<blockquote>

Here are Chris's current benchmark results:

Reference 4.02 Berkeley: 9147s
Our v4.07.2 x86_64: 6466s +41,5%
Aki's v4.7 x86_64: 6975s +31%
Metod's 4.07 ICC-P4: 6530s +40%
Our latest setiathome_4.07.3c_x86_64: 8048s +13% ???

</blockquote>

31% - that is not the newest version. it should be as fast as your 4.07.2 because i did not recompiled it with gcc4 yet ;) will do that later.

i have also a new boinc client for x86-64
yours:
2374 double precision MIPS (Whetstone) per CPU
6435 integer MIPS (Dhrystone) per CPU

my version:
2000 double precision MIPS (Whetstone) per CPU
7750 integer MIPS (Dhrystone) per CPU

it should give some more credits


my results for seti are a bit different. metods client was the fastest for me (not 4.07.2) until you released your athlon xp 4.07.3. it seems to by as fast as metods client so we don't need ICC :) maybe it is even faster in average.

I have also released some new boinc clients for Pentium4

---

<blockquote>
i have also a new boinc client for x86-64
yours:
2374 double precision MIPS (Whetstone) per CPU
6435 integer MIPS (Dhrystone) per CPU

my version:
2000 double precision MIPS (Whetstone) per CPU
7750 integer MIPS (Dhrystone) per CPU

it should give some more credits


</blockquote>

Nice job Aki :)

Yes, it's probably an older client Chris tested of yours as it was a while ago.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

Aki, Ned

Yes the banchmark with Aki's seti client is quite old and no longer valid.
The one downloadable on Aki's site should be exactly the same as our 4.7.2 release, since it was compiled the same way... :-)

The newest Boinc Client for x86_64 (4.19.3) that is now available from Ned's site is also giving the same high benchmarks as Aki's newest Boinc client. Again that is expected if the same settings and tools were used for the compile. :-)

My benchmarks for the new x86_64 Boinc client are (Athlon 64 3500+):
2002 Whetstone
7698 Dhrystone

The Bench of the new Athlon-XP Seti client on my x86_64 system is under way, I will post the results later today.

These clients really give a boost to those of you using Linux and an AMD processors!

Enjoy! ;-)

---

Chris Bosshard
Visit my homepage
astroinfo SETI page

OK here are the results for the Linux Atlon-XP client 4.07.3 on x86_64!

Benchmarks so far:

Reference 4.02 Berkeley: 9147s
Ned's v4.07.2 x86_64: 6466s +41,5%
Metod's 4.07 ICC-P4: 6530s +40%

New Ned's v4.07.3 Athlon-XP on x86-64: 5460s +67% :-D

The bench was done on my Athlon 64 3500+ running FC4 Test3.

I would suggest everybody running 64bit Linux to switch over to this new seti client, it certainly rocks!

---

Chris Bosshard
Visit my homepage
astroinfo SETI page

Ned, any chance of an i586 client? the k62's and pentium 1's (mmx and non-mmx) are left out... i think they need the most help :(

i've been compiling my own, but as slow as they are, it's not feasible to tune and benchmark.

Ned, any chance of an i586 client? the k62's and pentium 1's (mmx and non-mmx) are left out... i think they need the most help :(

i've been compiling my own, but as slow as they are, it's not feasible to tune and benchmark.

You can try to run my generic client which can be found at http://www.mkx.si/boinc/#seti. I never tried to run them on CPU older than P2, but they are supposed to run on any truely 32-bit ix86 CPU (eg. i386 or later).

My guess is that also Ned's i686 client should run on these platforms.

---

Metod ...

spacemeat

Metod is right.
His generic client or Ned's i686 client should work also on older CPU's, just give it a try.

We would of course appreciate some feedback if it works or not... ;)

---

Chris Bosshard
Visit my homepage
astroinfo SETI page

OK here are the results for the Linux Atlon-XP client 4.07.3 on x86_64!

Benchmarks so far:

Reference 4.02 Berkeley: 9147s
Ned's v4.07.2 x86_64: 6466s +41,5%
Metod's 4.07 ICC-P4: 6530s +40%

New Ned's v4.07.3 Athlon-XP on x86-64: 5460s +67% :-D

The bench was done on my Athlon 64 3500+ running FC4 Test3.

I would suggest everybody running 64bit Linux to switch over to this new seti client, it certainly rocks!

That's just awesome Chris :D

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

Ned, any chance of an i586 client? the k62's and pentium 1's (mmx and non-mmx) are left out... i think they need the most help :(

i've been compiling my own, but as slow as they are, it's not feasible to tune and benchmark.

I get a little confused with older architectures, so bare with me.

i686 is equivelent to pentium pro, so my i686 client will run on a pentium pro or pentium2 upwards. It won't run on an original pentium or pentium mmx (i586), i486 or i386 (not that you'd want to run seti on those earlier CPUs anyway).

