From: "mac teh knife" Newsgroups: comp.arch.fpga Subject: new computer Date: Sun, 16 Jun 2002 20:52:43 -0700 Lines: 22 X-Priority: 3 X-MSMail-Priority: Normal X-Newsreader: Microsoft Outlook Express 6.00.2600.0000 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 NNTP-Posting-Host: rgl-c-66-168-87-25.chartertn.net Message-ID: <3d0d331a_3@news.chartertn.net> X-Trace: 16 Jun 2002 20:53:46 -0400, rgl-c-66-168-87-25.chartertn.net Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.icl.net!newsfeed1.cidera.com!Cidera!news.chartertn.net!rgl-c-66-168-87-25.chartertn.net Xref: chonsp.franklin.ch comp.arch.fpga:18486 We have started development using the new fpgas. We are evaluating the virtex 2 and stratix devices. What I'm finding out is my 2 year old machine ain't got what it takes to crunch the files that can fill up these multi million gate chips. I was wondering if anybody would care to share with us the machine (PC) they are using. Also has anybody evaluated linux vs window performance as far as fpga applications are concerned. I'm looking at a P4 2.4ghz / 2 gig ram and 533 mhz front bus. I'm also looking at a Xeon 2.4 g with 400 mhz front bus. Does anybody know if there is a performance difference between these two type of processors? Mac the knife is of course not my real name, I'm just so tired of spam that posting to use net generates. thanks Jerry ###### From: "Pete" Newsgroups: comp.arch.fpga References: <3d0d331a_3@news.chartertn.net> Subject: Re: new computer Lines: 34 X-Priority: 3 X-MSMail-Priority: Normal X-Newsreader: Microsoft Outlook Express 6.00.2600.0000 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 Message-ID: Date: Tue, 18 Jun 2002 14:36:41 +0100 NNTP-Posting-Host: 62.252.176.68 X-Complaints-To: abuse@ntlworld.com X-Trace: newsfep1-win.server.ntli.net 1024407394 62.252.176.68 (Tue, 18 Jun 2002 14:36:34 BST) NNTP-Posting-Date: Tue, 18 Jun 2002 14:36:34 BST Organization: ntlworld News Service Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.stueberl.de!newspeer1-gui.server.ntli.net!ntli.net!newsfep1-win.server.ntli.net.POSTED!53ab2750!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18512 As a 'softie', I was dismayed to find that I needed to buy a 1.4GHz athlon with 1GB DDR ram just to be able to compile a design with handel-C, and then be able to do a PAR in less than one working day. this was for a XCV2000e which I managed to fill. This as when 1.4Ghz athlons were the hottest chip on the market. Pete "mac teh knife" wrote in message news:3d0d331a_3@news.chartertn.net... > We have started development using the new fpgas. We are evaluating the > virtex 2 and stratix devices. > What I'm finding out is my 2 year old machine ain't got what it takes to > crunch the files that can fill up > these multi million gate chips. I was wondering if anybody would care to > share with us the machine (PC) > they are using. Also has anybody evaluated linux vs window performance as > far as fpga applications are concerned. > > I'm looking at a P4 2.4ghz / 2 gig ram and 533 mhz front bus. I'm also > looking at a Xeon 2.4 g with 400 mhz > front bus. Does anybody know if there is a performance difference between > these two type of processors? > > Mac the knife is of course not my real name, I'm just so tired of spam that > posting to use net generates. > > thanks > Jerry > > > ###### Message-ID: <3D10286B.83A006FA@pac.net> From: Mark X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en MIME-Version: 1.0 Newsgroups: comp.arch.fpga Subject: Re: new computer References: <3d0d331a_3@news.chartertn.net> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Lines: 67 NNTP-Posting-Host: 207.214.212.160 X-Complaints-To: abuse@prodigy.net X-Trace: newssvr14.news.prodigy.com 1024468938 ST000 207.214.212.160 (Wed, 19 Jun 2002 02:42:18 EDT) NNTP-Posting-Date: Wed, 19 Jun 2002 02:42:18 EDT Organization: Prodigy Internet http://www.prodigy.com X-UserInfo1: Q[RGWXCEGZT]SWH]^JKBOW@@YJ_ZTB\MV@BNMRQIMASJETAANVW[AKWZE\]^XQWIGNE_[EBL@^_\^JOCQ^RSNVLGTFTKHTXHHP[NB\_C@\SD@EP_[KCXX__AGDDEKGFNB\ZOKLRNCY_CGG[RHT_UN@C_BSY\G__IJIX_PLSA[CCFAULEY\FL\VLGANTQQ]FN Date: Wed, 19 Jun 2002 06:42:19 GMT Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.imp.ch!news.imp.ch!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!newsfeed.cwix.com!prodigy.com!newsmst01.news.prodigy.com!prodigy.com!postmaster.news.prodigy.com!newssvr14.news.prodigy.com.POSTED!8a85c063!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18577 Hello, From a single recent very unofficial PC performance comparison .... Target: Virtex II XC2V6000 Xilinx: ngdbuild, map, par, trce using Design Manager. 4.1i, sp3 PC A: 650 MHz Pentium III, 1GB PC100 Execution time: ~1 hr: 30 min (+/- 5 min) PC B: 2.4 GHz Pentium IV, 2 GB PC2100 Execution time: 40 min Caveate: This is only one comparison of a single run on each computer. Most of the time (A: 1hr, 11 min B: 35 min) was spent in par, where the CPU utilization was 100%, according to NT 4.0, sp6 task manager. par execution time ratio: 71/35 ~= 2.03. Memory speed ratio: 266/100 ~= 2.66. CPU speed ratio: 2400/650 ~= 3.69. Although CPU utilization was 100%, the execution time ratio seems to imply that memory bandwidth was the limiting factor. But, the CPU was not <100%, so it's unclear why the execution time was not shorter. (IOW, I don't know what else the CPU was doing.) Screen saver, virus scanner, email, etc. were off. I've heard several times that FPGA tools are CPU intensive, but I think that machine specification should consider factors in addition to CPU speed, e.g., memory bandwidth, size of memory (to hold database), and M/B/chipset. We also considered a dual-CPU M/B. Unfortunately, no tools, that we have, can take advantage of multiprocessor PCs. A multiprocessor PC, in our case, would allow us to do synthesis/simulation, while the Xilinx tools are running. (I've tried to script the flow, but cannot since we're using ChipScope.) I believe that Xilinx on Linux needs Wine (from Xilinx installation notes), which I "heard" elsewhere ends up being slightly slower than native code. For front-end tools, we use the Mentor tools, which I believe have been ported to Linux, but, I haven't tried any, yet.... (I've been hoping ISD Mag will do another "annual" comparison.) Microsoft and Linux seem close as fas as speed, but, Linux seems to be ahead wrt stability. I haven't used Altera since the Flex 10K and MaxPlusII, so I don't have any experience with the latest Altera FPGAs. I also have no experience with the Xeon CPUs. Hope this helps, Mark mac teh knife wrote: > We have started development using the new fpgas. We are evaluating the > virtex 2 and stratix devices. > What I'm finding out is my 2 year old machine ain't got what it takes to > crunch the files that can fill up > these multi million gate chips. I was wondering if anybody would care to > share with us the machine (PC) > they are using. Also has anybody evaluated linux vs window performance as > far as fpga applications are concerned. > > I'm looking at a P4 2.4ghz / 2 gig ram and 533 mhz front bus. I'm also > looking at a Xeon 2.4 g with 400 mhz > front bus. Does anybody know if there is a performance difference between > these two type of processors? > > Mac the knife is of course not my real name, I'm just so tired of spam that > posting to use net generates. > > thanks > Jerry ###### From: rickman Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Wed, 19 Jun 2002 13:06:59 -0400 Organization: Arius, Inc Lines: 89 Message-ID: <3D10BA33.BBB6983B@yahoo.com> References: <3d0d331a_3@news.chartertn.net> <3D10286B.83A006FA@pac.