Hello community,


AFAIK (O.K., the infos are a few weeks old ;-), S@h will send out a stock S@h v7 CUDA app in future (AFAIK, a further development of the x38g_cuda32 app) (e.g. the x38g_cuda32 app of LIv0.38 could crunch already S@h v7 WUs).
Maybe S@h will send out then as stock CUDA app the Lunatics CUDA app (so not longer needed to install manually an opti app (at least for nVIDIA GPUs)).
(Maybe my infos are not longer up-to-date)


BTW. IIRC, the Lunatics crew will release only one S@h v7 CPU app for all CPU instruction sets. The app detect the best instructions set for usage.
Maybe S@h will send out also the Lunatics CPU app as stock app automatically?

This would be very nice, so we wouldn't need to play with app_info.xml files.. ;-)
(and, S@h would get the max performance of all PCs out there)


Anyway, I ask me, why we don't test this S@h v7 CUDA app here at S@h Beta Test?
(..and why not this upper mentioned new CPU app?)


Thanks.


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

---

There are 2 optimized CPU apps.

One supports SSSE3 instructions the other one is SSE2 for AMDs and older CPUs.
You will not get opti CPU app as stock at seti main.

Cuda app is still in alpha so no official beta for now.
But there is already a working app as well.

---

With each crime and every kindness we birth our future.

There are 2 optimized CPU apps.

One supports SSSE3 instructions the other one is SSE2 for AMDs and older CPUs.
You will not get opti CPU app as stock at seti main.

Yes, in Raistmer's latest (r365) beta app package.
I installed it and it will crunch (more) latest if my machine get not longer S@h WUs.. ;-)
(a few WUs already crunched and results validated/granted well)

We don't know what happen behind the curtain.. ;-)

Cuda app is still in alpha so no official beta for now.
But there is already a working app as well.

I looked online here and there and found at least one member which use a x41e_cuda32 app (in S@h v7 mode) here at S@h Beta Test.


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -

---

Cuda app is still in alpha so no official beta for now.
But there is already a working app as well.

I looked online here and there and found at least one member which use a x41e_cuda32 app (in S@h v7 mode) here at S@h Beta Test.

About half a dozen, all of them Lunatics Alpha testers or higher. You weren't implying somebody is running unauthorised, were you?

No, I don't know either why the CUDA app hasn't made it into official testing yet.
Apart from the obvious reason - not enough time, not enough people and too many other things going wrong, that need attention.

I don't know who is a Lunatics alpha tester.
But I think the one I saw is one of them.


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

---

Are the Lunatics alpha testers any different then the Guys who run

SETI@home v7 Anonymous platform (NVIDIA GPU) or

SETI@home v7 Anonymous platform (CPU). Just for the credit.

After all niether can be downloaded from seti@home direct.

Looks more like sour grapes. (I can't run it so niether should they.)

Perhaps Eric should ensure that they should be downloadable to all. or not used at all over here.

Michael.

The difference between alpha and beta test is, that the app in alpha status could be (very) buggy.

If the app run well on the alpha tester machines, then the app will be beta tested on more different machines.

So if the S@h v7 CUDA app run well on alpha machines, then maybe we will see a new thread about here in the forum, so all S@h beta tester could test it.

I guess it could be like this.


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

---

Are the Lunatics alpha testers any different then the Guys who run

SETI@home v7 Anonymous platform (NVIDIA GPU) or

SETI@home v7 Anonymous platform (CPU). Just for the credit.

After all niether can be downloaded from seti@home direct.

Looks more like sour grapes. (I can't run it so niether should they.)

Perhaps Eric should ensure that they should be downloadable to all. or not used at all over here.

Michael.

Where should we test those apps if not here on beta ?

For CPU and ATI apps there is already a open beta test here.

---

With each crime and every kindness we birth our future.

Are the Lunatics alpha testers any different then the Guys who run

SETI@home v7 Anonymous platform (NVIDIA GPU) or

SETI@home v7 Anonymous platform (CPU). Just for the credit.

After all niether can be downloaded from seti@home direct.

Looks more like sour grapes. (I can't run it so niether should they.)

Perhaps Eric should ensure that they should be downloadable to all. or not used at all over here.

Michael.

Yes, the Lunatics alpha testers are different, very observant and competent, willing to both report and discuss any discrepancies they see. They're also willing to do extensive offline testing as we request and don't complain about the credits they might otherwise be accruing. In short, they're essential to the overall Lunatics effort.

Eric is very much aware of what we're doing. A limited number of alpha and beta testers here works very well, giving confidence about accuracy and reliability. It is expected that the GPU apps will become the apps which the project releases.

