Message boards : Number crunching : Credit always low
Author | Message |
---|---|
Hammeh Send message Joined: 11 Nov 08 Posts: 63 Credit: 211,283 RAC: 0 |
Hey!! Lately, I have started my computers crunching Rosetta@home again. However, after reviewing the completed tasks, I find that my main quad core computer is constantly being granted less credit than it is claiming whilst my dual core one is the opposite! Is there any reason for this? |
Mod.Sense Volunteer moderator Send message Joined: 22 Aug 06 Posts: 4018 Credit: 0 RAC: 0 |
Looks like the Intel has a 2MB L2 cache as well, for 2 active processes. That is 4x larger then the AMD which is trying to run 4 active processes. Experience has shown that Rosetta workloads tend to run better when large L2 cache available. This is probably due to the relatively large program size. Credit is based on actual models completed. There is some variation from task to task and model to model as to how many operations are required to complete a given model, so the numbers are really most meaningful when averaged over a period of time (which you seem to have done, I'm just trying to be specific for future readers). Bottom line is that the raw CPU speed of the AMD is faster, and so the claims are high, but the actual ability to manage memory and complete work is not keeping up with the Intel. You are witnessing part of why it is very difficult to compare CPUs to one another. And why one vendor perfers to reference one benchmark while another vendor will always use another, some benchmarks run better with large L2 cache, others do more floating point operations then integer, etc. Rosetta Moderator: Mod.Sense |
Sid Celery Send message Joined: 11 Feb 08 Posts: 2073 Credit: 40,602,258 RAC: 5,342 |
The main difference is that your quad core is AMD, and your dual core is Intel (Core 2 duo). In my opinion. Your Intel is better suited for Rosetta and is therefore granting more credit, relatively. The Intel might be generating more models per task and for that reason the granted credit might be higher. The only reason the AMD has a higher RAC might be because it has twice as many cores. I'm in exactly the same configuration and can confirm the same results. But the dual core Intel (2.2GHz) has a total RAC of ~750 and the quad core AMD ~1300 (2.5GHz), so it's significant but not especially so. In all other respects the W7 laptop is a pain while Vista desktop very smooth. Boinc is its saving grace. It's all incidental to the results we're seeing here at Rosetta, especially right now. |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
I have got the impression, that the granted credits are just based on the CPU-type. My i7 920@3.6 GHz is claiming ~70 credits and gets ~40 credits granted. I have looked up some non-oced i7 920, they are getting granted the same ~40 credits, although my machine is oced by 30 percent... Actually, this is the reason, why I don't use my i7 for Rosetta very much. Joe |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
|
dcdc Send message Joined: 3 Nov 05 Posts: 1830 Credit: 119,208,549 RAC: 2,517 |
I have got the impression, that the granted credits are just based on the CPU-type. My i7 920@3.6 GHz is claiming ~70 credits and gets ~40 credits granted. I have looked up some non-oced i7 920, they are getting granted the same ~40 credits, although my machine is oced by 30 percent... Actually, this is the reason, why I don't use my i7 for Rosetta very much. |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
Credits are granted based on the actual work done and faster machines receive more credit than slower ones accordingly to their actual contribution. Well, in this case, I do not understand, why my 920@3.6 does not get more credits than a 920@2.67? Jochen |
mikey Send message Joined: 5 Jan 06 Posts: 1895 Credit: 8,922,080 RAC: 1,112 |
Credits are granted based on the actual work done and faster machines receive more credit than slower ones accordingly to their actual contribution. From the link below: "Credit shall be granted by the actual work done from the host. In order to achieve that we grant credit based on the number of models (decoys) completed for each work unit. This relates exactly to the actual scientific contribution. For each type of WU we determine a credit/model ratio which we will grant as "work credit". A faster machine which completes more models in the same time will receive more credits." Unfortunately credits cannot be compared directly to another machine over the short run, over the long term yes they can, but the long term may be weeks or even months. This has always been, and probably always be, a bone of contention but it is what it is. There are other variables too, do you crunch for other projects? Is a hard drive failing on one machine and writing slowly? Is one piece of ram in one machine flaky? Waaay too many variables involved from machine to machine, and we haven't even talked about the different units that get crunched. A very long time ago a standard unit for crunching was discussed, and thrown out, that would have been sent to every machine that would set the standard for what credits that machine gets. It was thrown out because the Developers thought people wanted to get credit from crunching, not running the same old unit week after week after week. |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
I had this rig running Rosetta for at least 3 months last year... I would consider this a long term. My rig is claiming the correct amount of credits, but just gets granted, what the same CPU@stock-clock is granted. It looks like, there is a table with CPU-types and WU-types to look up the granted credit - and this does not take oced rigs into consideration. I am sorry for my poor english, I can not explain it any better. Jochen |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
I had this rig running Rosetta for at least 3 months last year... I would consider this a long term. My rig is claiming the correct amount of credits, but just gets granted, what the same CPU@stock-clock is granted. There is no table for how much credit a CPU-type should get. Let your i7 run and compare your Average Credit for that machine with a machine with the same CPU. Let your average credit to settle (1-2 weeks)... and then compare. |
dcdc Send message Joined: 3 Nov 05 Posts: 1830 Credit: 119,208,549 RAC: 2,517 |
