new PC specs for Xilinx tools
in [FPGA]
|
Prev: Difference among Virtex Families, FPGA Books
Next: The UK Embedded Masterclass 2010 - 6th and 11th May Cambridge & Reading
From: General Schvantzkoph on 8 Jan 2010 10:22 On Fri, 08 Jan 2010 09:10:34 +0000, HT-Lab wrote: > "whygee" <yg(a)yg.yg> wrote in message > news:hi5896$h18$1(a)speranza.aioe.org... >> hi, >> >> Tom Kotwal wrote: >>> - Core 2 Duo vs i7? >> depends if you can find a laptop with it. It seems to be quite worth it >> but the i7 is much more expensive... It may have changed since the last >> time I've looked, > > This month edition of PCPro (UK edition but I believe they use the same > articles worldwide) tested 12 of the latest laptops and the winner is > the Dell Studio 15 which comes with a 1.6GHz Core I7-720QM. According to > the article the £653 (ex. VAT) Dell Studio beats many £2000 high-end > Core2 laptops in performance. Unfortunately the battery life is not that > great but then again I don't think you will be running a P&R session at > the airport :-) > > Hans. > www.ht-lab.com The benchmarks in magazines and on the hardware sites are useless. They are mostly testing gaming and multimedia performance which is completely irrelevant. When you see benchmarks where the iCore7 exceeds the performance of the Core2 it's because they are measuring the performance of the multimedia extensions to the instruction set. Those instructions aren't used by any CAE tools. What's important is basic integer performance which really boils down to cache size and latency.
From: whygee on 8 Jan 2010 10:34 General Schvantzkoph wrote: > On Fri, 08 Jan 2010 09:10:34 +0000, HT-Lab wrote: > The benchmarks in magazines and on the hardware sites are useless. They > are mostly testing gaming and multimedia performance which is completely > irrelevant. When you see benchmarks where the iCore7 exceeds the > performance of the Core2 it's because they are measuring the performance > of the multimedia extensions to the instruction set. Those instructions > aren't used by any CAE tools. What's important is basic integer > performance which really boils down to cache size and latency. and bandwidth too :-) However, I just tested the Atom-based ACER ONE with Actel's Libero and it does not seem to lag much behind the Core2 duo Toshiba that I have used since 2008. OK, my designs are small, so a lot of OS overhead and GUI clutter comes into the equation, but my point is : a cheap "netbook" is enough for small/medium designs. Price is lower and portability is better than "heating lap bricks". For the price of a PDA and about 2x the size, I can do full VHDL development so I'm satisfied :-) regards, yg -- http://ygdes.com / http://yasep.org
From: HT-Lab on 8 Jan 2010 11:25 "General Schvantzkoph" <schvantzkoph(a)yahoo.com> wrote in message news:7qp0tdFhpkU5(a)mid.individual.net... > On Fri, 08 Jan 2010 09:10:34 +0000, HT-Lab wrote: > >> "whygee" <yg(a)yg.yg> wrote in message >> news:hi5896$h18$1(a)speranza.aioe.org... >>> hi, >>> >>> Tom Kotwal wrote: >>>> - Core 2 Duo vs i7? >>> depends if you can find a laptop with it. It seems to be quite worth it >>> but the i7 is much more expensive... It may have changed since the last >>> time I've looked, >> >> This month edition of PCPro (UK edition but I believe they use the same >> articles worldwide) tested 12 of the latest laptops and the winner is >> the Dell Studio 15 which comes with a 1.6GHz Core I7-720QM. According to >> the article the �653 (ex. VAT) Dell Studio beats many �2000 high-end >> Core2 laptops in performance. Unfortunately the battery life is not that >> great but then again I don't think you will be running a P&R session at >> the airport :-) >> >> Hans. >> www.ht-lab.com > > > The benchmarks in magazines and on the hardware sites are useless. They > are mostly testing gaming and multimedia performance which is completely > irrelevant. You may be right although I wouldn't call them useless, with any benchmark you have to use it as a rough indicator. > When you see benchmarks where the iCore7 exceeds the > performance of the Core2 it's because they are measuring the performance > of the multimedia extensions to the instruction set. Those instructions > aren't used by any CAE tools. Are you sure, do you know what each vendors is using for their algorithms? I wouldn't be surpriced if they use some of the SIMD capabilities. > What's important is basic integer > performance which really boils down to cache size and latency. Right, so a quick google showed that the SPECint for the slowest i7-920 is nearly 20% higher than the fastest Extreme QX9770. Benchmarking is notoriously difficult and even the SPECint won't tell you if a PC is fast or not given that it depends on so many other factors. PCPro seems to be using a multitude of applications ranging from office apps to multitasking so I guess it is better than nothing. Hans www.ht-lab.com