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Trace: UmFuZG9tSVZohSNDCJ16g3ngWsxR5irq+1ib0/s3dToS2oTynGYOZpPnVjHYKRPY X-Complaints-To: abuse@rcn.com NNTP-Posting-Date: 19 Jun 2002 17:07:00 GMT X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.icl.net!newsfeed.icl.net!netnews.com!feed2.news.rcn.net!feed1.news.rcn.net!rcn!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18587 I am looking at a new machine as well. I would like to get an idea of what a faster P3 would do in your tests. Any chance you can run your PAR on a faster P3 or even a laptop? I believe they had P3s running at 1.3 GHz before they released the P4. It would be very instructive to see how that compares to the current P4s. Mark wrote: > > Hello, > > From a single recent very unofficial PC performance comparison .... > > Target: Virtex II XC2V6000 > Xilinx: ngdbuild, map, par, trce using Design Manager. 4.1i, sp3 > > PC A: 650 MHz Pentium III, 1GB PC100 > Execution time: ~1 hr: 30 min (+/- 5 min) > > PC B: 2.4 GHz Pentium IV, 2 GB PC2100 > Execution time: 40 min > > Caveate: This is only one comparison of a single run on each computer. > Most of the time (A: 1hr, 11 min B: 35 min) was spent in par, where the CPU > utilization was 100%, according to NT 4.0, sp6 task manager. par execution > time ratio: 71/35 ~= 2.03. Memory speed ratio: 266/100 ~= 2.66. CPU speed > ratio: 2400/650 ~= 3.69. Although CPU utilization was 100%, the execution time > ratio seems to imply that memory bandwidth was the limiting factor. But, the > CPU was not <100%, so it's unclear why the execution time was not shorter. > (IOW, I don't know what else the CPU was doing.) Screen saver, virus scanner, > email, etc. were off. > > I've heard several times that FPGA tools are CPU intensive, but I think that > machine specification should consider factors in addition to CPU speed, e.g., > memory bandwidth, size of memory (to hold database), and M/B/chipset. We also > considered a dual-CPU M/B. Unfortunately, no tools, that we have, can take > advantage of multiprocessor PCs. A multiprocessor PC, in our case, would allow > us to do synthesis/simulation, while the Xilinx tools are running. (I've tried > to script the flow, but cannot since we're using ChipScope.) > > I believe that Xilinx on Linux needs Wine (from Xilinx installation notes), > which I "heard" elsewhere ends up being slightly slower than native code. For > front-end tools, we use the Mentor tools, which I believe have been ported to > Linux, but, I haven't tried any, yet.... (I've been hoping ISD Mag will do > another "annual" comparison.) Microsoft and Linux seem close as fas as speed, > but, Linux seems to be ahead wrt stability. > > I haven't used Altera since the Flex 10K and MaxPlusII, so I don't have any > experience with the latest Altera FPGAs. I also have no experience with the > Xeon CPUs. > > Hope this helps, > Mark > > mac teh knife wrote: > > > We have started development using the new fpgas. We are evaluating the > > virtex 2 and stratix devices. > > What I'm finding out is my 2 year old machine ain't got what it takes to > > crunch the files that can fill up > > these multi million gate chips. I was wondering if anybody would care to > > share with us the machine (PC) > > they are using. Also has anybody evaluated linux vs window performance as > > far as fpga applications are concerned. > > > > I'm looking at a P4 2.4ghz / 2 gig ram and 533 mhz front bus. I'm also > > looking at a Xeon 2.4 g with 400 mhz > > front bus. Does anybody know if there is a performance difference between > > these two type of processors? > > > > Mac the knife is of course not my real name, I'm just so tired of spam that > > posting to use net generates. > > > > thanks > > Jerry -- Rick "rickman" 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 URL http://www.arius.com 4 King Ave 301-682-7772 Voice Frederick, MD 21701-3110 301-682-7666 FAX ###### From: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Wed, 19 Jun 2002 17:15:45 +0000 (UTC) Organization: Unknown Lines: 14 Message-ID: References: <3d0d331a_3@news.chartertn.net> <3D10286B.83A006FA@pac.net> <3D10BA33.BBB6983B@yahoo.com> NNTP-Posting-Host: soda.csua.berkeley.edu X-Trace: agate.berkeley.edu 1024506945 43095 128.32.247.226 (19 Jun 2002 17:15:45 GMT) X-Complaints-To: usenet@agate.berkeley.edu NNTP-Posting-Date: Wed, 19 Jun 2002 17:15:45 +0000 (UTC) Originator: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!cyclone.bc.net!news-hog.berkeley.edu!ucberkeley!agate.berkeley.edu!agate!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18606 In article <3D10BA33.BBB6983B@yahoo.com>, rickman wrote: >I am looking at a new machine as well. I would like to get an idea of >what a faster P3 would do in your tests. Any chance you can run your >PAR on a faster P3 or even a laptop? I believe they had P3s running at >1.3 GHz before they released the P4. It would be very instructive to >see how that compares to the current P4s. I'd definatly look at the Athlons as well. Although I don't have the numbers to back them up, my intuition is that the tools are memory and cache bound as much as CPU bound, and the athlons have a much better cache heirarchy. -- Nicholas C. Weaver nweaver@cs.berkeley.edu ###### Message-ID: <3D117A60.223F7756@pac.net> From: Mark X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en MIME-Version: 1.0 Newsgroups: comp.arch.fpga Subject: Re: new computer References: <3d0d331a_3@news.chartertn.net> <3D10286B.83A006FA@pac.net> <3D10BA33.BBB6983B@yahoo.com> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Lines: 109 NNTP-Posting-Host: 207.214.211.51 X-Complaints-To: abuse@prodigy.net X-Trace: newssvr13.news.prodigy.com 1024555457 ST000 207.214.211.51 (Thu, 20 Jun 2002 02:44:17 EDT) NNTP-Posting-Date: Thu, 20 Jun 2002 02:44:17 EDT Organization: Prodigy Internet http://www.prodigy.com X-UserInfo1: O@ZQBWKDBZVS@^LYMRKNOPDA[X_LPO@FKY\@LWQHBATBTSUBYFWEAE[YJLYPIWKHTFCMZKVMB^[Z^DOBRVVMOSPFHNSYXVDIE@X\BUC@GTSX@DL^GKFFHQCCE\G[JJBMYDYIJCZM@AY]GNGPJD]YNNW\GSX^GSCKHA[]@CCB\[@LATPD\L@J\\PF]VR[QPJN Date: Thu, 20 Jun 2002 06:44:17 GMT Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!newsfeed.cwix.com!prodigy.com!newsmst01.news.prodigy.com!prodigy.com!postmaster.news.prodigy.com!newssvr13.news.prodigy.com.POSTED!8a85c063!