I have two older AMD processors running at about at 900 and 1000 Mhz, and have recently started a "Pay it Forward" plan, adding machines of friends for whom I fix their computers. So I've added an Intel running at 1.8 Ghz.

But the faster Intel is actually slower in integer and floating point operations per second. What gives?

It seems like this kind of information is useful -- if only to let us BOINCers know what kind of processors are better than others. So -- not including merged computers or anything else -- is there a list top to bottom of the best processors and speeds for the project?

---

Netburst P4's such as the 1.8 your on about have always suffered badly in the standard boinc benchmarks.

As for the list I'm not sure if there is one but roughly in terms of the most common recent architecture's it would be along the lines of :

core2duo>core duo>Athlon 64>Pentium-M (centrino)>Athlon XP> netburst intel's.

---

There are projects outside of BOINC...

I have two older AMD processors running at about at 900 and 1000 Mhz, and have recently started a "Pay it Forward" plan, adding machines of friends for whom I fix their computers. So I've added an Intel running at 1.8 Ghz.

But the faster Intel is actually slower in integer and floating point operations per second. What gives?

It seems like this kind of information is useful -- if only to let us BOINCers know what kind of processors are better than others. So -- not including merged computers or anything else -- is there a list top to bottom of the best processors and speeds for the project?

Hi Ariztophanes,
It's been my experience that the mix of processes and services typically running on a computer with the BOINC-related stuff has a very significant inluence on the BOINC benchmark results--and on the efficiency of getting science results, too.

For examples, I have three computers, all of which are visible. The generally benchmarh at around the following numbers:

1. "Laptop": Intel Celeron, 1.40GHz: 1245 FP MIPS, 2580 Integer MIPS.
This is a Windows XP Home SP2 bottom-of-the-line Dell B-120 laptop with 512MB RAM (533MHz DDR2) running with 1MB Level 2 processor cache (my only upgrade in the order) and a 400MHZ Front-Side Bus, all stock off-the-shelf. It also runs Norton Antivirus real-time protesction, ZoneAlarm, and Windows Defender, and a "typical" mix of serrvices and background processes including the print spooler, since this is my primary computer that I use for office work as well as web surfing, etc.

2. "gw": A desktop box that originally housed a Gateway 200MHz Intel 386, but was upgraded some time ago with an Intel motherboard with an Intel Pentium 3, 733MHz, 512MB of compatible RAM: 650 FP MIPS, 1145 Floating Point MIPS.
This Windows 2000 Pro SP4 box is a 100% "cruncher"--modest as it is--and so runs with relatively few background processes and services. Only about 19 services normally run, and only 15 processes. Except when I'm checking it, I use the Windows Task Manager to stop even the BOINC Manager and Windows Explorer, and lower the task priority on any process I can other than the network communications real-time process. I push the task priority on the BOINC Client (boinc.exe) to High before closing the Task Manager, but leave the Rosetta science application at it's normal priority of Low. (Every new Rosetta job resets it to Low anyway.)

3. "Compaq1245": A prehistoric Compaq 1245 laptop "upgraded" to an AMD K6 processer at around 380MHz, with all of 168MB of RAM.
This beast benchmarks at around 175 MIPS Floating Point and 225 MIPS Integer.
It is set up much like the previous computer, except about 20 Services and only 12 total processes including the System Idle Process. (It does not have the three processes required by the network interface on the previous machine).

All machines communicate on 100mbps 802.3/"Ehternet" to a broadband connection.

You may want to use the Managemet Console and the Task Manager to check out what services and processes are usually running, and see if some of those can be disabled or stopped.

Although you can run the BOINC banchmark manually to see what effect such changes are having, the real difference will be best shown by seeing how much in the way of total "Granted Work Credit" is generated over some period of time such as a day. That will represent as closely as any measure can the actual amount of science work being delivered to Rosetta.

I hope to (or maybe someone else will!) start a thread on techniques for squeezing the most science out of the machines we already have at our disposal. For now, I have chosen to pay for mortgage, gas, and food rather than a shiny new cruncher... :-)

Good luck, and keep crunching!

We will be able to see what systems Rosetta runs best on over the next few weeks using the new credit system.

Per core and per MHz, I'd expect the Athlon64s, AthlonXPs, Pentium-Ms and P3s to be pretty well matched. I don't think the core architecture would be that far ahead of these either. The P4s trail per MHz, but obviously have higher clock rates to account for some of the deficit.

HTH
Danny

---

But the faster Intel is actually slower in integer and floating point operations per second. What gives?

Ariztophanes:

That's an example of IPC (Instruction Per Clock cycle) that AMD has been chanting about ever since the P4 line came out with clock cycles high enough to match AMD's cpu performance.

The Pentium III line had an IPC roughly the same as AMD's Athlon cpus. So a 1Ghz PIII roughly equalled a 1Ghz Athlon XP. When the P4 line came out, Intel went for speed, and lowered the IPC. AMD came out with a new naming standard. A P4 running at 3Ghz will perform roughly the same as an Athlon 3000+ running at 2Ghz. The P4 has an IPC of say.. 2. The Athlon line has an IPC of say.. 3. So the P4 can perform up to (3Ghz * IPC of 2) 6 Giga operations per second, while the Athlon 3000+ running at 2Ghz can perform up to (2Ghz * IPC of 3) 6 Giga operations per second.

This explanation is vastly simplified - and there are strengths of each architecture that don't match at the P4 running 3Ghz vs the Athlon running at 2Ghz.

With the Pentium M, Core and Core2 cpus, Intel dumped the speed is everything and went back to a system with lower clock speeds and higher IPCs.

So when you're asked which is faster - a 2Ghz Core Uno cpu, a Pentium 4 running at 3Ghz, or an Athlon 64 at 2Ghz - the answer is that they should all perform about the same. (Although you'll notice in some of the benchmarks that one architecture favors certain types of applications, and another architecture favors different types of applications.)

If you are really interested in the issue, there's some wonderful indepth articles I ran across on Ars Technica which describe the architectures in more detail than most of us need. Ask, and an Ars Technicaian will post links if one of us non-AT folks don't beat them to posting.

---

Thanks, all!

I see there's more to it than meets the eye.

It's nice to know everyone out there is willing to help!

---