Message boards :
Number crunching :
amd vs intel
Message board moderation
Author | Message |
---|---|
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
I have been noticing on my completed WU's that it is fairly easy to tell an intel user vs an AMD user. Its all in the credits. An amd user will typically get about 30 claimed credits per WU while the typical intel user gets 18-20. This is consistent all the way down to the duron processors. I have looked at many of my finished WU's and get so disappointed when i get results that are paired up with intel users. I have to assume that something in the BOINC process is hinky since all processors should report back with genenrally the same claimed credit. They don't. on a side note, a smaller portion of users use macs and they all seem to have higher claimed credit than the AMD and intel. |
Tern Send message Joined: 4 Dec 03 Posts: 1122 Credit: 13,376,822 RAC: 44 |
An amd user will typically get about 30 claimed credits per WU while the typical intel user gets 18-20. My AMD average claimed credit is 19 - and that's with an optimized BOINC client. My Macs do vary in average claimed from 28 to 57... but the high-end of that is on a VERY slow machine. My overall average claimed is 25, granted 22. We know the benchmarks are off, for example, the same CPU on Linux will benchmark very differently than on Windows. Thus all the push for using the duration correction factor on credit as well as estimated times, or completely revising the benchmark system. I don't think it's as simple as Mac gets more than AMD gets more than Intel, although you may well see that in some percentage of the cases - there's just too many variables. Believe me, it's being looked at! |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
An amd user will typically get about 30 claimed credits per WU while the typical intel user gets 18-20. Actually, and like Bill said, most, if not all, of my claimed credit will be in the teens. I'm using roughly the equivalent of Bill's processor as well. He's running an AMD Athlon64 3700+ San Diego with a 10% overclock and I'm running an Athlon64 3200+ Winchester with a 25% overclock. Sometimes his extra 512K cache will help, but if you look at results between both of us, they are roughly equivalent... We are also both using optimized science apps. Anyway, from time to time I see Intel processors claiming around the same as my AMD, but almost always I will either get a substantial amount more granted than what was claimed, or it will be slightly higher, like I claim 14 for a unit and get 16... Other times, I'll be the "average" (3-result quorum and mine is the middle result)... Brian |
mcbeth Send message Joined: 12 Mar 03 Posts: 81 Credit: 176,108 RAC: 0 |
What I can say is, I have an P4 3.2GHz whith Hyperthreading. I'm using the optimised seti and boinc clients. My claimed credits are mainly beetween 15 and 18. The granted are normally around 20. so I think they are not so different from the amds from Brian and Bill mcbeth mcbeth |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Hi Thats not true. My claimed credits are around 30 and more but getting granted 20 or less. The most crunchers because thats happend are p4 crunchers and linux users without optimizations. Running Athlon XP 2500+ at 2.2 Ghz. greetz Mike With each crime and every kindness we birth our future. |
mcbeth Send message Joined: 12 Mar 03 Posts: 81 Credit: 176,108 RAC: 0 |
so maybe slower comps claim for more credits? mcbeth mcbeth |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
I have looked at many of my finished WU's and get so disappointed when i get results that are paired up with intel users. Well that should soon change now AMD processors are outselling Intel processors ;) Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
W-K 666 Send message Joined: 18 May 99 Posts: 19075 Credit: 40,757,560 RAC: 67 |
This has been discussed several times. The main problem is the benchmark as it only measures the core of the cpu(s) and not the system. The actual Seti application is greatly affected by the L2 Cache memory, the more the better and new Intel's usually have more. For Intel HT and some? multi cpu machine it divides the benchmark by the number of processors, presumably on the assumption it can do that much more work in the same time, this is a fallacy most reasearch would say a factor of 1.8 at best and 1.5 is not uncommon. The benchmark problem is being looked at but it is run by the BOINC client and therefore has to be acceptable to all projects past, present and future. Using the correction factor in BOINC 5.2.2 as a quick fix in calculating claimed credit is a possible route but I think they possibly want to do more studies and collect more data first. I know that if I multiply the claimed credit on each of my families 5 computers by the correction factor and then divide by 2 the result is near to the average granted credit. Optimising of the BOINC client to increase the benchmark scores means that the benchmark figures have to be taken with a pinch (no a bucket) of salt. Any reasonable study of benchmarking will say you must use exactly the same benchmark on all machines, and as the source is open software and you will get different results for different compilers using the same computer then I guess that cannot happen. The optimised apps bring the claimed credit down because of quicker processing times, and probably only the users of these boards use them, and few of their friends, maybe! But they allow you to do more units per day and therefore your credits/day do go up and you climb the stats ladder quicker. One point though is make sure your computer is running as best as you can make it. The Dual P3 computer on my account when it first started here was not very heathy, it had been running for about 4 years without any major maintenance, hardware or software and as I used it for work and play (two teenage sons) and had gone from Win98SE thru 2000 to XP pro on upgrades. Once I got my new machine I cleaned it, installed win2000 from scratch and installed BOINC Seti and Einstein. I then adjusted voltages and memory timings etc and the claimed credits for Seti went from 50+ on standard app when first run in March, to about 22.5 with optimized app as it is now, and the time to crunch a unit fell from 10 to 14 hrs, to, 4 to 5 hrs. Andy |
