I was just looking at your results

3800 Results (124)
8x WUs Still pending
1x dreaded 58.7

4600 Results (120)
12x WUs still pending
3x dreaded 58.7s

It also looks like the 4600 got a few noisy results and had a few abnormal claims...this could contribute to the difference in numbers

-citroja

I would like to know more about reading results....
I assume you go to a particular computer and look under results.
Where do you see the 58.7 designation?
What do you mean/see as moisy and abnormal claims?

the 58.7 (or 58.69) is the claimed/granted credit...the reason that they are considered "dreaded" or "killer" is that for credit/hour is so much lower than those of other results that actually take longer to crunch. As a result the RAC for the host that crunches these WUs suffers noticeably!

there are more detials in this thread

-citroja

---

My AMD 3800+ SIngle scores a 16.658333 Credit per HOur :)

I was just looking at your results

3800 Results (124)
8x WUs Still pending
1x dreaded 58.7

4600 Results (120)
12x WUs still pending
3x dreaded 58.7s

It also looks like the 4600 got a few noisy results and had a few abnormal claims...this could contribute to the difference in numbers

-citroja

I would like to know more about reading results....
I assume you go to a particular computer and look under results.
Where do you see the 58.7 designation?
What do you mean/see as moisy and abnormal claims?

---

I was just looking at your results

3800 Results (124)
8x WUs Still pending
1x dreaded 58.7

4600 Results (120)
12x WUs still pending
3x dreaded 58.7s

It also looks like the 4600 got a few noisy results and had a few abnormal claims...this could contribute to the difference in numbers

The other thing (from BOINC wiki)...
Right after joining a BOINC Powered Project and getting the first credits granted are working against a very short time, and slightly odd things can happen to the numbers. The odd numbers for Recent Average Credit at the beginning of a Participants tenure at a Project are strictly an artifact of the equation used to generate them and are the result of the extremely short time span and the lack of previous data

I put this in there because the age of the 4600 is less than 30 days old...which means that it falls into the "odd numbers" category...I would wait a week or so more to let the 4600 complete a few more WUs before I would start to go into any kinda of panic mode. It is good that you are concerned but I would still wait it out.

-citroja

---

Okay,

taskmanager shows only boinc running 50/50 for each cpu, total 100% system usage.

Have fiddled with virtual memory a couple of times, can't remember exactly where it started so I'm turning the page file OFF on the amd 4600 for a while.

Scott - I've seen handles and threads but have never researched and don't know what they mean.

SCOTT - System cache sizes can also have a huge effect on benchmark scores, depending on the chipsets on the motherboards... Turn off indexing services for all of the drives, that info is in the cache, great place for it...

Need more direction on this. I seem to remember seeing this in the hard drive properties??

I'm still confused because the 4600 machine doesn't show low benchmarks, they are higher then the 3800 machine but the rac is lower.


What is I change the question? Is it normal for the amd 3800 - -
Measured floating point speed 2093.46 million ops/sec - -
Measured integer speed 3696.8 million ops/sec
to have a RAC of 920 and as high as 988?

---

I have parsed a lot of computers same as mine from various users and i have found that sometimes there is a result that is totally exagerated.
I have noticed that such exagerated results always come from the same user.
There must be a bug in the benchmark or some people have another modified version.

From the Boinc website, about the benchmarks:

boinc.berkeley.edu/benchmark.php

From the page:

Whetstone does 8 different groups of tests (repeatedly of course), times how long they took to finish, and produces a number, [ops performed]/[time]. These tests all use floating point math operations of the CPUs being tested.

Unfortunately that second sentence is flat wrong. When Whetstone was originated circa 1962, three of the tests were designated as floating point. The FPUs developed since that time will be used for two more, so it's reasonable to include them. The three remaining tests are not and never will be floating point tests.

                                                       Joe

using ( cntrol alt del) shows only boinc running.

Have sandra but no idea what I'm looking for to identify problem.

Look at Processes Tab when control alt del
Click on CPU header twice to sort processes by CPU usage.

Do you have exactly the same security software installed on all systems? (AntiVirus, Firewall, Anti-Spyware)

From the Boinc website, about the benchmarks:

boinc.berkeley.edu/benchmark.php

Interesting thread... I have a Intel Core 2 quad in my newest computer, and have seti set up to use 3 processors at a max of 50% and it scores 15.79 based on the formula... Present Bionc scores of 2451 million floating and 5129.38 million integer per processor. I just started using this machine for seti, so I am not sure how well it will do... (I will be using this system for a lot of 3d stuff, and Bionc will have to be turned off to do any rendering or to work in high polygon scenes..)

I am not sure how accurate the bionc benchmark is either, when it is running, taskmanager is at about 70 percent... That tells me that the benchmark is not optimized for the processor/cache or the opp system (Xp 64 bit)... It doesn't report the system properties as they actually are either... (Bionc reports 1 meg cache... there is 8 meg in the chip... I assume it is looking at what is allocated to the first processor)

I ran a few other benchmarks on it and all of them came up with different results... The highest numbers I came up with were 13 trillion floating and 26 trillion integer for all 4 processors combined... It is a beta 64 bit benchmark I tested for a friend of mine, so I am not sure how relevant it is to compare it to a 32 bit machine (it is a self booting benchmark full 64 bit, not sure what opp system it is based on, he is in the process of writing the definition files for different processors and chipsets, and I let him run it on mine when the system was first built and no opp system on it, running at 2.66 ghtz)...

One thing I can tell you about benchmarks is that most people don't set the machine up to run them properly...

I am guilty of that as well, most of my benchmarks were in a normal boot (Ie: approx... 13000 handles, 500 threads, and 40 processes running during testing...

The biggest change I found in settings had nothing to do with the programs running... I turned off virtual memory, and all of the benchmarks improved, some by almost twice...