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18622 Unfortunately, the only other faster P3's we have are dual 800 MHz, which are assigned to the CAD/EDA group. Nicholas Weaver mentioned the Athlons. We bought a K7 1000 PC with 1GB PC133 about two years ago. A coworker ran a similar version of the design on that PC and his P3 650. I believe he said the time was 1:30 on the P3 and 1:15 on the K7. I haven't kept track of CPU's as much as I should, since I'm so busy, so I don't know what the differences are between the K7 with the latest Athlons. I wish I had had the time to look at the Athlons. According to NT's task manager, neither his runs nor my runs have reached 1GB yet, approx 750 MB. Xilinx recommends 2GB memory for the XC2V6000. The FPGA resources are intentionally kept as minimal as possible, which is about 25% slice utilization. (The design is parallel processing flow through the FPGA.) We expect at least 65% utilization at the end of integration Also, my unofficial guestimate of the execution time, is that it will go up at least by the square as the utilization doubles. It went up quite noticably from <10% to 25% utilization. I'm sure part of this depends on the design and how hard the tools have to work during par. Mark rickman wrote: > I am looking at a new machine as well. I would like to get an idea of > what a faster P3 would do in your tests. Any chance you can run your > PAR on a faster P3 or even a laptop? I believe they had P3s running at > 1.3 GHz before they released the P4. It would be very instructive to > see how that compares to the current P4s. > > Mark wrote: > > > > Hello, > > > > From a single recent very unofficial PC performance comparison .... > > > > Target: Virtex II XC2V6000 > > Xilinx: ngdbuild, map, par, trce using Design Manager. 4.1i, sp3 > > > > PC A: 650 MHz Pentium III, 1GB PC100 > > Execution time: ~1 hr: 30 min (+/- 5 min) > > > > PC B: 2.4 GHz Pentium IV, 2 GB PC2100 > > Execution time: 40 min > > > > Caveate: This is only one comparison of a single run on each computer. > > Most of the time (A: 1hr, 11 min B: 35 min) was spent in par, where the CPU > > utilization was 100%, according to NT 4.0, sp6 task manager. par execution > > time ratio: 71/35 ~= 2.03. Memory speed ratio: 266/100 ~= 2.66. CPU speed > > ratio: 2400/650 ~= 3.69. Although CPU utilization was 100%, the execution time > > ratio seems to imply that memory bandwidth was the limiting factor. But, the > > CPU was not <100%, so it's unclear why the execution time was not shorter. > > (IOW, I don't know what else the CPU was doing.) Screen saver, virus scanner, > > email, etc. were off. > > > > I've heard several times that FPGA tools are CPU intensive, but I think that > > machine specification should consider factors in addition to CPU speed, e.g., > > memory bandwidth, size of memory (to hold database), and M/B/chipset. We also > > considered a dual-CPU M/B. Unfortunately, no tools, that we have, can take > > advantage of multiprocessor PCs. A multiprocessor PC, in our case, would allow > > us to do synthesis/simulation, while the Xilinx tools are running. (I've tried > > to script the flow, but cannot since we're using ChipScope.) > > > > I believe that Xilinx on Linux needs Wine (from Xilinx installation notes), > > which I "heard" elsewhere ends up being slightly slower than native code. For > > front-end tools, we use the Mentor tools, which I believe have been ported to > > Linux, but, I haven't tried any, yet.... (I've been hoping ISD Mag will do > > another "annual" comparison.) Microsoft and Linux seem close as fas as speed, > > but, Linux seems to be ahead wrt stability. > > > > I haven't used Altera since the Flex 10K and MaxPlusII, so I don't have any > > experience with the latest Altera FPGAs. I also have no experience with the > > Xeon CPUs. > > > > Hope this helps, > > Mark > > > > mac teh knife wrote: > > > > > We have started development using the new fpgas. We are evaluating the > > > virtex 2 and stratix devices. > > > What I'm finding out is my 2 year old machine ain't got what it takes to > > > crunch the files that can fill up > > > these multi million gate chips. I was wondering if anybody would care to > > > share with us the machine (PC) > > > they are using. Also has anybody evaluated linux vs window performance as > > > far as fpga applications are concerned. > > > > > > I'm looking at a P4 2.4ghz / 2 gig ram and 533 mhz front bus. I'm also > > > looking at a Xeon 2.4 g with 400 mhz > > > front bus. Does anybody know if there is a performance difference between > > > these two type of processors? > > > > > > Mac the knife is of course not my real name, I'm just so tired of spam that > > > posting to use net generates. > > > > > > thanks > > > Jerry > > -- > > Rick "rickman" 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 URL http://www.arius.com > 4 King Ave 301-682-7772 Voice > Frederick, MD 21701-3110 301-682-7666 FAX ###### From: rickman Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Sat, 22 Jun 2002 12:59:09 -0400 Organization: Arius, Inc Lines: 43 Message-ID: <3D14ACDD.F5262E@yahoo.com> References: <3d0d331a_3@news.chartertn.net> <3D10286B.83A006FA@pac.net> <3D10BA33.BBB6983B@yahoo.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Trace: UmFuZG9tSVaHhmvN8ed3/Kro5haaK4ibKHzB4OorDxnsObuiWHFIw4rgQ/DICvAE X-Complaints-To: abuse@rcn.com NNTP-Posting-Date: 22 Jun 2002 16:58:53 GMT X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.wirehub.nl!news-hub.cableinet.net!blueyonder!newsfeed1.cidera.com!Cidera!nntp.abs.net!feed2.news.rcn.net!feed1.news.rcn.net!rcn!