From: General Schvantzkoph on 8 Jan 2010 12:04 On Fri, 08 Jan 2010 16:25:22 +0000, HT-Lab wrote: > "General Schvantzkoph" <schvantzkoph(a)yahoo.com> wrote in message > news:7qp0tdFhpkU5(a)mid.individual.net... >> On Fri, 08 Jan 2010 09:10:34 +0000, HT-Lab wrote: >> >>> "whygee" <yg(a)yg.yg> wrote in message >>> news:hi5896$h18$1(a)speranza.aioe.org... >>>> hi, >>>> >>>> Tom Kotwal wrote: >>>>> - Core 2 Duo vs i7? >>>> depends if you can find a laptop with it. It seems to be quite worth >>>> it but the i7 is much more expensive... It may have changed since the >>>> last time I've looked, >>> >>> This month edition of PCPro (UK edition but I believe they use the >>> same articles worldwide) tested 12 of the latest laptops and the >>> winner is the Dell Studio 15 which comes with a 1.6GHz Core I7-720QM. >>> According to the article the £653 (ex. VAT) Dell Studio beats many >>> £2000 high-end Core2 laptops in performance. Unfortunately the battery >>> life is not that great but then again I don't think you will be >>> running a P&R session at the airport :-) >>> >>> Hans. >>> www.ht-lab.com >> >> >> The benchmarks in magazines and on the hardware sites are useless. They >> are mostly testing gaming and multimedia performance which is >> completely irrelevant. > > You may be right although I wouldn't call them useless, with any > benchmark you have to use it as a rough indicator. > >> When you see benchmarks where the iCore7 exceeds the performance of the >> Core2 it's because they are measuring the performance of the multimedia >> extensions to the instruction set. Those instructions aren't used by >> any CAE tools. > > Are you sure, do you know what each vendors is using for their > algorithms? I wouldn't be surpriced if they use some of the SIMD > capabilities. > >> What's important is basic integer >> performance which really boils down to cache size and latency. > > Right, so a quick google showed that the SPECint for the slowest i7-920 > is nearly 20% higher than the fastest Extreme QX9770. > > Benchmarking is notoriously difficult and even the SPECint won't tell > you if a PC is fast or not given that it depends on so many other > factors. PCPro seems to be using a multitude of applications ranging > from office apps to multitasking so I guess it is better than nothing. > > Hans > www.ht-lab.com The only way to really find out what the fastest machine for your workload is to do the benchmarking yourself. Whenever I build a new machine I spend a few days benchmarking it versus my other machines. I'm primarily interested in NCVerilog performance and secondarily on Xilinx and Altera performance. I run a suite of Verilog simulations that generally take a few hours to complete. I measure single core performance and then multiple cores. My most recent machine is an iCore7 920 with 12G of RAM, it's the first machine that I've built that really disappointed me. My second most recent machine is a Core2 8400 with 8G. When running on a single core the Core2 is about 10% faster then the iCore7 on a clock for clock basis. However the Core2 can run at a much faster clock then the iCore7. I have both machines overclocked using a Thermalright Ultra 120 extreme. The Core2 is running at 4GHz, the fastest I could get the iCore7 to run is 3.3GHz. In both cases I used sys_basher (available in the Fedora repositories for F10, F11 and F12, and in the EPEL repository for CentOS5) to determine the fastest reliable clock speed for each machine. When you combine the actual clock rate with the per clock NCsim performance the Core2 beats the iCore7 by a substantial margin. On throughput the two machines are about even. The iCore7 has four cores vs two for the Core2. However the undersized cache on the iCore7, 8M/four cores, cripples it so that the two cores on the Core2 do as much work as the four cores on the iCore7. The iCore7 does out perform the Core2 on a clock for clock basis when running Xilinx tools. However the difference is smaller then the difference in actual clock speed so the the Core2 is a little faster even on Xilinx tools.
From: Prevailing over Technology on 10 Jan 2010 16:06
On Jan 7, 9:24 am, Tom Kotwal <tkot...(a)gmail.com> wrote: > Hi All, > > I'm speccing out a new windows PC that I'll use with Xilinx tools, > probably Webpack and Modelsim, and I'm looking for some advice to make > sure the tools will run fast. I know memory is important, but what > else? Also, what pitfalls should I watch out for? > > I'm not sure how relevant this info is, but I'm probably going to > target something in the ballpark of a Virtex-5 LX50. Also, I'm > planning on getting a laptop because I'll need to travel quite a bit > with it. > > Some specific questions: > > - Have people had any problems with Windows 7? 32 bit or 64? Is it > useful to have Win 7 Pro so that I can use XP mode? > - Core 2 Duo vs i7? > - How important is cache size? > - How much memory should I get? Is 4GB enough? > > Thanks! > > -Tom If you're only building an LX50, you should be fine with 4GB. I have a few different machines and have built LX50s on a 2 GB machine. Most new machines that we've procured are i7 based, although not because of Xilinx requirements. Some of our proprietary software takes advantage of the multithreading. The Xilinx tools do no generally but the i7's multi-channel memory allows you to do something else while running the Xilinx tools. We just installed a few Windows 7, 64-bit machines (both Professional and Home Premium) and haven't had any problems to report so far. On Windows Vista Home Premium, we had a few anomalies with the "Clock Report" function. On Windows Vista Home Premium, 64-bit, we saw problems when opening new files. The problem didn't exist on 32-bit machines and the problem didn't show up on 64-bit machines after all the files were established. -- Steve Knapp Prevailing Technology www.prevailing-technology.com |