I had this rig running Rosetta for at least 3 months last year... I would consider this a long term. My rig is claiming the correct amount of credits, but just gets granted, what the same CPU@stock-clock is granted. Hi Jochen Chilean is right - you're looking at the wrong numbers. A difference between claimed and granted credit is irrelevant - you should be looking at credit per hour or RAC, which will both show that your overclocked machine gets more credit than your stock-speed machine. HTH Danny |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
Thanks for all your answers. I will wait. There were not many 920s around last year, so I could not compare the RAC. But still, looking at the granted credits for similar WUs on a stock 920 and on my oced 920... We will see in a couple of weeks. Jochen |
dcdc Send message Joined: 3 Nov 05 Posts: 1830 Credit: 119,208,549 RAC: 2,517 |
Thanks for all your answers. I will wait. There were not many 920s around last year, so I could not compare the RAC. But still, looking at the granted credits for similar WUs on a stock 920 and on my oced 920... We will see in a couple of weeks. WUs are made up of a varying numbers of decoys (models) - the number depends on how many your machine can fit into your WU time preference. Your overclocked machine might be fitting 30% more decoys into each WU, which will result in it getting 30% more granted credit on average. The claimed credit is just based on the computer's benchmark scores x WU time so is often inaccurate. Danny |
mikey Send message Joined: 5 Jan 06 Posts: 1895 Credit: 8,922,080 RAC: 1,112 |
Thanks for all your answers. I will wait. There were not many 920s around last year, so I could not compare the RAC. But still, looking at the granted credits for similar WUs on a stock 920 and on my oced 920... We will see in a couple of weeks. You cannot compare your 920 to someone elses 920, their motherboard could be more efficient, ie more expensive, their memory could be faster, their harddrive could be faster, they could be using their video card to crunch with and you are not, you are playing games sometimes and they are not, etc ,etc, etc. Yes it would be nice if you could find another identical machine that that person does exactly the same thing you do, but in real life you cannot. I have a bunch of computers on line that only crunch, nothing else, that makes them faster than any other pc like them that their user does something, anything, with besides crunching. I also have my cache set fairly low, less than a day, that reduces the flucuations caused by returning units once every few days or even once a week or more. In short there are MANY variables that will cause even 2 identical machines to have a different RAC than each other, some of which we can control, some of which we cannot. Do you have the screen saver running? That can slow down the pc too! |
Mod.Sense Volunteer moderator Send message Joined: 22 Aug 06 Posts: 4018 Credit: 0 RAC: 0 |
Another key reason it is not meaningful to compare RAC of one machine to that of another participant is that you have no idea how many hours per day their machine is on, nor how many projects they are running. They could easily have a 8 core machine running one climate prediction model and 7 Rosetta tasks for example. They might have powered down for the night when they installed Windows updates, etc. It seems one concern is that your overclocking is not recognized. This is not the case. A successfully OC'd machine will claim more credit per hour, and produce more models and therefore more granted credit per hour then the same machine running the same tasks without the OC. But note that the granted credit may still be consistently below the claim, even though the credit granted per hour has indeed increased. Rosetta Moderator: Mod.Sense |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
A successfully OC'd machine will claim more credit per hour, and produce more models and therefore more granted credit per hour then the same machine running the same tasks without the OC. Is there a guide how to calculate this credit per hour value? I did a quick forum search, but could not find something. Jochen |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
A successfully OC'd machine will claim more credit per hour, and produce more models and therefore more granted credit per hour then the same machine running the same tasks without the OC. BOINC Manager shows you an X-Y graph for the total credit of your machine, just pick two point (1 day) and divide by 24. Best way to figure it out IMO. |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
What about taking the granted credit of 100 WUs and calculate the credits per hour for each workunit and use these numbers to calculate an average? Jochen |
mikey Send message Joined: 5 Jan 06 Posts: 1895 Credit: 8,922,080 RAC: 1,112 |
What about taking the granted credit of 100 WUs and calculate the credits per hour for each workunit and use these numbers to calculate an average? You can do that, that is sort of what Boinc does now, but what are you going to compare it too? The only thing that shows you is the trend on your own machine and as we have all pointed out, your machine cannot be compared to any other machine in any kind of meaningful way. Yes your 920 will be faster than my P3 or P4 but when trying to compare your machine to anything else there are just too many unknown variables. I have seen an old IBM pc XT overclocked from 4.77mhz to over 1ghz! It was the fastest XT on the planet, but could only do 1 thing, add numbers. And by that I mean 1 + 1 = 2 + 1 = 3, etc, etc ,etc. It was not comparable to any other XT at the time because it didn't do the same thing any other XT did! |
Jochen Send message Joined: 6 Jun 06 Posts: 133 Credit: 3,847,433 RAC: 0 |
I did the above for my 920 and 3 other 920s: my 920: 17,7409880459578 Computer-Id 1143608: 17,4120945379865 Computer-Id 1141244: 17,0467817199461 Computer-Id 1159340: 18,0163583022697 These numbers are actually credits per hour per core. My 920 and the last 920 seem to oced to the same level, the other two do not seem to be oced (or they are using optmizted BOINC versions). Looking at these numbers, it does not look like overclocking does make any difference. I still doubt very much, that it would make any difference, if I would rum my computer at stock clocks. And just do not try to tell me, my computer is not overclocked correctly. Jochen |
Message boards :
Number crunching :
Credit always low
©2024 University of Washington
https://www.bakerlab.org