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18693 Nicholas Weaver wrote: > > In article <3D10BA33.BBB6983B@yahoo.com>, > rickman wrote: > >I am looking at a new machine as well. I would like to get an idea of > >what a faster P3 would do in your tests. Any chance you can run your > >PAR on a faster P3 or even a laptop? I believe they had P3s running at > >1.3 GHz before they released the P4. It would be very instructive to > >see how that compares to the current P4s. > > I'd definatly look at the Athlons as well. Although I don't have the > numbers to back them up, my intuition is that the tools are memory and > cache bound as much as CPU bound, and the athlons have a much better > cache heirarchy. > -- > Nicholas C. Weaver nweaver@cs.berkeley.edu But in laptops the Athons suck. The problem is that they are targeted to the low end of the market and the support chip sets are very low end. I don't think I have seen one that will support 1 GB of memory and none work with the slightly faster DDR memory. I am moving to a laptop for convenience and want to figure out which of the three are really faster in the laptop environment. So far I am pretty sure that the Athlon is at the bottom of the heap. Tom's hardware web page has a good article on Athlon chip sets and a new unnamed model with a better chip set. But it still is much slower than the P4s, mainly because of the memory. I am just not sure about the P3s. -- Rick "rickman" 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 URL http://www.arius.com 4 King Ave 301-682-7772 Voice Frederick, MD 21701-3110 301-682-7666 FAX ###### Path: chonsp.franklin.ch!not-for-mail From: Neil Franklin Newsgroups: comp.arch.fpga Subject: Re: new computer Date: 22 Jun 2002 20:00:40 +0200 Organization: My own Private Self Lines: 60 Message-ID: <6u660bi46v.fsf@chonsp.franklin.ch> References: <3d0d331a_3@news.chartertn.net> <3D10286B.83A006FA@pac.net> <3D10BA33.BBB6983B@yahoo.com> <3D14ACDD.F5262E@yahoo.com> NNTP-Posting-Host: chonsp.franklin.ch X-Trace: chonsp.franklin.ch 1024768840 611 10.0.3.2 (22 Jun 2002 18:00:40 GMT) X-Complaints-To: news@chonsp.franklin.ch NNTP-Posting-Date: 22 Jun 2002 18:00:40 GMT X-Newsreader: Gnus v5.7/Emacs 20.4 Xref: chonsp.franklin.ch comp.arch.fpga:18694 rickman writes: > Nicholas Weaver wrote: > > > > In article <3D10BA33.BBB6983B@yahoo.com>, > > rickman wrote: > > >PAR on a faster P3 or even a laptop? I believe they had P3s running at > > >1.3 GHz before they released the P4. It would be very instructive to > > >see how that compares to the current P4s. > > > > I'd definatly look at the Athlons as well. Although I don't have the > > But in laptops the Athons suck. The problem is that they are targeted > to the low end of the market and the support chip sets are very low > end. Generally Athlon chipsets are not up to speed. That is bad because the processor is. The boards/chipsets for Athlon MP (multiprocessor) chips seem to be better because less cost limits have lead to them pulling all stops. This is so even if you only use them as single processor machines. > I am moving to a laptop > for convenience That will though cost speed, in any case. Laptops are good enough for office work. But eng/sci computations (and so most likely PAR) will lag on them. > unnamed model with a better chip set. But it still is much slower than > the P4s, mainly because of the memory. Warning: P4s are incredibly bad for long duration jobs! They are fast so long one has jobs ths compute heavily for a short time and then take a break. The problem comes from heating up, without breaks to cool off, and then stepping down the clock to stave off overheating. Athlons just heat up and run, so long the cooling system is good enough to prevent them being killed. All our (university physics dept) engineering and science computation machines are now being specified as Athlon only. > I am just not sure about the > P3s. Good on heavy integer, can even beat P4 on that. Is PAR integer code or FP dominated? Also good solid optimised on IO in the chipsets. We specify P3 for all servers. But that is not much use for PAR. -- Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/ Hacker, Unix Guru, El Eng HTL/BSc, Programmer, Archer, Roleplayer - Make your code truely free: put it into the public domain ###### From: rickman Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Sun, 23 Jun 2002 03:09:16 -0400 Organization: Arius, Inc Lines: 78 Message-ID: <3D15741C.D46FC639@yahoo.com> References: <3d0d331a_3@news.chartertn.net> <3D10286B.83A006FA@pac.net> <3D10BA33.BBB6983B@yahoo.com> <3D14ACDD.F5262E@yahoo.com> <6u660bi46v.fsf@chonsp.franklin.ch> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Trace: UmFuZG9tSVY5UDWkf4XOT3EFtXsCkq26uA4NKwjX2MrQq3Ky7H6MRf4ycUXjtbM8 X-Complaints-To: abuse@rcn.com NNTP-Posting-Date: 23 Jun 2002 07:08:50 GMT X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!feed2.news.rcn.net!feed1.news.rcn.net!rcn!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18704 Neil Franklin wrote: > > rickman writes: > > But in laptops the Athons suck. The problem is that they are targeted > > to the low end of the market and the support chip sets are very low > > end. > > Generally Athlon chipsets are not up to speed. That is bad because the > processor is. The boards/chipsets for Athlon MP (multiprocessor) chips > seem to be better because less cost limits have lead to them pulling > all stops. This is so even if you only use them as single processor > machines. That may be true in some context, but the benchmarks I have read indicate that the Athlons in standard desktops run quite well beating P3s at 20% higher clock speeds. The advantage over P4s is even more. But laptops are where they seem to be filling the low end and the chip sets are exclusively designed for that. > > I am moving to a laptop > > for convenience > > That will though cost speed, in any case. Laptops are good enough for > office work. But eng/sci computations (and so most likely PAR) will lag > on them. Of course laptops will not match desktops. But a new laptop is much better than the desktop I am currently using. I am just trying to get information on which processor is currently best for FPGA work in laptops. I can do quite well without a desktop for the time being. > > unnamed model with a better chip set. But it still is much slower than > > the P4s, mainly because of the memory. > > Warning: P4s are incredibly bad for long duration jobs! They are fast > so long one has jobs ths compute heavily for a short time and then > take a break. The problem comes from heating up, without breaks to > cool off, and then stepping down the clock to stave off overheating. Can you cite a source for this info? I have never heard that P4s won't run at full speed for long periods. This may be true in laptops, I don't doubt, but I can't belive they would design a desktop that won't crunch at top clock all day. > Athlons just heat up and run, so long the cooling system is good > enough to prevent them being killed. All our (university physics dept) > engineering and science computation machines are now being specified > as Athlon only. > > > I am just not sure about the > > P3s. > > Good on heavy integer, can even beat P4 on that. Is PAR integer code > or FP dominated? > > Also good solid optimised on IO in the chipsets. We specify P3 for all > servers. But that is not much use for PAR. I was asking about info specific to FPGA work and also laptops if available. Most of the benchmarks I have seen are targeted to office apps or graphics. This thread contained some of the few specs on FPGA applications I have seen. -- Rick "rickman" 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 URL http://www.arius.com 4 King Ave 301-682-7772 Voice Frederick, MD 21701-3110 301-682-7666 FAX ###### From: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Sun, 23 Jun 2002 16:38:05 +0000 (UTC) Organization: Unknown Lines: 46 Message-ID: References: <3d0d331a_3@news.chartertn.net> <3D14ACDD.F5262E@yahoo.com> <6u660bi46v.fsf@chonsp.franklin.ch> <3D15741C.D46FC639@yahoo.com> NNTP-Posting-Host: soda.csua.berkeley.edu X-Trace: agate.berkeley.edu 1024850285 16384 128.32.247.226 (23 Jun 2002 16:38:05 GMT) X-Complaints-To: usenet@agate.berkeley.edu NNTP-Posting-Date: Sun, 23 Jun 2002 16:38:05 +0000 (UTC) Originator: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!cyclone.bc.net!news-hog.berkeley.edu!ucberkeley!agate.berkeley.edu!agate!