Not very important, IMHO, but the latest Anonymous Platform, reports
4 cores on an i7-2600 with HT is on?

Also a completely different matter, SSE3 should run faster on i7's,
compaired to SSSE3x or SSE4.1 and this has been tested?!

Maybe the AR has a big influence, but also the memory speed/timings?!
I've seen i7's (XXX) running SSSE3x and have shorter runtime, at 0.4 AR!
Probably the best, trying this myself ;-)

---

Not very important, IMHO, but the latest Anonymous Platform, reports
4 cores on an i7-2600 with HT is on?

The core count is physical rather than logical. Perhaps we'll change that fairly soon, there are other reasons to consider improving the capability checking.

Also a completely different matter, SSE3 should run faster on i7's,
compaired to SSSE3x or SSE4.1 and this has been tested?!

Maybe the AR has a big influence, but also the memory speed/timings?!
I've seen i7's (XXX) running SSSE3x and have shorter runtime, at 0.4 AR!
Probably the best, trying this myself ;-)

It is absolutely the best policy to check such things on your own system, and include VLAR and VHAR tasks too. The advice that SSE3 is probably better is based on observations, but we don't have a huge test lab full of machines on which to perform thorough testing.

                                                                   Joe

I had asked this about a year or so ago. As I was seeing my dual CPU Xeon systems reporting "1 x 2992 MHz" instead of 2 or 4. Also my dual 6 core Xeon box reports "Speed: 8 x 2390 MHz" instead of 12 or 24. IIRC I was informed that the bit of software they is use for the CPU detection comes form a 3rd party. So there isn't a lot they can do about it. I'll see if I can find that thread and drop a link to it.

Edit: Here it is Lunatics MB opt apps limited to counting 16 CPUs?

It is a bit funny comparing the older app to the newer one on my 12/24 core box.

SSSE3x Win64 Build 76 , Ported by : Jason G, Raistmer, JDWhale
CPUID: Intel(R) Xeon(R) CPU E5645 @ 2.40GHz
Speed: 16 x 2390 MHz 

SSSE3x Win64 Build 386 , Ported by : Jason G, Raistmer, JDWhale
CPUID: Intel(R) Xeon(R) CPU           E5645  @ 2.40GHz 
Speed: 8 x 2390 MHz

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

For what it's worth (if anything) ...

I'm running a 64bit win-7 with 16Gb mem core I-7 860 2.80Gh

I'm running one Lunatics app on CPU's 0,2,4,6 and another Lunatics app on CPUs 1,3,5,7.

running each test for 20 days I have found:

SSSE3 is 1.47% faster than SSE3
SSE41 is 3.84% faster than SSE3
SSSE3 is 10.57% faster than SSE41

I'm trying to test AK_v8b2_win_SSE3x.exe*32 V.S. AK_v82_win_x64_SSSE3x.exe but the server keeps crashing :-)

Ed F

For what it's worth (if anything) ...

I'm running a 64bit win-7 with 16Gb mem core I-7 860 2.80Gh

I'm running one Lunatics app on CPU's 0,2,4,6 and another Lunatics app on CPUs 1,3,5,7.

running each test for 20 days I have found:

SSSE3 is 1.47% faster than SSE3
SSE41 is 3.84% faster than SSE3
SSSE3 is 10.57% faster than SSE41

I'm trying to test AK_v8b2_win_SSE3x.exe*32 V.S. AK_v82_win_x64_SSSE3x.exe but the server keeps crashing :-)

Ed F

Your math doesn't make sense. If SSE41 > SSSE3 > SSE3 then how can SSSE3 be > SSE41 ?

---

It's just what the results were for each of the 20 day run ... go figure ... perhaps the HT conflict between the apps created a difference???

Ed F

IMHO, it appears inpossible to test this, unless you'll use 3 identical
hosts with 3 copies of a SETI MB, (or AstroPulse) WU's, which
are identical, not just 3 MB WU's.