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
Just as a FYI, I did have one result left at the bottom of my current results listing where I was using the regular 4.18 non-optimized client. My claimed credit was 28, so yes, I used to be claiming around 30 and was getting usually less. This is probably because, like others have said, the benchmark isn't looking at processors equally/"fairly"/"accurately". As an example, even with the optimized client and optimized BOINC app, it still thinks it's going to take me about 2 hours and 36 minutes per work unit. In reality, it only takes me about 1 hour (55-65 minutes). I think that overall, I'm probably still getting the same amount of GRANTED credits, but I can complete a unit faster, so it is boosting the credit/day for me, which is what WinterKnight said... Brian |
[B@H] Ray Send message Joined: 1 Sep 00 Posts: 485 Credit: 45,275 RAC: 0 |
I have 2 Intel units running, both take about the same time 1:30 to 1:50. One is a Celeron (2.93Gig) with a 256K cache and 764 Megs memory, and claims 28 to 32 creits per WU. On this system I only get one ar two at a time and try to turn them in early to bring the granted credit up for all running the same WU. The other system is a P4 (2.4Gig) with a 512K cache and 512 Megs Memory, and claims 18 to 23 credits per WU. For this system I usually let the WU's age a few days on the system before running, a lot of times higher claiming systems will control how much credit they get that way. Both are runnning the optomizer SETI and BOINC units to keep them about the same. There is a big diferance in other programs also, for Predictor and CPDN the P4 dues them faster. For Rosetta the Celeron uses only half the time as the P4 on the same type on WU. For there 1acf__abrelax units the P4 takes about 4 hours and the celeron about 2 hours. They take advantage of the clock speed and main mamory rather than the L2 cache so they faver the AMD and Celeron processors. On my P4 I will be changeing the memory from 512 Meg's PC2100 to 1 gig PC3200 memory, that should allow it to keep up with the celeron on all programs. The memory comming out of the P4 will go into the celeron in place of a 256 meg chip to bring that up to 1 Gig also. To bad that we can't inexpensively increase the L2 cache on a CPU that is in use like we can the system memory, that would make them all run faster. Cheers Ray Pizza@Home Rays Place Rays place Forums |
ewells2420 Send message Joined: 24 Jul 05 Posts: 9 Credit: 25,431 RAC: 0 |
on avg its taking my Intel pent D processor 2.8 gig with 512mb ram and 1mb cache an hour to do every WU it gets. im using Trux's opted boinc client and TMR's opted seti app. oh yeah the evil part .... it runs 2 WU's at same time always =P I went throught about 15 WU's last night. |
[B@H] Ray Send message Joined: 1 Sep 00 Posts: 485 Credit: 45,275 RAC: 0 |
on avg its taking my Intel pent D processor 2.8 gig with 512mb ram and 1mb cache an hour to do every WU it gets. im using Trux's opted boinc client and TMR's opted seti app. oh yeah the evil part .... it runs 2 WU's at same time always =P I went throught about 15 WU's last night. If you are running other programs as well as SETI you may want to increase the memory, with 2 units running and one being Predictor or Rosetta everything will slow as they use a lot of memory for each. SETI will run good with the 512 if it is the only program running. That would really scream if it had HT on both processors in the Pent D. 4 at a time would be great. Would be good it they made a board that would hold 2 Pentium (or Pentium D) processors, that would cost a lot less than a 2nd system. Pizza@Home Rays Place Rays place Forums |
ewells2420 Send message Joined: 24 Jul 05 Posts: 9 Credit: 25,431 RAC: 0 |
*tries to imagine 2 dual CORE processors on one MB* thats like having 4 processors on one mobo. You would be able to run 4 seti WU's at same time if thats all your running, which is all i run right now. |
[B@H] Ray Send message Joined: 1 Sep 00 Posts: 485 Credit: 45,275 RAC: 0 |
*tries to imagine 2 dual CORE processors on one MB* thats like having 4 processors on one mobo. You would be able to run 4 seti WU's at same time if thats all your running, which is all i run right now. Bsides running 4 at a time, think of the cost savings over 2 systems, only one of most other things. And a lot less power used compaired to 2 or 4 systems, but they don't build that board. Well they do build some for the zenon cpu's Pizza@Home Rays Place Rays place Forums |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
I forgot to mention i am running an AMD anthlon64 3200 with 1gb ram |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