                                                                  Joe

I found the old Message 1145334 of Richard.

(...)
For the longer-term future, the plan is to reduce the number of applications still further, by including all the separate optimisations in, perhaps, a single application which is capable of running without error on a MMX CPU, but which uses SSE2, SSE3, SSSE3, SSE4.1, AVX..... as appropriate if the CPU and motherboard support them. That will eliminate the repeated threads we both know from these boards where somebody has mis-read the difference between SSE3 and SSSE3, or "knows" that SSE4.1 is better because it's a bigger number...

This was for ~ one month..

If S@h v7 will be released, it will be an all in one Lunatics CPU app?

If, will S@h send out this app automatically (no usage of app_info.xml file needed)?

If it will be maybe two or three apps, BOINC detect the instruction sets and choose the correct CPU project app?

E.g. this is from BOINC V6.12.33 on my Intel Core2 Duo E7600, max SSE4.1 instruction set:

Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 lm vmx tm2 pbe

(Hmm, no SSE3? IIRC BOINC mentioned this as ht, or not?)


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

---

I found the old Message 1145334 of Richard.

(...)
For the longer-term future, the plan is to reduce the number of applications still further, by including all the separate optimisations in, perhaps, a single application which is capable of running without error on a MMX CPU, but which uses SSE2, SSE3, SSSE3, SSE4.1, AVX..... as appropriate if the CPU and motherboard support them. That will eliminate the repeated threads we both know from these boards where somebody has mis-read the difference between SSE3 and SSSE3, or "knows" that SSE4.1 is better because it's a bigger number...

This was for ~ one month..

If S@h v7 will be released, it will be an all in one Lunatics CPU app?

If, will S@h send out this app automatically (no usage of app_info.xml file needed)?

If it will be maybe two or three apps, BOINC detect the instruction sets and choose the correct CPU project app?

E.g. this is from BOINC V6.12.33 on my Intel Core2 Duo E7600, max SSE4.1 instruction set:

Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 lm vmx tm2 pbe

(Hmm, no SSE3? IIRC BOINC mentioned this as ht, or not?)

'longer-term future' means in the next few years, Not in the next few days, weeks or months.

Prescott New Instructions (PNI) is also known as SSE3.

Claggy

O.K., if so..

Then S@h should use the CPU instruction set infos of BOINC for to send out automatically the best CPU app.

If the new Lunatics CUDA app will be the new stock S@h v7 CUDA app, then we couldn't longer speak about 3rd party apps. The work of S@h and Lunatics are very close, they work together.
So it would be O.K. - IMHO - to send out the Lunatics CPU instruction set apps as stock S@h apps.

So the project would get automatically the max performance of all PCs out there.


JMHO


[EDIT: BTW. Why BOINC don't say SSE3? It would be better for inexperienced users.]


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

---

I found the old Message 1145334 of Richard.

(...)
For the longer-term future, the plan is to reduce the number of applications still further, by including all the separate optimisations in, perhaps, a single application which is capable of running without error on a MMX CPU, but which uses SSE2, SSE3, SSSE3, SSE4.1, AVX..... as appropriate if the CPU and motherboard support them. That will eliminate the repeated threads we both know from these boards where somebody has mis-read the difference between SSE3 and SSSE3, or "knows" that SSE4.1 is better because it's a bigger number...

This was for ~ one month..

If S@h v7 will be released, it will be an all in one Lunatics CPU app?

No, that won't happen for S@h v7, maybe never. The Lunatics code will probably move toward an all-in-one for x86 architecture, but not in the next couple of months. The project code needs to be buildable for other architectures too, and not rely on specific compiler features and capabilities.

                                                                 Joe

For your info, there is NV-compatible V7 application in rev365 package too. Though it's OpenCL, not CUDA for now.

---

News about SETI opt app releases: https://twitter.com/Raistmer

BTW. Why BOINC don't say SSE3? It would be better for inexperienced users.]

Since this is actually correct many people get confused over instruction sets
The problem has todo with amd calling its similar looking and named SSE3 the same but both are in fact different.
As the programmers from lunatics know all about this issue.
Even now many people think newer amd models contain sse3 and ssse3 but in reality the sse3 from amd should have been named sse3a or something like that.
Second even though on many selling sites the newest amd models been reported to have ssse3 but i can asure you they do not.
Hell as far as i know even the new BD FX ones will lack this set.
Why simply because Intel will never allow its biggest competition to use this on their product unless amd pays billions to them for use it.
I am glad about raistmer and the others diving into the opencl standard because the more it is being used by a bigger audience.
The sooner nvidia is going to spend more time on opencl together with its competition Amd and others, believe me for us that would be much better.
It would be best if cuda died out and only opencl would make it to be the real standard

BTW. Why BOINC don't say SSE3? It would be better for inexperienced users.]