Each MB WU has too much differences todo a thorough test.
And on top of this it'll be hard enough to find/make 3 identical hosts.

---

I agree, but I am using the same host with 2 BOINCs running on it ... This should cover the identical PC idea ... BUT
I can't run the identical WU mix so I just run the "common mix" for 20 days and ASSUME the 250 (+ or -) WU on each BOINC will "avg out" enough to give an indication ... perhaps not a result.

Ed F

For what it's worth (if anything) ...

I'm running a 64bit win-7 with 16Gb mem core I-7 860 2.80Gh

I'm running one Lunatics app on CPU's 0,2,4,6 and another Lunatics app on CPUs 1,3,5,7.

running each test for 20 days I have found:

SSSE3 is 1.47% faster than SSE3
SSE41 is 3.84% faster than SSE3
SSSE3 is 10.57% faster than SSE41

I'm trying to test AK_v8b2_win_SSE3x.exe*32 V.S. AK_v82_win_x64_SSSE3x.exe but the server keeps crashing :-)

Ed F

I'm doing this kind of testing as well with my five i7-860 boxes. I ran the SSSE3x 32-bit vs the SSSE3x 64-bit for several months and came up with the 64-bit being about 9% faster. Now I am running SSE3 64-bit vs SSSE3x 64-bit. Right now it looks like the SSE3 might have a 5% advantage over the SSSE3x app, but it is still to early for me to say.
I know benchmark testing has showing that in the case of i7's SSE3 is generally faster, but differing hardware can have different results. however it seems like my hardware may fit in with those results.
Once I finish this testing I will be testing running fewer than 8 tasks at once. As when I had my home i7-860 running during the winter with 4 tasks (with HT enabled) at once. It was producing nearly the same RAC as my others running 8 at once. With only dual channel memory there may be an optimal number to run between 4-8.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Thats true HAL.
I found out SSSE3 is faster on Mobos with fast memory transfer.
Triple or quad channel memory might help too.
Unfortunately SSSE3 version dont work on my FX while SSE4.1 works fine.
Intel compler switches hmmmm.
But AMD CPUs only supports dual channel memory so i have to stick with SSE3.
Maybe i upgrade to 1866 RAMs and try again.

---

With each crime and every kindness we birth our future.

Once I finish this testing I will be testing running fewer than 8 tasks at once. As when I had my home i7-860 running during the winter with 4 tasks (with HT enabled) at once. It was producing nearly the same RAC as my others running 8 at once. With only dual channel memory there may be an optimal number to run between 4-8.

since I'm not comparing anything except the 2 hardware instruction sets against each other, I thought I'd give myself a little CPU back while attempting to compare 64bit vs 32bit while maintaining/improvine overall RAC for the box ... soooo this is my new setup

CPUs 0,2,4 64bit
CPUs 1,3,6 32bit
cpu 5 has 2 copies of lunatic's 32bit GPU for my fermi card
CPU 7 has 1 copy of lunatic's 32 bit GPU for my non-fermi card

This frees up 2 cpu's from SAH and consumes some for the GPU feeding and allows me to use the non-fermi card with only 1 WU.

We'll see what happens ... assume the servers stay up for 4 weeks ;-)

Ed F

For what it's worth (if anything) ...

I'm running a 64bit win-7 with 16Gb mem core I-7 860 2.80Gh

I'm running one Lunatics app on CPU's 0,2,4,6 and another Lunatics app on CPUs 1,3,5,7.

running each test for 20 days I have found:

SSSE3 is 1.47% faster than SSE3
SSE41 is 3.84% faster than SSE3
SSSE3 is 10.57% faster than SSE41

I'm trying to test AK_v8b2_win_SSE3x.exe*32 V.S. AK_v82_win_x64_SSSE3x.exe

well... for what it's worth ...

32 bit SSSE3 is 29.81% faster than 64 bit SSSE3 on my 21 day same computer test (see above) (sure seems strange to me!!)

Ed F