied together in a carry scheme. Actel FPGAs fare reasonably well with a carry select architecture because it maps efficiently into their logic element. shahzad2512@my-deja.com wrote: > > Different types of adders are available that offer different > advantages. These adders are: > Ripple Carry Adders > Carry lookahead adders > Carry Select adders > Conditional Sum adders > These adders when implemented in FPGAs deliver different performance in > terms of speed and area. Both these things are critical for FPGAs. I > implemented a Logiblox 4 bit adder and then implemented CLA adder. CLA > was equivalent in resources to the logiblox and was faster. But bigger > CLA consumes more resources than the Logiblox adder. Adders are the > basic units that are extensively used in many designs. > Has anyone done a thorough research on which adder is the best for say > Xilinx FPGA, Altera, Atmel FPGAs. If someone has done a thorough work > on Adders for say, Altera FPGAS, it will be worth sharing the > experience. If someone knows that this topic has been discussed it will > be helpful to mention the relevant site. > Regards, > SHAHZAD > > Sent via Deja.com http://www.deja.com/ > Before you buy. -- -Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email ray@andraka.com http://www.andraka.com or http://www.fpga-guru.comArticle: 25771
Not to mention that you would have to trust this company to tell you their sales and pay you accordingly. I wouldn't trust a company unless I knew the principles personally. I've had to police companies in the past, and it isn't fun. Besides, policing adds to the 30-40 hours that they defined. -Simon Ramirez, Consultant Synchronous Design, Inc. <artaaron@my-deja.com> wrote in message news:8q8ktr$hs1$1@nnrp1.deja.com... > In my experience it takes an incredibly well defined project > to even consider a fixed price contract. And without some > knowledge of the market and the company itself the revenue off > of sales is a crap shoot. Doesn't seem to make a lot of sense > with the money that can be made elsewhere in this job market but > maybe there's more details that we don't know about. > > - Arthur AaronArticle: 25772
Thanks for the feedback everyone.
I did a little bit of digging around, and concluded that I'll have to
pay real money for a real FPGA board. The XESS XS-40 board did a very
respectable job these past two months, but my current project requires
more gates.
I checked out the various 'third party vendors' listed under Xilinx's
webstore (click on "prototype platforms".) After looking at all the
sites, I've narrowed down my search to two contendors :
1) AVNET's Xilinx Development System (XCV300-432 FPGA on PCI board)
prototype board can be used standalone (not plugged into PCI slot)
$2495 (with Xilinx Foundation Express)
2) XESS Virtex-300 (XCV300-240 FPGA on prototype board)
$799 (no software, so I'd need to buy Foundation Express
elsewhere)
Either way, I need Xilinx's software tool (Foundation Express), so
factoring this as an additional cost to #(2), the total cost of either
choice comes out the same.
(Don't laugh, but these past two months, I 'scraped by' with the Xilinx
Student Edition 1.5 software. This was only a temporary situation until
I verified FPGA-board was viable for prototyping some work I was doing.)
Article: 25773You may want to check for a presense of combinatorial loops, which could be a cause of what you are seeing. -- Yury Wolf-Sonkin / RealTime Data LLC / In article <39C29203.2587E967@xilinx.com>, Mujtaba Hamid <mujtaba.hamid@xilinx.com> wrote: > Your design is going into an infinite loop at that point. You can go to > Options > Simulation Options in MTI and increase the iteration limit. But > even if hte default iteration limit is reached (5000), it probably means that > there is an infinite loop at that time in simulation. > > What you might wanna do is to run simulation for maybe 1 ns less than at the > time when this error happens. Then step through the simulator and see which > line in the source code MTI hangs at. > > Mujtaba > > Tomasz Brychcy wrote: > > > Hello, > > > > Simulation of model before synthesis is correct. When i simulate the same > > model but after synthesis during simulation in MTI 5.2c occur the problem: > > > > iteration limit reached. Possilbe zero delay oscillation. > > > > I do not know what do? > > > > Please about help > > > > I attach files: > > > > file before synthesis: controllogic(synth).v > > file after synthesis: controllogic.v > > testbench: tb_controllogic.v > > > > P.S I know that the problem occur exactly when signal ReadCn goes to high > > state. > > > > Tomek > > > > -- > > Department of Electrical Metrology > > Technical University of Zielona Gora > > > > T.Brychcy@ime.pz.zgora.pl > > Sent via Deja.com http://www.deja.com/ Before you buy.Article: 25774
My experience envolves a design based on PCI 32/33 on Spartan family.
You have to be careful sizing your device. For example, when Xilinx
states that PCI 32/33 should take up only 55% of Spartan 30 XL, what
they really mean that in order to get PCI Master/Slave functional you
need to add some 15%-20% of additional logic on top of the core. That
required logic is extensively described in PCI Logicore Implementation
Guide. Obviously, Xilinx's marketing team minimizes the device
utilization of PCI logicore to make the product more attractive.
-- Yury
In article <39C1EB9F.196D852E@ti.uni-mannheim.de>,
Lars Rzymianowicz <larsrzy@ti.uni-mannheim.de> wrote:
> Marc Reinert wrote:
> > I'm going to develop a PCI board (33Mhz/32Bit) on one of these
devices.
> > My idea is to have a lot of space left in my FPGA beside the
> > PCI-Interface.
> > Is there a core (or better a complete VHDL-Programming) aviable from
> > Xilinx (or anybody else)? Does it work well and are any features of
the
> > PCI-Bus not supported? Is there a free version aviable on the
internet?
>
> Hi Marc,
>
> Xilinx has the logiCORE PCI macrocell, which is a customizable
> 33/66/32/64 PCI interface. Check the Xilinx website for it, it's
> price is around $5000, i guess.
> Since you are a University, you might consider to join the Xilinx
> Univ. Prg. XUP. There you can also apply for a donation of the core.
> I tried that, but never got any reply on several emails/faxes.
> But maybe you have more luck ;-)
>
> Lars
> --
> Address: University of Mannheim; B6, 26; 68159 Mannheim, Germany
> Tel: +(49) 621 181-2716, Fax: -2713
> email: larsrzy@{ti.uni-mannheim.de, atoll-net.de, computer.org}
> Homepage: http://mufasa.informatik.uni-mannheim.de/lsra/persons/lars/
>
Sent via Deja.com http://www.deja.com/
Before you buy.
Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z