BOINC Benchmarks

[Wayward_Son]

I was browsing through the top Rosetta computers (sorry, Mudd, looks like you got bumped to #3) to find other hex core AMD procs and see what kind of RAC they pull down. I noticed something that I think is strange.

Here is the top AMD processor in the project, a 1090T like mine with four gigs of RAM:

http://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1355806

Drystone: 3172.78
Whetstone: 10030.4

RAC: 4213.33

Here is the second AMD proc, a 1055T with eight gigs:

http://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=1316967

Drystone: 3254.62
Whetstone: 16762.51

RAC: 3768.22

And here are my specs, with my chip running at 4.0, RAM running at 1333 @ 7-7-7-24-1T, NB running at 2600 and HT Link running at 2400:

Drystone: 4116.22
Whetstone: 9510.92

RAC: 1613.57


Here's what I think is strange: My Drystone score, whatever that is, is much higher than both processors. My Whetstone, though, is lower than the first computer and WAY lower than the second. What, exactly, are these two scores measuring? Is the 1055T's Whetstone score due to his extra RAM?

Also, I've been running BOINC almost non-stop for a week now (took 12 hours off to try something new on my overclock) and I kind of expected my RAC to top out by now. Clearly I have a long way to go yet if these two other AMD procs are any indication.

[dejo]

your oc may not be stable and have some failing work units. just a guess. I find the same thing with the intel stuff. but am a rookie with rosetta

[Wayward_Son]

I have had a few failed work units but only a few, less than ten out of 517 results so far. And none since I ironed out a few glitches in my overclock (I tried running it for a couple days at a Prime-stable 4.1, but I was getting failed work units so I backed it down to 4.0 for 24/7 crunching).

Anyway, the failed units will affect RAC but I don't see how they could affect the BOINC processor benchmarks.

[dejo]

how long has the box been running? that has a huge impact on rac. if not close to a month it will still be climbing

[Wayward_Son]

Oh, only a week so far. One week today I think. I figured my RAC was still climbing, I'm more curious about my Whetstone score being lower than the other two systems.

[dejo]

I think whetstone is raw integer math, drystone is floating point. It does seem odd how some box's score compared to others. but I have an intel box that is still climbing after close to a month. with only 1 week you score is less than half of what it will be for sure, as the calculation is based on at least 2 weeks worth of data. and starts at 0 you the math kills your rac. it will come up a bunch, just give it time.

[bmwbaxter]

sorry for poking in here.... so your saying that it can take up to a month for your RAC to max out?

[dejo]

I have one that is still climbing and been close to a month. it will get close after about 3 weeks but can still climb quite a bit from there.

[bmwbaxter]

I have one that is still climbing and been close to a month. it will get close after about 3 weeks but can still climb quite a bit from there.

sweet. in a week i have went from 40th to 15th. and still have 2 more computers to bring online. although i will be retiring my 2 slower machines. too much money to run 6 machines.

[dejo]

according to rosetta, it takes an average of 2 weeks to come up with your rac. so after just one week you are getting an average of way less than half, as the first week cant really be an average because it is being averaged as it is running. after 2 weeks, the first week was still starting an average. after 3 weeks you will get a pretty good idea of where it will fall, but will still be climbing

I have a i5 661 that is pretty much topped out at 2000ppd, also have an i7 950 that is still climbing slowly now at 4500ppd

[bmwbaxter]

according to rosetta, it takes an average of 2 weeks to come up with your rac. so after just one week you are getting an average of way less than half, as the first week cant really be an average because it is being averaged as it is running. after 2 weeks, the first week was still starting an average. after 3 weeks you will get a pretty good idea of where it will fall, but will still be climbing

I have a i5 661 that is pretty much topped out at 2000ppd, also have an i7 950 that is still climbing slowly now at 4500ppd

nice. i am gonna be running my Sig Rig, a core 2 quad and an Phenom x4. so i will be running a total of 16 tasks 24/7 hoping to be able to OC the core 2 quad and the x4 to around 3.2GHz.

[dejo]

should get you some ppd

[Wayward_Son]

What is ppd?

[dejo]

ppd = rac or points per day

[muddocktor]

Hmm, shows me back at #2 again, by about 25 rac. And that machine that's trying to bump me off is dual hex core, so I don't feel too bad that he's running at around my rac, with me having half his processing power. :D I guess I'll have to look at bumping my overclock up a couple hundred megahertz once I get her transplanted into my main system when I get in from the rig. ;)

[dejo]

cant have you sitting at #3 can we Jim. hehe. that machine is spanking some medical research. how you been. and how is the 875k doing for you.

