What one possibly can say here it seems, is that the SSSE3x build may be faster on true multicore CPU's, like the Core2 range, while the SSE3 is faster on HT enabled CPU's like my i3, and the ATOM.

Well, yes and no.

That's exactly the effect I saw on my i5 - somewhat to everyone's surprise - so it's good to have it confirmed.

But I'm pretty sure it isn't anything directly related to the number of cores, or hyperthreading as such. I'd bet (subject to checking by the real hardware gurus) that it's more to do with the internal micro-architecture of the various generations of Intel CPUs.

What one possibly can say here it seems, is that the SSSE3x build may be faster on true multicore CPU's, like the Core2 range, while the SSE3 is faster on HT enabled CPU's like my i3, and the ATOM.

Well, yes and no.

That's exactly the effect I saw on my i5 - somewhat to everyone's surprise - so it's good to have it confirmed.

But I'm pretty sure it isn't anything directly related to the number of cores, or hyperthreading as such. I'd bet (subject to checking by the real hardware gurus) that it's more to do with the internal micro-architecture of the various generations of Intel CPUs.

Well, I was just speculating. It would be interesting to know the real reason why some CPU's although they are capable of the SSSE3x, runs MB faster with SSE3. Maybe in time some hardware gury will answer that.

The reasoning put forward by Joe Segur involved that much of Alex's Core2 (SSSE3 & up in our builds) optimisation was there related to handling misaligned memory accesses. A big problem traditionally when dealing with large amounts of data at strange strides. Since the later architectures have more sheer transistors dedicated to handling those weird alignments of data, the extra code really becomes extra unneeded overhead. I agree with that so far.

I further proposed that having L3 cache also closer resembles later Pentium 4 (SSE3) architectures, so the compiler optimisations for those may be dealing with memory more appropriately as well.

As far as I'm aware, neither of us have done sufficient testing to verify those completely, but this test helps push further along those directions. Understanding what the architecture does well comes in handy, and more efficient memory handling is going to be a very useful thing to have confirmed.

Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

I have just installed lunatics v0.39 on my four PCs http://setiathome.berkeley.edu/hosts_user.php however I am not sure that I have used the optimal apps. In the past I simply looked for the most advanced feature that the CPU would support (SSE4 or SSSE3) and used that. However, following the advice that the SSE4 app only works best for fast processors and that SSE3 is better than SSSE3 for Intel i3-7, this time I used the SSE3 app for my Intel(R) Xeon(R) CPU X3430 @ 2.40GHz and SSSE3 for the other three machines (Intel(R) Core(TM)2 Duo CPU E4500 @ 2.20GHz; Intel(R) Core(TM)2 Duo CPU E7500 @ 2.93GHz and Pentium(R) Dual-Core CPU T4200 @ 2.00GHz). None of my processors support hyperthreading. Is this correct?

---

I have just installed lunatics v0.39 on my four PCs http://setiathome.berkeley.edu/hosts_user.php however I am not sure that I have used the optimal apps. In the past I simply looked for the most advanced feature that the CPU would support (SSE4 or SSSE3) and used that. However, following the advice that the SSE4 app only works best for fast processors and that SSE3 is better than SSSE3 for Intel i3-7, this time I used the SSE3 app for my Intel(R) Xeon(R) CPU X3430 @ 2.40GHz and SSSE3 for the other three machines (Intel(R) Core(TM)2 Duo CPU E4500 @ 2.20GHz; Intel(R) Core(TM)2 Duo CPU E7500 @ 2.93GHz and Pentium(R) Dual-Core CPU T4200 @ 2.00GHz). None of my processors support hyperthreading. Is this correct?

It's a long time since I looked at the List of Intel Xeon microprocessors - boy, are there a lot of them!

Your X3430 looks like a 45nm Lynnfield with SSE4.2 and TurboBoost - so yes, that sounds like the same class of beastie as an i3. I'd give it a try with the SSE3, but as you may have gathered, this isn't an exact science yet. If you feel like it, give it a comparison run with SSSE3x - that's likely to be the only other good one, SSE4 certainly wasn't any benefit on my E5320 Clovertowns.

