I'm afraid the way that the cpus were recorded in the data it would be pretty difficult to break down like that. What you are seeing are all of the different unique values stored in the cpu type field.

> Nice stats :)
>
> But what about sorting by CPU Family, would that be possible ?
>
> e.g.
> intel Pentium / 2 / 3 / 4
> intel Celeron
> AMD K6 / K6-2 / K6-2+ / K6-III / Athlon / AthlonXP / Athlon64
>
> etc...
>

---

Yea,

Give poor Zeno and Zain a break...there are apparently 2400+ ways to say Intel and AMD. (I'm guessing...there are 2526 CPU types listed...and I gave 126 to Apples & other systems)

Hmm..lets see...if all OSes gave same string for CPU info..how many would there really be?

Intel
80386, 80486, Pentium, Pentium Pro, Pentium-MMX, Pentium-II, Pentium-III, Pentium-IV, P2-Xeon, P3-Xeon, P4-Xeon, P2-Celeron, P3-Celeron, P4-Celeron, Pentium-M

And then all the various MHz speeds.

You could, I suppose, Include things like Pentium-4EE for those Prescott CPUs with 2 MB of L3 cache.
-----------
So far I've got the only
Intel P3, Pentium-III, 866Mhz, (Coppermine)
Intel P4, Pentium-IV, 2400Mhz, (Northwood)
Intel P4, Pentium-IV, 2775Mhz, (Prescott)
and
AMD K7-XP, Athlon XP, 1525Mhz, (Thoroughbred)

machines.

> http://www.setisynergy.com/stats/cpus.php?project=sah

Nice stats, Zain!

Now I know that I have one of the only 8 (active/visible) hosts with a GenuineIntel Mobile Intel(R) Pentium(R) 4 CPU 2.80GHz processor that run Seti :-)

---

Is the stats for BOINC or only S@H?

Mine were calculated using the SAH data.

> Is the stats for BOINC or only S@H?
>

---

The best idea would be if someone wrote a special app (No i haven't checked if there is one available now) to measure the cpu's IPC as good old SetiSpy did and divide the value with how many cpu's (hyperthreading or dual) the machine have.

That would be the most ideal way to see which machine does the most work 24H/Day..

As it is for the moment it seems like the hyperthreaded Prescotts give the most bang for the buck with this present buggy claimed/granted credit system.

If a A64 cpu makes a wu in about 7000 Secs that machine would claim about 35-40 credits and luckily that machine is granted 35-40 credits but nowadays if the same WU was calculated on a hyperthreaded machine that machine would claim 22-28 credits for the same WU when it took perhaps 10000 Secs instead.

In unlucky conditions if your A64 cpu crunches away more than 12 WU's/day and got granted credit of a prescott machines reference that computer would gain 12 wu's multiplied with 22 - 28 credits = 264 to 336 credits.
The same northwood/prescott compo would get 16+ WU's multiplied with 22 - 28 equals about 400 Credits / Day (Hyperthreaded 2 SETI per HT Core).

If the same machine got his own credits that equals to 12 multiplied with about 35 and the machine would be granted 420 credits instead of todays 264 - 336 credits.

So if you want to undermine the creditgrant system i would say go for an Prescott with duacl channel memory with good throughput (Geil UltraX PC3200, Crucial ballistics ) with ratio 2-2-2-5 and perhaps even be able to run it at Command Rate 1T.. Then that machine would spur out packets faster than many other machines and be granted a whole bunch of credits..

Take this for an example.. Your prescott machine spurs out 18-20 WU's / Day overclocked and with good memory and all of the WU's granted to that machine was calculated by an old AMD 1000 Mhz.. Multiply 20 Wu's with granted credit of perhaps 60 and just look what could happen with this crazy way of granting credits. Your system grants 1200 Credits on a single day when in fact the system was only worth 480 credits..

See how wrong it could be..

Everyone here knows that a A64 cpu has extraordinary good IPC on only 1 core but is double hindered to be granted creds just because of the current solution to let the server decide granted credit level and because of all the Hyperthreaded cpu's out there..

Well don't get me wrong, we're all in this because of science :) but some fair judgement of high output Prescotts and A64's or even Xeons and FX cpus out there should be taken for granted in the server calculates granted credits..

Their validator should have a calculated coefficient to calculate the effecient granted credit value and taking in considiration of atleast how many cpu's that machine have and even perhaps O/S dependancy.. Then we can get rid of unaccurate granted credits in the future.

My very long 2 cents in this matter.

Kind Regards.. Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Vyper,

Would the ultimate goal be for Machine A to get identical credit to Machine B on the exact same WU?

Regardless of whether machine A was a Celeron 500 and Machine B was an Athlon 64 FX-55?

> Vyper,
>
> Would the ultimate goal be for Machine A to get identical credit to
> Machine B on the exact same WU?
>
> Regardless of whether machine A was a Celeron 500 and Machine B was an Athlon
> 64 FX-55?
>
>

That would be pretty much it, like it was in good old SAH.. One completed WU got one extra completed.
And yes i know that they did this to get rid of cheating but the credit script has flaws. That's all.

If they would take number of cpus in consideration then it would level it up rather ok.. But taking an 12 creds grant for a regular A64 is low just because of the huge amount of Intel HT computers out there crunching away fast and effecient but can gain much more creds to their system because fast Athlons crunched it first and the server granted 38 creds to a system that only required 25..

//Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

> Vyper,
>
> Would the ultimate goal be for Machine A to get identical credit to
> Machine B on the exact same WU?
>
> Regardless of whether machine A was a Celeron 500 and Machine B was an Athlon
> 64 FX-55?
>
>

Just found a good example. Look at this one.
http://setiweb.ssl.berkeley.edu/workunit.php?wuid=8830036
Look that computer 204319 required only 14 creds for that WU and got 43.53 just because that the Pentium M was the main computer for the validator to relate to.
This is plain .. Bah..

Get the point every poor 1 cpu people :)

//Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Vyper,

You're partially correct...but from a programming standpoint I will restate what has been said in other threads.

Credit is based on benchmark scores for a machine, and the current benchmarks don't measure a machine the same way as as seti crunching a WU.

So, to be completely accurate, some benchmark based on seti code itself would have to be used to make all machines claim identical credits.

This may happen some time in the future after any/all mechanical problems with boinc are ironed out.