[muddocktor]

I'm doing ok, but I've been out on the rig since 12/22 and won't be coming in until Tuesday. So my Christmas and New Years are crap. :cry: But I do have an eVGA 570 waiting for me to get in and play with though. :D I treated myself, since I was getting screwed on the Holidays.

The 875k is presently on autopilot running Rosetta unattended right now. I have it at 4.2 IIRC while crunching. I've been playing around with it some and it's a tricky beast once you get past 4.2. So I plan to do some more playing around with it when I get in and see if bclk is giving me some fits by lowering it back down to around 180 or so and upping the multi some more. I was able to get 4260 relatively stable but getting to 4.3 long enough to run some benches was really tricky.

[Wayward_Son]

So any ideas on the benchmarks? I'm curious about the Whetstone speed of those two rigs I linked and why they're higher (significantly higher in the case of that 1055T) than mine.

[bmwbaxter]

should get you some ppd

thats what i am hoping for. just trying to catch up to the guys who are running a 980x :drool:

[muddocktor]

So any ideas on the benchmarks? I'm curious about the Whetstone speed of those two rigs I linked and why they're higher (significantly higher in the case of that 1055T) than mine.

I really don't know about that. I have seen some systems very close to mine in configuration that give much lower benchmark scores on either whetstone or drystone for some reason. And my 875k seems to give better numbers than my 930, but then it is also overclocked 200 MHz more too.

Here's what my rigs are showing:

875k @ 4.2, Win7 HP 64
Drystone: 3811.13
Whetstone: 13566.42

930 @ 4.0, Vista Ult. 64
Drystone: 3665.25
Whetstone: 11661.26

980X @ 4.0, Win7 Bus 64
Drystone: 3644.45
Whetstone: 11638.41

QX9650 @ 4.0, Win7 HP64
Drystone: 4087.62
Whetstone: 11848.33

940BE @ 3.6 (I think), WinXP Pro
Drystone: 3272.63
Whetstone: 8152.96

Q6600 @ 3.7, Win7 HP32 (retired for now)
Drystone: 3656.95
Whetstone: 7476.25

T9600 @ 2.8, Vista HP32
Drystone: 2902.14
Whetstone: 6085.07

So you see that you can take those benchmark numbers with a large grain of salt. Because the 875k shows higher numbers than my 980X, which just smokes these work units.

[bmwbaxter]

875k
Drystone: 3811.13
Whetstone: 13566.42

whats your 875k overclocked to?
my sig rig only gets like:
Drystone)3500~
Whetstone)8200~

[dejo]

I think he said it is running at 4.2

[Bobnova]

My 980x hasn't been heard from in the last three days, donno what happened to it. I'll find out sunday when I get home.

On the benchmarks: Running linux inflates the second score hugely. Doesn't have any effect on PPD, i think it's just for them to get a guess on WU length before you've completed enough to have a statistical model.

[muddocktor]

I just edited in the speeds and OS info into my previous post. Just to give some more info on the setups on my systems that are crunching.

[bmwbaxter]

I just edited in the speeds and OS info into my previous post. Just to give some more info on the setups on my systems that are crunching.

well i just upped my OC on my sig rig and have noticed a decent increase in the boinc benchmarks(more so in my whetstones), so once i have stress tested my OC i should get a few more RAC :)

[bmwbaxter]

I really don't know about that. I have seen some systems very close to mine in configuration that give much lower benchmark scores on either whetstone or drystone for some reason. And my 875k seems to give better numbers than my 930, but then it is also overclocked 200 MHz more too.

Here's what my rigs are showing:

875k @ 4.2, Win7 HP 64
Drystone: 3811.13
Whetstone: 13566.42

930 @ 4.0, Vista Ult. 64
Drystone: 3665.25
Whetstone: 11661.26

980X @ 4.0, Win7 Bus 64
Drystone: 3644.45
Whetstone: 11638.41

QX9650 @ 4.0, Win7 HP64
Drystone: 4087.62
Whetstone: 11848.33

940BE @ 3.6 (I think), WinXP Pro
Drystone: 3272.63
Whetstone: 8152.96

Q6600 @ 3.7, Win7 HP32 (retired for now)
Drystone: 3656.95
Whetstone: 7476.25

T9600 @ 2.8, Vista HP32
Drystone: 2902.14
Whetstone: 6085.07

So you see that you can take those benchmark numbers with a large grain of salt. Because the 875k shows higher numbers than my 980X, which just smokes these work units.

are you sure that you don't have your whetstones and dhrystones reversed?


Also, with my new OC (4GHz) I am now getting:
Whetstones= 3820
Dhrystone=9181