Since this is actually correct many people get confused over instruction sets
The problem has todo with amd calling its similar looking and named SSE3 the same but both are in fact different.
As the programmers from lunatics know all about this issue.
Even now many people think newer amd models contain sse3 and ssse3 but in reality the sse3 from amd should have been named sse3a or something like that.
Second even though on many selling sites the newest amd models been reported to have ssse3 but i can asure you they do not.
Hell as far as i know even the new BD FX ones will lack this set.
Why simply because Intel will never allow its biggest competition to use this on their product unless amd pays billions to them for use it.
I am glad about raistmer and the others diving into the opencl standard because the more it is being used by a bigger audience.
The sooner nvidia is going to spend more time on opencl together with its competition Amd and others, believe me for us that would be much better.
It would be best if cuda died out and only opencl would make it to be the real standard

I don't belief SSE3 from AMD is different from INTEL SSE3, it's a 'standard' (instruction-set)
Since INTEL still is the leader in CPUs and probably has patented SSE3, SSSE3, (SSE4.1(?))etc. (production process, wafer structure, etc.).
AMD has implemented some other features, but they are/were hardly used, IMO!

And with AMD and NVIDIA GPU, it's not much different, looking at the
architecture, they're quite different. 5000/6000 v.s. 400/500 GPUs

And CUDA will probably die a slow dead and OpenCL, will be "the new standard".

Hi All!
Thanks for all your keen interest in helping to test out newer Cuda builds. I can confirm that Eric requested & received latest (x41e) builds about a week ago, though no doubt has been extremely occupied with more urgent matters delaying whatever he wanted to do with them (presumably checking & installing here). I am hesitant to make current builds available beyond alpha crew, simply because I have no intention of overriding the project plans for automated release & testing.

You've all been testing elements of the CUDA refinements already, mostly on main, through builds including & after x38g. In fact Raistmer's OpenCL builds derive from & extend some key GPU code worked out from V7 stock, refined in MATLAB & ported through Cuda, so many of you are already testing shared technologies between OpenCL<->Cuda.

Only the last few nagging (long standing) issues remain on the Cuda front, i.e. the most challenging ones, which you've all been instrumental in pinning down over time. The accrued knowledge from that has been so successful in isolating & understanding things, that beyond the expected Beta here, vast amounts of work are already on the drawing board to take things to the 'next level'. I don't make it here often, because that's more alpha/dev work than beta, certainly experienced alpha testers & fellow developers play a key role in making sure I get the information I need from all sources, as Joe described earlier.

Given the relative stability of development on the Cuda front, I would encourage everyone, that is able to, to assist Raistmer with his OpenCL work at this time, as there are technical hurdles to overcome due to the rapid technological developments in this area, including major language revision, suspected driver/SDK problems & underlying infrastructure changes.

We all hope that OpenCL will ultimately become a universally applicable & refined GPU supercompute architecture that was promised. Raistmer's work, with your help there will be critical in ensuring that happens, in addition to potentially having much greater impact that just 'here' for Seti@Home (Think Big :) ).

I myself happen to prefer Cuda development, but can fully appreciate the long term directions OpenCL shows glimpses of.

Meanwhile 'Don't Panic', Cuda V7 is closer than you think... 'You've been soaking in it Jan!' :-D

Best Ragards,
Jason

Thanks Jason_gee, for the clear explanation about CUDA , I read in a few magazines, OpenCL would probably been used more and more, also since
ATI and NVIDIA GPU, are both capable in handling OpenCL, which is also
'less bound', (free?), owned by the Knohnos Group.

CUDA, is NVIDIA.(Will not run on an ATI GPU)

I'm, b.t.w. running the ATI rev.365 app. for MB WU's (A.R.0.001 to A.R. >10{?})
on 2 HD5870s, 2 on each GPU, period_iterations 2.

Should try more 'settings'..............

---

How is the status of the new S@h v7 CUDA app - ready for beta testing?


If my machine (stock - no app_info.xml file) ask for GPU work (S@h v7) BOINC messages: no tasks - for CPU tasks available..


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

---

How is the status of the new S@h v7 CUDA app - ready for beta testing?


If my machine (stock - no app_info.xml file) ask for GPU work (S@h v7) BOINC messages: no tasks - for CPU tasks available..


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC (@ SETI@home). - SETI@home needs your help. -

Probably any week now, just waiting for the revised 6.98 CPU app too,

Claggy