ok i just ran through the first 3 pages of results and consistantly got AMD processors with higher claimed credits vs intel. Only 1 intel p3 scored higher than the amds. I would love to provide links to all these WU's but it seems pointless at this time. People seem to be more concerned with finishing the WU that the actual credit gained. I simply cant fathom why the obvious differences arent making more people wonder what the heck is going on with the process itself. Since this is a scientific endeavor it would seem logical that the folks at seti would take a look at the #'s and compare avg. claimed credits for amd vs intels vs macs Personally, I already know that both my AMD boxes claim similar credits while the "optimized" or plain intel claimed credits are much lower, sometimes even half the credit. THis doesn't seem optimized to me and even seems that something is actually going wrong with the computations in some way. Not sure why everyone worrys about the speed they comlete a WU if that WU is done poorly or underclaims credits. It would seem that the processes that are optimized aren't generating more credit for the time spent at all. Lower claimed credit/faster finishing speed vs normal speed and higher credit. |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Quite simple: Claimed credit is proportional to benchmark score times CPU time needed for a WU. It has no relation to the "quality" of a result. Athlons simply have higher benchmark scores while taking longer to complete a WU. This results in higher claimed credit. Usually 3 WU are used to determine granted credit, the highest and the lowest is thrown out. This leads to granted credit being much more even than claimed credit when comparing CPUs. Regards Hans |
Astro Send message Joined: 16 Apr 02 Posts: 8026 Credit: 600,015 RAC: 0 |
Oh boy, where to start. I use an optimized application (science app) and a standard boinc client (manager). This means that I'm only claiming on average 12.48 credits/wu for my AMD64 3700, something like 15 for my AMD athlonXP2200+, and about the same for my P4 1.8. However I'm being granted approx 22 credits/wu. I don't have the exact figure but I looked at 400+ wus and claimed like 2900 credits, but was granted 4800. I know I'm claiming less per wu, but I'm doing almost twice as many as I would with a standard science app. If I used the standard app, then I'd be claiming roughly what I was getting and only being granted in the end approx 1/2 as much credit in total. The optimized science apps are known by the developers and they seem to be fine with them, and who am I to argue with them, it's their project. If you read the boards where people are discussing their own optimization attempts you can tell they care about getting the numbers right. Each CPU/machine crunches numbers a little differently given the same formula. Example, An AMD might complete a calculation and come up with 18.91234567, where the Intel might come up with 18.91234569, not a big difference, but different. Some round to different places as well. It's up to the project directors to determine the accuracy required by themselves. hope this helps tony |
W-K 666 Send message Joined: 18 May 99 Posts: 19075 Credit: 40,757,560 RAC: 67 |
ok i just ran through the first 3 pages of results and consistantly got AMD processors with higher claimed credits vs intel. Only 1 intel p3 scored higher than the amds. I would love to provide links to all these WU's but it seems pointless at this time. People seem to be more concerned with finishing the WU that the actual credit gained. I simply cant fathom why the obvious differences arent making more people wonder what the heck is going on with the process itself. Since this is a scientific endeavor it would seem logical that the folks at seti would take a look at the #'s and compare avg. claimed credits for amd vs intels vs macs Personally, I already know that both my AMD boxes claim similar credits while the "optimized" or plain intel claimed credits are much lower, sometimes even half the credit. THis doesn't seem optimized to me and even seems that something is actually going wrong with the computations in some way. It's quite simple really. Your credits and RAC is based on Granted not claimed credits, With the 'averageing' Granted credits are in the range 20 to 25 credits, normally, so what matters is number of units you can do in a unit of time. My Pent M (claimed 11 to 15) granted averages about 22 would do about 6 to 8 Seti units/day using standard application but on Optimised can do 16 to 18 per day. So If I did:
|
Brian Silvers Send message Joined: 11 Jun 99 Posts: 1681 Credit: 492,052 RAC: 0 |
Athlons simply have higher benchmark scores while taking longer to As a FYI... My non-optimized setup was getting claimed 25-35... That was using an AMD Athlon64 3200+ overclocked to 2.5GHz. It took roughly 7500-8000 seconds per unit. TMR-optimized client with non-optimized BOINC was getting claimed amounts of 11 to 16. Throw in TMR-optimized BOINC and claimed amount went up to around 14-18. Now I'm on Trux 5.3.1 BOINC and TMR 8.1 Seti client. Claimed credit is now up to 17-23... On all of the above, my "average" time in seconds (just a visual average, I haven't actually calculated the average) was 3700... I've bumped up to an A64 3700+ @ 2.6GHz now, and my times are now down to 3400 (visual average, not computed), which typically SMOKES most Intel processors. Meanwhile, I was still getting granted credit in the upper teens, to sometimes 20s, and even a few low-30s. If anything, prior to me switching to the Trux BOINC app, my AMD was so fast and was claiming so little that I was probably dragging down the average, which is what I think Skildude is getting at... Bottom line, from my lurking in these forums in the past, there is still a lot of confusion in regards to the credit system, which is what prompted this thread. As I said earlier, my faster completion times allowed me to finish units faster, which helped offset the drop in the Claimed Credit. Claimed 2 X 14 = 28, which is in the range of where I was at before the non-optimized seti app... Granted credit was also probably nearly the same, at about 18-25 per work unit. Prior to the optimized client, I would routinely get LESS than what I claimed... Now just to totally blow Skildude's mind (hehe), how do I know what the "right amount" of claimed credit is supposed to be anyway? :shrug: Brian |
©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.