I thought that AMD K6 was equivalent to pentium so it won't run on that, but K6-2 and K6-3 were equivalent to pentium2, so it should run on those?

Please try the i686 version and if it doesn't run, I'll certainly consider a generic version for older architectures.

Also, due to the complex nature of these highly optimized clients, and the nature of the static builds, I suspect that even if I did, say, an i386 version, they may not run on the older distro's likely to be runing on such older hardware. To be honest, people are more likely to have success compiling their own non-static versions on the specific platform they are going to be using.

Also, doesn't Berkeley only do an i686 version?

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

OK here are the results for the Linux Atlon-XP client 4.07.3 on x86_64!

Benchmarks so far:

Reference 4.02 Berkeley: 9147s
Ned's v4.07.2 x86_64: 6466s +41,5%
Metod's 4.07 ICC-P4: 6530s +40%

New Ned's v4.07.3 Athlon-XP on x86-64: 5460s +67% :-D

The bench was done on my Athlon 64 3500+ running FC4 Test3.

I would suggest everybody running 64bit Linux to switch over to this new seti client, it certainly rocks!

that is a really good improvement, but

my Athlon64 3.200+ running on Windows XP SP1

with the optimized SSE2 Client

http://setiweb.ssl.berkeley.edu/results.php?hostid=318394&offset=20

it seems to be faster or ...

only for Info

Greetings from Germany NRW
Ulli S@h Berkeley's Staff Friends Club m7 ©

Ned, any chance of an i586 client? the k62's and pentium 1's (mmx and non-mmx) are left out... i think they need the most help :(

i've been compiling my own, but as slow as they are, it's not feasible to tune and benchmark.

I get a little confused with older architectures, so bare with me.

i686 is equivelent to pentium pro, so my i686 client will run on a pentium pro or pentium2 upwards. It won't run on an original pentium or pentium mmx (i586), i486 or i386 (not that you'd want to run seti on those earlier CPUs anyway).

I thought that AMD K6 was equivalent to pentium so it won't run on that, but K6-2 and K6-3 were equivalent to pentium2, so it should run on those?

Please try the i686 version and if it doesn't run, I'll certainly consider a generic version for older architectures.

Also, due to the complex nature of these highly optimized clients, and the nature of the static builds, I suspect that even if I did, say, an i386 version, they may not run on the older distro's likely to be runing on such older hardware. To be honest, people are more likely to have success compiling their own non-static versions on the specific platform they are going to be using.

Also, doesn't Berkeley only do an i686 version?

Ned

i'll have to give it a try on K6-2, but i learned from using Gentoo not to bootstrap those CPU's as i686 because it isn't supported properly.

other than that i have about a half dozen pentium systems (133-266MHz) doing nothing. i had them in a cluster at one point, but BOINC refused to run on anything less than 64MB RAM (which i know it doesn't need)...

Spacemeat,

Sounds like you may need a lower client then. What would you specifically like, i386 or i586?

I'll try and do you one just for you when I get some time - might be a week or so. Don't expect the same level of performance increases we are getting now though - probably more like 30% as those architectures don't support many of the optmization flags we currently use.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

Spacemeat,

Sounds like you may need a lower client then. What would you specifically like, i386 or i586?

I'll try and do you one just for you when I get some time - might be a week or so. Don't expect the same level of performance increases we are getting now though - probably more like 30% as those architectures don't support many of the optmization flags we currently use.

Ned

i586 covers everything else i have (except BeOS and sparc). i'm going to try i686 on k6-2 and see if that works (it just might). if so, i won't worry about the others for a while because they still can't run boinc anyway

unfortunately no, i686 did not work for me on K6-2.

unfortunately no, i686 did not work for me on K6-2.

Thanks for testing. I'll do you an i586 client at some point :)

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

... I would suggest everybody running 64bit Linux to switch over to this new seti client, it certainly rocks!

Will do! My A64 3200+ is getting a serious inferiority complex!

---

unfortunately no, i686 did not work for me on K6-2.

I'd be interested to know if my generic FFTW client runs on your K6-2 ...

---

Metod ...

unfortunately no, i686 did not work for me on K6-2.

I'd be interested to know if my generic FFTW client runs on your K6-2 ...

just started it up, no errors using setiathome-4.07-fftw.i686-pc-linux-gnu ~ yay. i'll let it run a while and note what happens. this is a k62-300. i am also running a P2-300 and an ultrasparc 333 so i can compare any improvements with these similar speed grades.

Please let us know how the i686 seti client compares to other clients you've run.

Ned

Maybe this has been discussed somewhere already :-/
Do you prefer a test result using the reference wu from BOINC download site or is it sufficient for you when I give results from real wu?

rattelschneck

Rattelschneck

Well the Reference WU is certainly the best thing, but you would also need to rund the Ref. WU on your earlier client to compare...

If the client works and the results are accepted and validate everything should be fine.

---

Chris Bosshard
Visit my homepage
astroinfo SETI page