first : do i need to wait my BOINC queue is empty before doing that ?
i dont want to lose any of these 500+ WUs

no. benching is independent of BOINC - in fact the current bench script suspends BOINC for the duration of the bench, to ensure smooth hasslefree running.

edit: please open a new thread, how to set up a bench is rather offtopic.

...
So what if the latest Lunatics app is a Knat's cock slower (on your hardware) if the validation rate is improved, less -12's, less inconclusives, less wasted time, less time waiting for an extra wingman to validate an inconclusive, So the Science is returned sooner and Credit granted sooner,

Claggy

I have no particular issues myself with this specific test, since it illustrates quite well the architectural changes both drivers and applications are undergoing.

There are two main, perhaps not so obvious, issues to consider that are (& aren't) reflected in the numbers seen here.

We have two main issues at play:
- Later drivers refining toward meeting Microsoft's WDDM specification, do by design function slower on these older cards. These cards simply do not have the same memory related features, so an emulation layer is needed. This is in part frustrating for older card owners, even on XP as driver changes are necessary for compatibility there as well. Microsoft's expanded design of the old XP model is a costly forward-looking one devoted to reliability & security, both of which have been major problems under XP's now quite dated driver model (in it's pure 190.xx driver form at least). Basically, IMO, those improvements are worth it, but do have an associated performance cost.
- The second main question is whether comparing old applications is reasonable, considering that there are known stability issues, precision/correctness issues, compatibility issues, and bugs outright that can compromise uptime & validation rates with any application (old stock or new opt). These issues are reduced as the builds progress, as typically are driver bugs and performance issues... So the choices become between raw throughput, & validation rate & subsequent scientific value.

From the standpoint of global reliability & validation concerns, IMO the stock builds should have been replaced a long time ago. While x branch still suffers some obscure key limitations inherited from the 6.09 codebase, it (x41o) is submitted for Beta V7 consideration on the principles of being a solid, far more refined basis for the future, though several known long standing issues inherited from 6.09 remain.

So as it stands, IMO, if there needs to be any choices made, it should be to only stick with any outdated stock or opt applications if there is some pressing need, such as that experienced by the new hardware limitations on older cards like multiple GTX295's in one rig not 'liking' the newer driver model. Despite being a stock application still distributed, 6.09 should certainly not be any more considered as a validation reference, or an example of efficiency (which is what controls general RAC trends after all) but instead be put in some sort of museum.

Jason

Yeah, This driver model allows more than 4 GTX295 gpus to work, but after 5 gpus, in the weeds for all, so the model sucks, I'd love to be able to use XP Pro x64 video drivers in Win7 Pro x64, but I tried that already and I got nowhere as the OS said they weren't allowed to install on this OS.

So You can guess what I'm doing with 5 cards and a block...

Oh and I'm using 275.50 as My video driver.

---

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

One important thing not to forget is that the version numbers for different monitor drivers coming for different cards made by different manufacturers do not necessary match when it comes to the numbers for these drivers!

The second most recent driver for one particular monitor card could have a version number higher than the most recent monitor driver for a different card coming from the same manufacturer or possibly a different manufacturer of just another type or kind of monitor card.

<various bits snipped>

Gladly. NDA stands for 'Non-disclosure agreement', which the Cuda 3.1 beta libraries are subject to. That just means 3.1 based builds can't be released yet, or chat about Cuda 3.1 in specifics etc...

Jason

I'm sure that was meant to say Cuda 4.1 :-)

<various bits snipped>

Gladly. NDA stands for 'Non-disclosure agreement', which the Cuda 3.1 beta libraries are subject to. That just means 3.1 based builds can't be released yet, or chat about Cuda 3.1 in specifics etc...

Jason

I'm sure that was meant to say Cuda 4.1 :-)

Back in 2010 when that post was written ;-) (I think the thread title has migrated a bit since then), I'm sure he did mean CUDA 3.1

I'm sure that was meant to say Cuda 4.1 :-)

LoL, not likely given that's a quote from a very dated post. Cuda 4.1 is out now of course :)

---

"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'm sure that was meant to say Cuda 4.1 :-)

LoL, not likely given that's a quote from a very dated post. Cuda 4.1 is out now of course :)

Wishful thinking then. Oh well I guess we have to wait until April for Nvidia to release drivers as whql to go with the Cuda 4.1 compiler.

I'm sure that was meant to say Cuda 4.1 :-)

LoL, not likely given that's a quote from a very dated post. Cuda 4.1 is out now of course :)

Wishful thinking then. Oh well I guess we have to wait until April for Nvidia to release drivers as whql to go with the Cuda 4.1 compiler.

?? nah 4.1's working fine here at the moment (using 290.53 beta here), just attempting to solve a few long standing annoyances & try to help the Ati crowd get going to make sure I can get a Kepler cheap enough ;)

Naturally something I said ~2 years ago about Cuda 3.1 beta has nothing to do with current Cuda 4.1 status

---

"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'm sure that was meant to say Cuda 4.1 :-)

LoL, not likely given that's a quote from a very dated post. Cuda 4.1 is out now of course :)

Wishful thinking then. Oh well I guess we have to wait until April for Nvidia to release drivers as whql to go with the Cuda 4.1 compiler.

IIRC, BOINC said CUDA version 4010 in the first few messages with nVIDIA driver 285.58 WHQL (WinXP 32bit). This mean this driver support already CUDA V4.1 .


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

---