Message boards :
Number crunching :
AMD's
Message board moderation
Author | Message |
---|---|
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Just noticed there are only 8 AMD based rigs in the Top Hosts 100. Just thought I would point it out. Dave |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Very easy for to explain. SETI@home run faster on Intel than on AMD CPUs. Why? Because the Intel Compiler support mostly the Intel CPUs. This we know, so we buy Intel. It's a pity, but it's the true. 'lunatics.kwsn.net/discussion-forum/..' AMD - max. SSE3 Intel - SSSE3, max. SSE4.1 I would like it, if the compiler would also support the AMD chips (and their extensions) to 100 %. ____________ [Optimized project applications, for to increase your PC performance (double RAC)!][Overview of abbreviations, which are used often in forum and their meaning.] |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
All the 100 are Cuda based rigs and therefore the CPU is used to load the GPU's. Lots of the rigs only run 50 to 75% CPU's, so CPU crunching is mainly on VLAR's. Whilst Intel is better at this type of task there is a premium to pay in cost. It was just an observation. Dave |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I guess the first 100 are mostly 'old' PCs. Only new added nVIDIA GPUs. If you would like to build now a new fast machine, the CPU isn't longer important. The nVIDIA GPUs make the RAC. ;-) ____________ [Optimized project applications, for to increase your PC performance (double RAC)!][Overview of abbreviations, which are used often in forum and their meaning.] |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Possibly older machines but they still have to have PCI Express slots. By the way Sutaru do you have a link to the cc_config switches ? Dave |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Possibly older machines but they still have to have PCI Express slots. all the switches and logging flags are documented here: How do I configure my client using the cc_config.xml file? and here: Client configuration Claggy |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Intel took the crown when they released the core2 generation of CPUs. Before GPU crunching, the floating point performance of the Intel core2 architecture simply blew AMD out of the water. Now that GPU crunching is doing most of the work, there is less advantage for the CPU platform. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Thank's Claggy for that. I seem to recall that FredW turned of 1 of his GPU's with an <ignore_cuda_dev> but I can't see it in the list. I have a faulty GPU on a 295 and whilst I can turn it of in XP using Device Manager Vista 32 will not allow it. Any Idea how to do it. Dave |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Thank's Claggy for that. <ignore_cuda_dev>, <ignore_ati_dev> ignore (don't use) a specific NVIDIA or ATI GPU. You can ignore more than one. New in 6.10.19 Fill in the device number of the GPU you do not want to use. Can be used for multiple GPUs, one line per GPU. Reminder: Zero will disable device zero, it does not disable the line, only removal does. Claggy |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
I am using 6.10.18 thats probably why it didn't work. I put the line <ignore_cuda_dev>1</ignore_cuda_dev> in to disable the second GPU and I get an urecognised tag error. Dave |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
I am using 6.10.18 thats probably why it didn't work. Hi Dave, Worked for me 'cause I am on 6.10.32. I've now disabled that line since I found that clocking the GPU memory down to 750 allowed that weaker GPU to work OK. F. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Hi Fred I just did an update to 6.10.19 to try it out. Will keep you posted, I tried underclocking but no joy. Dave |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
Last time I checked, the stock S@H app doesn't run faster on SSSE3 or SSE4.1 - I think it's just because Intel has manufactured a physically superior chip - and they had a greater market share in the first place. I do hope it stays that way. Take me back to the P4 Prescotts - talk about hot! They should've advertised it as a living room heater. Intel took the crown when they released the core2 generation of CPUs. Fit everything a GPU has into the size of a GPU - it wouldn't be anywhere near as great. - Luke. |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
Strangely enough, the first 3 systems in the top500 list are all Opteron based, and the Los Alamos National Laboratory has just ordered a new 45m dollars machine, "Cielo", based on the Opteron 6100. They must not be that bad. Tullio |
ML1 Send message Joined: 25 Nov 01 Posts: 20291 Credit: 7,508,002 RAC: 20 |
... Take me back to the P4 Prescotts - talk about hot! They should've advertised it as a living room heater. ... Usual legal disclaimer and reminder that the following is merely my uneducated and personal opinion: Intel and AMD have very different design constraints. That very definitely shows in their designs. In brief, Intel has superior fab and packaging available and so can afford to use the sheer brute force of silicon area and number of transistors to push their designs. That can be used either for better performance, or to instead "push a design to market more quickly". Or even, just to extend the life of an existing architecture. I think an extreme and brazen example of "extend the life of an existing architecture" were the plans to add (squander) multi-mega-bytes of FSB-cache onto the Intel Northbridge chips to procrastinate the obsolescence of the Intel FSB for the (old architecture) dual core CPUs at the time... At the moment, Intel appear to