64 bit windows can be set up to have a swap file in memory, that is how I have mine set now...

If you set 32 bit windows set up with a ram drive it works the same way...

Don't ask me why, but windows preferrs to have a swap file, some parts of it require one...

To be fair to everyone I should put what my setup is right now, and it seems to be stable, for now... (I am going to push this one to see what it can really do)

Core 2 Extreme Quad QX6700 12x mutiplier, 280 mhz bus (3.36 ghtz)
Intel D975xbx2 motherboard and all the software is installed for it
4 1 gig ddr2 presently running at 2x bus speed daul channel
Ati X550 video card with 64 bit drivers, 32 bit software
Full Xp64 bit install with all updates..

It is a shame that I can't find everything for the system in 64 bit so I can turn off the 32 bit emulator... I am sure that would speed the system up even more...

(Most of the programs I run on it are 64 bit)

Try changing the virtual ram settings on your systems and see if the benchmarks improve... If you have over 2 gig of memory, add the 3/gb command to windows so it can actually use above 2 gig... Fyi (3.4 gig is the most win xp 32 bit can use)...

System cache sizes can also have a huge effect on benchmark scores, depending on the chipsets on the motherboards... Turn off indexing services for all of the drives, that info is in the cache, great place for it...

Just my 2 cents...

Scott Shaver (subscribed to thread, I am curious as to why the faster Amd is slower)

It might be better to compare your machines to each other and to others in this site by comparing times to crunch different units with different angle ranges. Some are overclocked. Some are devoted to other projects. Their owners might be gone a lot. Some might use different crunchers, etc. MSattler, for example, whose Core 2 Quad had exceeded 4,000 RAC, encountered a broken motherboard which has ruined his RAC temporarily.

---

Here's the data from whence the data for the chart comes:

Zaphod, the Benchmark calcuation is the OLD way of claiming credit. The new way counts flops (floating point operations/second). There are many things which can change the credit. I find that using a stock seti application in windows results in getting about 85-90% of what one would get if using the benchmark.

I collect data on how projects compare. Here's a comparison chart for projects done by my X2 4800 (using stock boinc clients). See the seti specific data. You'll see credit depends on which version of seti you run. Simons' 2.0 application currently give me the best results.

NOTE: ignore the spike in ABCathome, this was the first version they ran in Alpha, it's been adjusted several times since this chart was created.

Also, this chart shows both Windows (left side) and Linux (right side) on the same chart, the blank spot is the dividing line. notice the low benchmark for the same machine when using linux.

[/quote]

The computer you show with a boinc benchmark of 2515/4688 should be claiming 15.00625 credits/run hour.

[/quote]


This benchmark is the 4600 dual core. Based on your estimate the machine should do about 720 credits a day, which it is doing.

BUT that leaves me more confused because of the 3800 machine - Measured floating point speed 2093.46 million ops/sec, Measured integer speed 3696.8 million ops/sec - This machine has a RAC of 903.

Have not checked yet on the comment regarding "the dreaded 58.7's"

The ram is 2 x 1g, paired ddr 3200, it is virtually identical to the ram in the amd 3800 that averages 900 and the amd 3800 that averages 5-700 while I'm using it.

Reminder, one amd 3800 only crunches boinc, one 3800 is my everyday computer, the amd 4600 is boinc only.

Thanks for the comments.

---

You may also want to look at how the cache and the memory setup...Dual channel mode vs. single channel can have huge effects. I have an Anthlon XP 3000+ (single core) that had 1.25GB memory (2x512MB and 1x256MB) with similar timings. I took the 256MB stick out and ran the memory in dual channel and saw a 7-10% improvment in my times.

Not to mention the fact that 2GB Ram doesn't mean a lot....is it DDR/DDR2, PC3200/4000/4200/etc.

The other thing to look at is WUs themselves....the 4600+ could have gotten a series of the dreaded 58.7's

Just a few things to look at...hope this helps.

-citroja

---

Zaphod, Boinc has it's own Benchmark code. it's run whenever you install boinc and then every 5 days thereafter. To find your benchmarks, go to "your account", then "computers - view", then select a computer.

The computer you show with a boinc benchmark of 2515/4688 should be claiming 15.00625 credits/run hour.

there are links to "credit", "granted credit", and "claimed credit" in the Unofficial Boinc Wiki on the [[RAC]] page.

Previous message did not maintain spacing so it may be hard to read benchmarks repost:


Can't make sense of BENCHMARKS.

Boinc Report Whetstone 2515 Dry 4688

Sandra Whet 15998 Dry 18985

---

[/quote]
(whetstone + dhrystone) X 3600/1728000=claimed credit/hour


tony[/quote]



Can't make sense of BENCHMARKS.

Boinc Report Sandra

Whetstone 2515 15998
Dry 4688 18985


Why are these different? Both seem to be in the same units - MIPS. If one was single core and the other dual core it seems the number would just double. Neither of these seem to work in the formula for computing estimated credits.

---

My AMD64 X2 4800 Toledo(OCed 5%) gets about the same (per core) as my single core AMD64 3700 Sandiego (OCed 10%). It gets a little more (per core) than my Mobile AMD64 3700 Newark (no OC).

Mobile AMD64 3700 claims 13.13/hour
AMD64 3700 claims 14.14/hour
AMD64 X2 4800 claims 14.39/hour/core

you can calculate yours by inputting your benchmarks into the following formula:

(whetstone + dhrystone) X 3600/1728000=claimed credit/hour

Seti doesn't use the benchmark for actual granted credit, but it is used at other projects (except ABCathome, Einstein, and Rosetta). Gives a decent comparison.

tony

---

using ( cntrol alt del) shows only boinc running.

Have sandra but no idea what I'm looking for to identify problem.

---