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18714 In article <3D15741C.D46FC639@yahoo.com>, rickman wrote: >That may be true in some context, but the benchmarks I have read >indicate that the Athlons in standard desktops run quite well beating >P3s at 20% higher clock speeds. The advantage over P4s is even more. >But laptops are where they seem to be filling the low end and the chip >sets are exclusively designed for that. Also, remember that the P4 started out with some REALLY REALLY crappy chipsets, either RDRAM (horribly expensive) or poor and therefore slow SDRAM implementations. Which did effect things. I hope Hammer comes out at speed, that will be a great foundation for CAD machines, especially since the memory-controller on die should help memory latency (massive L1 and L2, 30 cycles plus access time to get to main memory). >Of course laptops will not match desktops. But a new laptop is much >better than the desktop I am currently using. I am just trying to get >information on which processor is currently best for FPGA work in >laptops. I can do quite well without a desktop for the time being. You might want to reconsider, simply because the desktops are SO much cheaper. You can build yourself a really nice CAD machine for the desktop for a low cost, esp if you reuse your monitor. As for laptops, my GUESS would be a PIII laptop would be better than a P4, because the PIII has a better L1 cache, runs faster on a clock-per-clock basis (so better on a power basis). I'd THINK that simulated annealing would be memory and integer bound (but you could write it so that it is memory and FP bound, depending on how you do the cost function). Anyone at Brand X or Brand A care to comment? >Can you cite a source for this info? I have never heard that P4s won't >run at full speed for long periods. This may be true in laptops, I >don't doubt, but I can't belive they would design a desktop that won't >crunch at top clock all day. The P4 has a clock throttling thermal diode. In case of overheat, it cuts the clock in half. Early systems had poorly implemented cooling, so you could trigger it during normal use on the early desktops. -- Nicholas C. Weaver nweaver@cs.berkeley.edu ###### From: rickman Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Sun, 23 Jun 2002 14:15:13 -0400 Organization: Arius, Inc Lines: 67 Message-ID: <3D161031.480B750C@yahoo.com> References: <3d0d331a_3@news.chartertn.net> <3D14ACDD.F5262E@yahoo.com> <6u660bi46v.fsf@chonsp.franklin.ch> <3D15741C.D46FC639@yahoo.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Trace: UmFuZG9tSVbzAllGrkKh5nHKwK1FJmoeqqJ9c6l0/80VFwB4xzqEUIkOHMtIyufT X-Complaints-To: abuse@rcn.com NNTP-Posting-Date: 23 Jun 2002 18:14:39 GMT X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.icl.net!news.maxwell.syr.edu!feed2.news.rcn.net!feed1.news.rcn.net!rcn!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18717 Nicholas Weaver wrote: > > In article <3D15741C.D46FC639@yahoo.com>, > rickman wrote: > >That may be true in some context, but the benchmarks I have read > >indicate that the Athlons in standard desktops run quite well beating > >P3s at 20% higher clock speeds. The advantage over P4s is even more. > >But laptops are where they seem to be filling the low end and the chip > >sets are exclusively designed for that. > > Also, remember that the P4 started out with some REALLY REALLY crappy > chipsets, either RDRAM (horribly expensive) or poor and therefore slow > SDRAM implementations. Which did effect things. What did it affect? I am not following you here. > >Of course laptops will not match desktops. But a new laptop is much > >better than the desktop I am currently using. I am just trying to get > >information on which processor is currently best for FPGA work in > >laptops. I can do quite well without a desktop for the time being. > > You might want to reconsider, simply because the desktops are SO much > cheaper. You can build yourself a really nice CAD machine for the > desktop for a low cost, esp if you reuse your monitor. I think I know my needs better than most other people know my needs. A desktop is nice, but I get really tired of packing it up when I have to drive to my other work locations. Whew, BIG box. I can't even get the 19" monitor box in my seat, it has to go in the back of the pickup! > As for laptops, my GUESS would be a PIII laptop would be better than a > P4, because the PIII has a better L1 cache, runs faster on a > clock-per-clock basis (so better on a power basis). This is one of those situations where having real data far surpasses armchair analysis. All of the laptop tests I have seen shows avaiable P4s outrunning available P3s and lasting just as long if not longer on batteries. They may be using larger batteries or they may have better power control. Don't know, don't care. > >Can you cite a source for this info? I have never heard that P4s won't > >run at full speed for long periods. This may be true in laptops, I > >don't doubt, but I can't belive they would design a desktop that won't > >crunch at top clock all day. > > The P4 has a clock throttling thermal diode. In case of overheat, it > cuts the clock in half. Early systems had poorly implemented cooling, > so you could trigger it during normal use on the early desktops. But you don't have any data or reports showing that this is a common problem? Do you know any brand names or models? -- Rick "rickman" 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 URL http://www.arius.com 4 King Ave 301-682-7772 Voice Frederick, MD 21701-3110 301-682-7666 FAX ###### Path: chonsp.franklin.ch!not-for-mail From: Neil Franklin Newsgroups: comp.arch.fpga Subject: Re: new computer Date: 23 Jun 2002 21:23:49 +0200 Organization: My own Private Self Lines: 49 Message-ID: <6uelexn6ii.fsf@chonsp.franklin.ch> References: <3d0d331a_3@news.chartertn.net> <3D14ACDD.F5262E@yahoo.com> <6u660bi46v.fsf@chonsp.franklin.ch> <3D15741C.D46FC639@yahoo.com> <3D161031.480B750C@yahoo.com> NNTP-Posting-Host: chonsp.franklin.ch X-Trace: chonsp.franklin.ch 1024860229 1154 10.0.3.2 (23 Jun 2002 19:23:49 GMT) X-Complaints-To: news@chonsp.franklin.ch NNTP-Posting-Date: 23 Jun 2002 19:23:49 GMT X-Newsreader: Gnus v5.7/Emacs 20.4 Xref: chonsp.franklin.ch