allocate a large swathe of silicon to boost the Marketing headline numbers for FLOPs and CPU cache. "Hyperthreading" (HT) is thrown in also to help give perhaps up to a 30% performance boost to overall throughput. However, the real biggie for the Intel HT is that it doubles the number of CPUs that the uneducated public see. I'm sure Marketing make a nice profit out of HT that is way out of proportion to the actual physical performance boost gained. That's all very good for "scientific computing", Boinc and s@h! It's also good for keeping prices high until AMD can catch up on those scores... However, all that is not so useful for servers or even for home users. Meanwhile, AMD has to be much more frugal about silicon area and so is constrained to focus on better overall system design. Whereas the Intel CPUs appear to be very top heavy with lots of FPU and cache resources that are often underutilised, AMD looks to have a much better balance for their designs. That's one of the reasons that AMD have never needed the design complexity and compiler hangups for using hyperthreading. The AMD designs are well balanced in the first place for the most useful tasks. Unfortunately, the public and gamers understand nothing of having a balanced design and instead focus on headline numbers. With various dubious and anti-competitive practices by Intel, that has left AMD chasing the less sexy ends of the market. Hence, AMD does very well in the server market and for supercomputing clusters. AMD also has the market for those who are price-performance conscious. I think we really do need to have Intel's big share meaningfully divided with a third competitor to re-inspire and reinvigorate computer design and development. There's been some very stagnant periods in the past decade. Then again, might various spin-offs from ARM slowly creep up into that area? Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
While working for BULL Italy at Trieste Area Science Park I have used a BULL/MIPS Risc machine which on the Lapack benchmark had the same performance of a DEC VAX 9000 costing ten times as much. The, retired. I used a Pentium I with Solaris and a PII Deschutes with Linux. Now I am using an Opteron 1210, a good chip although it is the lowest clocked Opteron. Tullio |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Unfortunately, the public and gamers understand nothing of having a balanced design and instead focus on headline numbers. With various dubious and anti-competitive practices by Intel, that has left AMD chasing the less sexy ends of the market. Hence, AMD does very well in the server market and for supercomputing clusters. AMD also has the market for those who are price-performance conscious. The very best example of this is clock speed. The newer Intel designs are saner, but look at the P4 at 4 GHz when AMD was down around 2 GHz. The P4 wasn't twice as fast simply because most of the instructions on the AMD part took fewer clocks to do the same work. ... but the public said "wow, look at all those GHz, that's fast!" and that was the end of that. Depends on what you want. On a related subject, I've seen SSE3, SSSE3 and SSE4 mentioned. I really haven't bothered to read up on these extensions, but they all add instructions to the existing x86 instruction set. If the running program does not use a new instruction, then it follows that a "better" instruction won't help. If the application doesn't do the kind of math suited to a new instruction, an optimized app. still won't benefit. More isn't always better, sometimes it's just more. |
Luke Send message Joined: 31 Dec 06 Posts: 2546 Credit: 817,560 RAC: 0 |
Unfortunately, the public and gamers understand nothing of having a balanced design and instead focus on headline numbers. With various dubious and anti-competitive practices by Intel, that has left AMD chasing the less sexy ends of the market. Hence, AMD does very well in the server market and for supercomputing clusters. AMD also has the market for those who are price-performance conscious. Your right. It's called the Megahertz myth I think... or was it the Gigahertz Myth? - Luke. |
ML1 Send message Joined: 25 Nov 01 Posts: 20291 Credit: 7,508,002 RAC: 20 |
... More isn't always better, sometimes it's just more. The uneducated state of (most?) journalists, reviewers, sales assistants and the general public is such that they appear to know no difference between those terms or any of: mhz (mHz): milli-Hertz, as in thousandths of a Hertz as in one cycle per thousand seconds; MHz: Mega-Hertz, million cycles per second; GHz: Giga-Hertz, a thousand million cycles per second (American Billion); mbit/s: milli-bits per second or bits per thousand seconds; kbit/s: kilo-bits per second, a thousand bits per second; Mbit/s: Mega-bits per second; MByte: Mega-Bytes, a million bytes, eight million bits; GByte: Giga-Bytes, a thousand million bytes, or eight thousand million bits; And so on... The units for broadband speed, data rate, clock speed, system memory storage, or disk storage seem to be badly confused. The only thing that sticks out is only the number before the units! Overhearing the bamboozlement of some innocent shoppers by PC shop sales assistants can be gruesome. Then again, the ignorance of some shoppers is just simply painful! It's an uneducated jungle out there! Regards, Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
LoL, I would be more cautious with those ;) Everyone blames HDD manufacturers for similar understanding M/G Byte terms. Actually MByte = 2^20, GByte=2^30. Programmers like to live in 2-states world and these terms were established before general public got access to computers ;) |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.