Message boards : Number crunching : Much more powerful cpu but far fewer granted credit
Author | Message |
---|---|
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. In practice almost all cpus with the same features like E4x00 or T7100/T7250 (mobile), get many more credit Even T2x00 processors with 1Mb cache L2 do better. I've tried everything: increasing FSB, increasing Minirosetta tasks priority, lowering memory timings, disabling other programs, adjusting targer cpu run time, moving from XP to Vista all of this, but nothing changes The granted credits still remain lower by 20-30% compared with the claimed a friend of my team with a T7250 cpu running at 2 ghz is getting many more credit, about 20-30% over his claimed so where is the problem? |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,675,695 RAC: 11,002 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. Ignore the claimed credit. The only thing that is important is granted credit / time. HTH Danny |
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. I know that since the time I joined the project in 2007. The question is simple: how is it possible that a T7250 at 2Ghz (my friend's netbook) get many more credits/time compared with my E5200 running at 3.2 Ghz with 4Gb at 900+Mhz 4-4-4-10 it is very clear that there is something wrong somewhere... |
Greg_BE Send message Joined: 30 May 06 Posts: 5691 Credit: 5,859,226 RAC: 0 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. try comparing the numbers on the caches and other features of the two cpu's the t7250 is at: http://www.intel.com/products/processor_number/chart/core2duo.htm and has 65 nm 2MB L2 2.00 GHz 800 MHz the e5200 is here: http://www.intel.com/products/processor_number/chart/pentium_dual-core.htm and has 45 nm 2MB L2 2.50 GHz 800 MHz it is possible that it is pentium technology vs core2duo technology that makes the difference as the L2 numbers are the same. i think that this page at intel may give you some insight. that's just what i have found at a quick glance. |
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
The fact that the L2 cache is essential for Rosetta had understood for a long time (since I had an Athlon X2 with 2x512Kb) and in fact I saw that the dual channel does not increase performance, a clear sign that the bandwidth between cpu and memory isn't a bottleneck even lowering the ram timings doesn't bring any performance's increase but then how do you explain that E2xx with 1Mb and almost all Intel 2Mb that I've seen (including the T7250 of my friend) are getting many more credits/hour? It can't be normal |
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
Core 2 Duo and Pentium Dual Core are commercial names, technically the only differences are the FSB's frequency, the cache L2 size and the silicon's quality All benchmarks in the net show that given the same frequency, FSB and cache's size, there is no difference in performance between Core 2 processors and Pentium Dual Core |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,675,695 RAC: 11,002 |
Your machine gets 15.5 credits per hour per core from one page of your results. My Q6600 (2.4GHz) gets something in the region of 16.1-20.5 credits per core per hour, but that has 2x the cache per core. I've got an E2180 that's got 1/2 the cache per core which gets 9.8 credits per core-hour. Taking the speed difference into consideration, all are multiplied up to 3.2GHz: Q6600 (2MB/core): 21.5-27.3 E4300 (1MB/core): 21.0 E2180 (0.5MB/core): 15.6 E5200 (1MB/core): 15.5 It looks a bit low for 3.2GHz - is your machine prime95 stable? Is speedstep kicking in? |
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
Your machine gets 15.5 credits per hour per core from one page of your results. My Q6600 (2.4GHz) gets something in the region of 16.1-20.5 credits per core per hour, but that has 2x the cache per core. I've got an E2180 that's got 1/2 the cache per core which gets 9.8 credits per core-hour. Taking the speed difference into consideration, all are multiplied up to 3.2GHz: I've tested the system with Orthos blend test that is heavier than Rosetta and it results stable, in fact I haven't compute errors or failed WUs from Win task manager and CPU-Z the processor is always at full frequency |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,675,695 RAC: 11,002 |
Is there something you're running that's hogging the cache? I've got an OC'd E2180 (2.66GHz) that's getting less credit that i'd expect - I might get a chance to play with it this weekend to find out why. |
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
Is there something you're running that's hogging the cache? I've got an OC'd E2180 (2.66GHz) that's getting less credit that i'd expect - I might get a chance to play with it this weekend to find out why. I suspected that eMule and uTorrent should penalize Rosetta's perfomance but I've tried some WUs with only BOINC, Opera Browser, Comodo Firewall, Avira antivir and Peerguardian2 and nothing changes Maybe I try to shut down that programs for a while... another thing that comes to mind is the fact that I always keep many tabs opened on Opera |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,675,695 RAC: 11,002 |
Is there something you're running that's hogging the cache? I've got an OC'd E2180 (2.66GHz) that's getting less credit that i'd expect - I might get a chance to play with it this weekend to find out why. Process explorer will probably be able to show you if any of those programs are responsible... |
Paul Send message Joined: 29 Oct 05 Posts: 193 Credit: 66,422,060 RAC: 9,629 |
If we assume you are running Windows, you must go into system monitor and look at your processes and find the CPU hog. With WindowsXP, get to system monitor running by hitting <CTRL><Alt><Del>. Click the processes tab. You will see all of the processes listed with the username, CPU% and memory consumption. If you click on CPU, it will sort all of the process by CPU usage. Click twice to sort from highest percentage to lowest. The ideal is to have Rosetta @ the top with the highest CPU utilization. One of my Windows systems consistently listed SVHOST as the highest process. Google svhost.exe and you will find ways to reduce amount of CPU this process consumes. Good Luck! Thx! Paul |
dcdc Send message Joined: 3 Nov 05 Posts: 1832 Credit: 119,675,695 RAC: 11,002 |
If we assume you are running Windows, you must go into system monitor and look at your processes and find the CPU hog. With WindowsXP, get to system monitor running by hitting <CTRL><Alt><Del>. Click the processes tab. You will see all of the processes listed with the username, CPU% and memory consumption. If you click on CPU, it will sort all of the process by CPU usage. Click twice to sort from highest percentage to lowest. The ideal is to have Rosetta @ the top with the highest CPU utilization. to add to Paul's suggestion, in Task Manager go: Processes > View > Select Columns > CPU Time Then you can sort by that column to see which process has used the most CPU time. HTH Danny |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. Probably because the TXXXX are Intel core 2. I haven't looked at the L2 Cache, but I'm willing to bet that your friend's laptop has more L2 cache than your PC. |
Greg_BE Send message Joined: 30 May 06 Posts: 5691 Credit: 5,859,226 RAC: 0 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. not quite true chilean, i already looked up the two cpu's he was refereing to. see here for the comparison. only difference i see is the nm architecture. |
Chilean Send message Joined: 16 Oct 05 Posts: 711 Credit: 26,694,507 RAC: 0 |
I can't understand why my pc is getting so far fewer granted credit compared to other computers equipped with cpus whose perfomances are at least equal to mine (Intel E5200 at 3.2 Ghz with 4 Gb DDR2-800 4-4-4-10) but in most cases lower than mine. hmmm... I would've lost 50 bucks right there. |
GT82 [HWU] Send message Joined: 26 Aug 07 Posts: 15 Credit: 154,103 RAC: 0 |
Thanks for the answers. Now I've raised the FSB to 320 (3200=320x10), which with the previous 5.98 WUs had not led to increases in performance also I turned off EIST and C1E in bios things seem to go much better, with Ralph's 1.43 I'm getting 20.5 credits per core-hour and even with the last 1.40 WUs crunched the results were higher Since currently the system has only one stick of memory I think that with another one (and dual channel) it could reach 22 credits per core-hour (+ 8% from my previous calculations). I will keep updated ;) |
Message boards :
Number crunching :
Much more powerful cpu but far fewer granted credit
©2024 University of Washington
https://www.bakerlab.org