comp.arch.fpga:18720 rickman writes: > Nicholas Weaver wrote: > > > > rickman wrote: > > > > >Can you cite a source for this info? Tests run by my office mate, just before I moved to this office im Mai (so I do not know the details, just the result: sci/eng get Athlons). > > I have never heard that P4s won't > > >run at full speed for long periods. This may be true in laptops, I It definitely was desktop machines. The only notebooks in out setup are students or professors private machines. All computer center stuff is desktop or server. > > >don't doubt, but I can't belive they would design a desktop that won't > > >crunch at top clock all day. The largest part of Intels users will be running Office, that does not crunch numbers for long times. > > The P4 has a clock throttling thermal diode. In case of overheat, it > > cuts the clock in half. Early systems had poorly implemented cooling, > > so you could trigger it during normal use on the early desktops. And first generation P4s get/got awfully hot. Them GHz come with lots of current usage. > But you don't have any data or reports showing that this is a common > problem? Do you know any brand names or models? Tests were on whatever demo/test systems our standard vendor (http://www.dalco.ch/index.php) delivered for that test series. And they are specialised into delivering machines for sci/eng, so these were most likely not the cheapest PC crap. They also delivered our 502 processor Beowulf (http://www.asgard.ethz.ch/). -- Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/ Hacker, Unix Guru, El Eng HTL/BSc, Programmer, Archer, Roleplayer - Make your code truely free: put it into the public domain ###### From: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Mon, 24 Jun 2002 03:06:26 +0000 (UTC) Organization: Unknown Lines: 45 Message-ID: References: <3d0d331a_3@news.chartertn.net> <3D15741C.D46FC639@yahoo.com> <3D161031.480B750C@yahoo.com> NNTP-Posting-Host: soda.csua.berkeley.edu X-Trace: agate.berkeley.edu 1024887986 34656 128.32.247.226 (24 Jun 2002 03:06:26 GMT) X-Complaints-To: usenet@agate.berkeley.edu NNTP-Posting-Date: Mon, 24 Jun 2002 03:06:26 +0000 (UTC) Originator: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.icl.net!newsfeed1.cidera.com!Cidera!news-hog.berkeley.edu!ucberkeley!agate.berkeley.edu!agate!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18737 In article <3D161031.480B750C@yahoo.com>, rickman wrote: >> Also, remember that the P4 started out with some REALLY REALLY crappy >> chipsets, either RDRAM (horribly expensive) or poor and therefore slow >> SDRAM implementations. Which did effect things. > >What did it affect? I am not following you here. It meant that the P4 initially had horrible memory performance compared to the DDR athlons at a reasponable price point (as RDRAM is expensive), as only SDRAM was offered in the Intel chipsets, and even that wasn't great. >> You might want to reconsider, simply because the desktops are SO much >> cheaper. You can build yourself a really nice CAD machine for the >> desktop for a low cost, esp if you reuse your monitor. > >I think I know my needs better than most other people know my needs. A >desktop is nice, but I get really tired of packing it up when I have to >drive to my other work locations. Whew, BIG box. I can't even get the >19" monitor box in my seat, it has to go in the back of the pickup! If your real goal is a mobile desktop, why not look at a Shuttle all-in-one system. You can get a P4 or Athlon, in a small aluminum cube, (SS50 is the P4, SS40 is the Athlon) and a LCD display, pack it up in a backpack setup and you have a mobile desktop, with some real CPU power in it, which should be in the <20 lbs and highly packable range. You lose about 20% memory bandwidth due to shared video but you do a lot better than a laptop, as you could have a >2.0 GHz P4 with DDR-DRAM. >> The P4 has a clock throttling thermal diode. In case of overheat, it >> cuts the clock in half. Early systems had poorly implemented cooling, >> so you could trigger it during normal use on the early desktops. > >But you don't have any data or reports showing that this is a common >problem? Do you know any brand names or models? There was some debate on the subject when the P4 first came out. Tom's Hardware shows what happens when you de-heatsink it. It tends to work fine (no slowdown) even in small cases (EG, SV50 case), at least for the desktop machines. -- Nicholas C. Weaver nweaver@cs.berkeley.edu ###### From: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Mon, 24 Jun 2002 03:07:12 +0000 (UTC) Organization: Unknown Lines: 14 Message-ID: References: <3d0d331a_3@news.chartertn.net> <3D15741C.D46FC639@yahoo.com> <1024857214.2262.0.nnrp-14.9e9832fa@news.demon.co.uk> NNTP-Posting-Host: soda.csua.berkeley.edu X-Trace: agate.berkeley.edu 1024888032 34898 128.32.247.226 (24 Jun 2002 03:07:12 GMT) X-Complaints-To: usenet@agate.berkeley.edu NNTP-Posting-Date: Mon, 24 Jun 2002 03:07:12 +0000 (UTC) Originator: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.icl.net!news.maxwell.syr.edu!news-hog.berkeley.edu!ucberkeley!agate.berkeley.edu!agate!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18735 In article <1024857214.2262.0.nnrp-14.9e9832fa@news.demon.co.uk>, Tim wrote: >Nicholas Weaver wrote > >> I'd THINK that simulated annealing would be memory and integer bound >> (but you could write it so that it is memory and FP bound, depending >> on how you do the cost function). > >Is simulated annealing the basis of the placement algorithms? Usually, yes. -- Nicholas C. Weaver nweaver@cs.berkeley.edu ###### From: rickman Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Mon, 24 Jun 2002 00:25:11 -0400 Organization: Arius, Inc Lines: 86 Message-ID: <3D169F27.B4357423@yahoo.com> References: <3d0d331a_3@news.chartertn.net> <3D15741C.D46FC639@yahoo.com> <3D161031.480B750C@yahoo.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Trace: UmFuZG9tSVYo1TjauP2Q+an1Ve0pGX3YIMQVkZkOj5p0M/5gVkr5etsO+1VBVE7N X-Complaints-To: abuse@rcn.com NNTP-Posting-Date: 24 Jun 2002 04:25:08 GMT X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!feed2.news.rcn.net!feed1.news.rcn.net!rcn!