Edit - if anybody's keeping notes, the processor I noticed the SSE3 app running fast on is a 32nm Arrandale i5 - that's a mobile version of the first-generation i5 series, before they added AVX.

The 486 was simpler...

The 486 was simpler...

I rather liked the Z80 myself ;-)

My first computer had a 6502A

---

Fingers, with a later upgrade to toes as an
auxiliary processor.

My first computer had a 6502A

Mine was a 6502b.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

This may be off thread (and may make no sense) but ...

With all the talk of sse3, ssse3, and sse4.1 (not to mention MMX etc) ...

Proposition:

On an Intel hyperthreading processor, wouldn't it make most sense to run the most architecturally divergent processes on the same CPU? I.E. something like - sse3 assigned affinity to even "cores" and sse4.1 to odd "cores" so that the chip can use idle registers (or whatever) on both halves of the chip??

I would have no idea what this would mean in real life but it seems like a good idea ...

Hyperthreading the same code on both sides of a chip should only provide an advantage based on timing of instruction arrivals but no advantage in utilizing idle hardware .

or am I all wet ??

Ed F

Thank you Sten-Arne. Your exhaustive test is really appreaciated.

Anyone tried the 11.12 drivers with 0.39 and ATI 6xxx card?

Or the 12.1 preview drivers?
http://support.amd.com/us/kbarticles/Pages/Catalyst121Previewdriver.aspx

---

Yes, both drivers are safe to use with both 5xxx and 6xxx cards.

The 100% CPU bug is now fixed.

---

With each crime and every kindness we birth our future.

Yes, both drivers are safe to use with both 5xxx and 6xxx cards.

The 100% CPU bug is now fixed.

Thanks Mike!
Guess i'll try the 12.1 preview with my HD6970.

Can anybody drop me a link for the prog that cleans any driver-left-overs?
Had a fresh install and can't remember the name/url.

---

http://www.guru3d.com/category/driversweeper/
there you go. I'll give the 11.12 a try as well

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

My first computer had a 6502A

National Semiconductor SC/MP (we eventually used the SC/MP-2 NMOS version) -- helped build a 4-processor multiprocessor system with shared-memory communication to run a computer-controlled Fabry-Perot etalon and its DAS, in 1978, and installed it in Antarctica in 1980. In the meantime I built a Philips 2650 kit published in Electronics Australia -- I never did build their PDP-8 clone based on TTL SSL chips despite working with a PDP-8/e for my PhD.

---

http://www.guru3d.com/category/driversweeper/
there you go. I'll give the 11.12 a try as well

Guru3D no longer supplies Driver Sweeper

"
Publisher: Guru3D.com / Phyxion (FnF)

Attention - Guru3D Driver Sweeper has been discontinued as such you will not find any download links here any longer.
"
http://downloads.guru3d.com/Guru3D---Driver-Sweeper-(Setup)_d1655.html


But the new version of Driver Sweeper 3.2.0 is here:
http://phyxion.net/item/driver-sweeper.html


I found the link here (Nvidia Forceware Tweak Guide):
http://www.tweakguides.com/NVFORCE_3.html

Similar Guide for ATI cards (somehow old - 2009) (ATI Catalyst Tweak Guide):
http://www.tweakguides.com/ATICAT_1.html

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

with V0.39
is it faster to run two work units on a gtx 580?
what about on a gtx 580 two or three???

---

with V0.39
is it faster to run two work units on a gtx 580?
what about on a gtx 580 two or three???

The basic rule, is to increase the number of wu until you get to 97% or higher GPU usage. This is assuming your GPU's have the memory to run multiple tasks. Every rig is a bit different, but try adding one at a time until you get to 97% or better usage. Beyond that you run into the law of diminishing returns.

Steve

---

Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website

You can easily run 3 with a 580.
Maybe 4.

Check what Steve said of course.

---

With each crime and every kindness we birth our future.