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18761 Nicholas Weaver wrote: > > In article <3D161031.480B750C@yahoo.com>, > rickman wrote: > >> Also, remember that the P4 started out with some REALLY REALLY crappy > >> chipsets, either RDRAM (horribly expensive) or poor and therefore slow > >> SDRAM implementations. Which did effect things. > > > >What did it affect? I am not following you here. > > It meant that the P4 initially had horrible memory performance > compared to the DDR athlons at a reasponable price point (as RDRAM is > expensive), as only SDRAM was offered in the Intel chipsets, and even > that wasn't great. I don't understand why you are talking about this. Perhaps we are having two different conversations. I am interested in today's computers and most specifically laptops. I don't see how the memory used in a computer a year or more ago is of interest to either of us. > >> You might want to reconsider, simply because the desktops are SO much > >> cheaper. You can build yourself a really nice CAD machine for the > >> desktop for a low cost, esp if you reuse your monitor. > > > >I think I know my needs better than most other people know my needs. A > >desktop is nice, but I get really tired of packing it up when I have to > >drive to my other work locations. Whew, BIG box. I can't even get the > >19" monitor box in my seat, it has to go in the back of the pickup! > > If your real goal is a mobile desktop, why not look at a Shuttle > all-in-one system. You can get a P4 or Athlon, in a small aluminum > cube, (SS50 is the P4, SS40 is the Athlon) and a LCD display, pack it > up in a backpack setup and you have a mobile desktop, with some real > CPU power in it, which should be in the <20 lbs and highly packable > range. You lose about 20% memory bandwidth due to shared video but > you do a lot better than a laptop, as you could have a >2.0 GHz P4 > with DDR-DRAM. I would never buy a machine with shared video memory because 20% performance loss is not worth the small price savings realized. I also am not interested in a cube that is not functional until you find a power outlet and a desk to set the keyboard, monitor and ect. Since you don't know how fast current laptops are, how can you say that a desktop with performance limitations is better? BTW, I can buy a mobile P4 to run at 2.0 GHz with DDR in a laptop. I am not asking for you to rethink my basic decisions. I am asking about performance of various laptops to use for FPGA design. The laptop CPUs I can buy today are about as good as the desktop CPUs I could not have bought until 6 months ago, or maybe even less. I am willing to pay the price premium to gain the portability and ease of setup. > >> The P4 has a clock throttling thermal diode. In case of overheat, it > >> cuts the clock in half. Early systems had poorly implemented cooling, > >> so you could trigger it during normal use on the early desktops. > > > >But you don't have any data or reports showing that this is a common > >problem? Do you know any brand names or models? > > There was some debate on the subject when the P4 first came out. Tom's > Hardware shows what happens when you de-heatsink it. It tends to work > fine (no slowdown) even in small cases (EG, SV50 case), at least for > the desktop machines. Yes, and all of that data has little bearing on the real point. If I buy machine X, will it have a heat problem when run at full speed, that is will it self limit due to heat? I will be asking that question as I make my selection. But I don't see how a thermal protection mechanism makes a chip unsuitable for computation use. -- Rick "rickman" 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 URL http://www.arius.com 4 King Ave 301-682-7772 Voice Frederick, MD 21701-3110 301-682-7666 FAX ###### From: rickman Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Mon, 24 Jun 2002 00:37:32 -0400 Organization: Arius, Inc Lines: 79 Message-ID: <3D16A20C.2809B2A3@yahoo.com> References: <3d0d331a_3@news.chartertn.net> <3D14ACDD.F5262E@yahoo.com> <6u660bi46v.fsf@chonsp.franklin.ch> <3D15741C.D46FC639@yahoo.com> <3D161031.480B750C@yahoo.com> <6uelexn6ii.fsf@chonsp.franklin.ch> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Trace: UmFuZG9tSVZMMcd9nJ+1UaC1I61gm2aNg2Ip/ZMg8kGkk6OYcAzsG4TpaGhdY6NI X-Complaints-To: abuse@rcn.com NNTP-Posting-Date: 24 Jun 2002 04:37:30 GMT X-Mailer: Mozilla 4.7 [en] (Win95; U) X-Accept-Language: en Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.imp.ch!news.imp.ch!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!feed2.news.rcn.net!feed1.news.rcn.net!rcn!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18762 Neil Franklin wrote: > > rickman writes: > > > Nicholas Weaver wrote: > > > > > > rickman wrote: > > > > > > >Can you cite a source for this info? > > Tests run by my office mate, just before I moved to this office im Mai > (so I do not know the details, just the result: sci/eng get Athlons). But if we don't have any info on how the tests were done, then we don't know under what conditions the results will change. Both Athlons and P4s have been changing over the last year. I also do not understand exactly how the CPU heats up as you do "computation". Have they changed the OS so that the CPU stops when you don't give it something to do? Does the clock slow down when running the IDLE task? I suppose that memory accesses are reduced since you should get more cache hits, but I don't see how the CPU heat will change significantly. > > > >don't doubt, but I can't belive they would design a desktop that won't > > > >crunch at top clock all day. > > The largest part of Intels users will be running Office, that does not > crunch numbers for long times. But even if 1% of the machines are running workhorse code, it would be immediately recognized if they were slower than other machines with much lower clock ratings. In my last position we had a bank of servers to crunch FPGA work and we would have known there was a problem. We had one P4 which was used by the group designing in the largest FPGAs we were using. We did not see a problem and this 1 GHz machine ran faster than the 750 MHz P3s. I can't belive Intel is making a chip or that DELL, etc are making machines that slow down running "normal" FPGA apps. > > > The P4 has a clock throttling thermal diode. In case of overheat, it > > > cuts the clock in half. Early systems had poorly implemented cooling, > > > so you could trigger it during normal use on the early desktops. > > And first generation P4s get/got awfully hot. Them GHz come with lots > of current usage. We are no longer buying first gen P4s. They are up to 2.5 GHz or so and have reduced geometries with lower power demands. I will check with some former coworkers and see if they have a newer, faster P4 server now. > > But you don't have any data or reports showing that this is a common > > problem? Do you know any brand names or models? > > Tests were on whatever demo/test systems our standard vendor > (http://www.dalco.ch/index.php) delivered for that test series. And > they are specialised into delivering machines for sci/eng, so these > were most likely not the cheapest PC crap. They also delivered our > 502 processor Beowulf (http://www.asgard.ethz.ch/). So you don't have any specific data on what machine type had what problem, just general recommendations? -- Rick "rickman" 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 URL http://www.arius.com 4 King Ave 301-682-7772 Voice Frederick, MD 21701-3110 301-682-7666 FAX ###### From: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Newsgroups: comp.arch.fpga Subject: Re: new computer Date: Mon, 24 Jun 2002 15:52:23 +0000 (UTC) Organization: Unknown Lines: 46 Message-ID: References: <3d0d331a_3@news.chartertn.net> <3D161031.480B750C@yahoo.com> <6uelexn6ii.fsf@chonsp.franklin.ch> <3D16A20C.2809B2A3@yahoo.com> NNTP-Posting-Host: soda.csua.berkeley.edu X-Trace: agate.berkeley.edu 1024933943 57280 128.32.247.226 (24 Jun 2002 15:52:23 GMT) X-Complaints-To: usenet@agate.berkeley.edu NNTP-Posting-Date: Mon, 24 Jun 2002 15:52:23 +0000 (UTC) Originator: nweaver@CSUA.Berkeley.EDU (Nicholas Weaver) Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news.mailgate.org!newsfeed.icl.net!news.maxwell.syr.edu!news-hog.berkeley.edu!ucberkeley!agate.berkeley.edu!agate!not-for-mail Xref: chonsp.franklin.ch comp.arch.fpga:18738 In article <3D16A20C.2809B2A3@yahoo.com>, rickman wrote: >I also do not understand exactly how the CPU heats up as you do >"computation". Have they changed the OS so that the CPU stops when you >don't give it something to do? Does the clock slow down when running >the IDLE task? I suppose that memory accesses are reduced since you >should get more cache hits, but I don't see how the CPU heat will change >significantly. The P4 uses alot of gated clocks, so large hunks of the datapath aren't clocked unless they are in use. Power consumption is very computation dependant on the P4. >But even if 1% of the machines are running workhorse code, it would be >immediately recognized if they were slower than other machines with much >lower clock ratings. In my last position we had a bank of servers to >crunch FPGA work and we would have known there was a problem. We had >one P4 which was used by the group designing in the largest FPGAs we >were using. We did not see a problem and this 1 GHz machine ran faster >than the 750 MHz P3s. I can't belive Intel is making a chip or that >DELL, etc are making machines that slow down running "normal" FPGA >apps. The INITIAL P4s ran very hot, and some people did notice the slowdown happening on cases with poor cooling, notably on Quake and similar sustained processor intensive tasks. It actually doesn't slow the clock, it just switches on and off at a 2 us period, so a roughly 50% duty cycle, using the existing STOPCLOCK ability. Do a google search for "P4 thermal diode" and a lot of the flameage still remains. The newer ones (post process shrink) run significantly cooler, and the cooling has also somewhat improved, and the understanding of the importance of cooling it. http://developer.intel.com/technology/itj/q12001/pdf/art_4.pdf Is a paper on it. An interesting observation: they wanted to reduce power 20%, but reduce the thermal design point by 40%. As such, they wanted the cooling to be omre in line with expected, and high expected, not peak cooling. -- Nicholas C. Weaver nweaver@cs.berkeley.edu ###### Path: chonsp.franklin.ch!not-for-mail From: Neil Franklin Newsgroups: comp.arch.fpga Subject: Re: new computer Date: 24 Jun 2002 23:12:32 +0200 Organization: My own Private Self Lines: 92 Message-ID: <6uit48xtxb.fsf@chonsp.franklin.ch> References: <3d0d331a_3@news.chartertn.net> <3D14ACDD.F5262E@yahoo.com> <6u660bi46v.fsf@chonsp.franklin.ch> <3D15741C.D46FC639@yahoo.com> <3D161031.480B750C@yahoo.com> <6uelexn6ii.fsf@chonsp.franklin.ch> <3D16A20C.2809B2A3@yahoo.com> NNTP-Posting-Host: chonsp.franklin.ch X-Trace: chonsp.franklin.ch 1024953152 1073 10.0.3.2 (24 Jun 2002 21:12:32 GMT) X-Complaints-To: news@chonsp.franklin.ch NNTP-Posting-Date: 24 Jun 2002 21:12:32 GMT X-Newsreader: Gnus v5.7/Emacs 20.4 Xref: chonsp.franklin.ch comp.arch.fpga:18765 rickman writes: > Neil Franklin wrote: > > > > Tests run by my office mate, just before I moved to this office im Mai > > (so I do not know the details, just the result: sci/eng get Athlons). > > But if we don't have any info on how the tests were done, then we don't > know under what conditions the results will change. Both Athlons and > P4s have been changing over the last year. I only know the tests were done fairly recently (stongly assume this year). I will be seeing this colleague again on Thursday and can ask him then. > I also do not understand exactly how the CPU heats up as you do > "computation". Have they changed the OS so that the CPU stops when you > don't give it something to do? That is standard in PCs since about 10 years in anything which is not DOS-based Windows (which ignores this hardware feature). My 5 year old AMD K6-2/350 running Linux has an pronounced increase in ventilator (thermaly controlled) noise as soon as I compile something, within split seconds. > Does the clock slow down when running > the IDLE task? Slow or even switch off until next interrupt occurs. > > > > >don't doubt, but I can't belive they would design a desktop that won't > > > > >crunch at top clock all day. > > > > The largest part of Intels users will be running Office, that does not > > crunch numbers for long times. > > But even if 1% of the machines are running workhorse code, it would be > immediately recognized if they were slower than other machines with much > lower clock ratings. That is aparently exactly the observation made. > lower clock ratings. In my last position we had a bank of servers to > crunch FPGA work and we would have known there was a problem. We had > one P4 which was used by the group designing in the largest FPGAs we > were using. We did not see a problem and this 1 GHz machine ran faster > than the 750 MHz P3s. a) perhaps they were sufficiently cooled, being server machines b) perhaps the problem only starts above 1GHz. > I can't belive Intel is making a chip or that > DELL, etc are making machines that slow down running "normal" FPGA > apps. 99% of their customers could not care. And cost saving ist cost saving. Lower price or can be spent on 0.1GHz more, which markets itsself better. That counts in todays PC marketplace. The victory of crapitalism. > > > > The P4 has a clock throttling thermal diode. In case of overheat, it > > > > cuts the clock in half. Early systems had poorly implemented cooling, > > > > so you could trigger it during normal use on the early desktops. > > > > And first generation P4s get/got awfully hot. Them GHz come with lots > > of current usage. > > We are no longer buying first gen P4s. They are up to 2.5 GHz or so and > have reduced geometries with lower power demands. Reduced geometry and increased clock. And power goes up linear with clocking. > I will check with > some former coworkers and see if they have a newer, faster P4 server > now. I will check on Thursday. -- Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/ Hacker, Unix Guru, El Eng HTL/BSc, Programmer, Archer, Roleplayer - Make your code truely free: put it into the public domain