Just trying to clarify the baseline for the comparison. So far as I can see, a Core i7 using 7.96 cores + a CUDA card should crunch far more than a Core i7 without the CUDA card. And the CUDA card should be a cheaper way of upgrading my Q9450 rig than a complete new Core i7 rig. Is that right, Who?

F.

If you ask me now then it would be a no. But if you ask me later on when we doesn't need to babysit our machine ta notice that too many errors occured so that you need to reboot to get the gpu to go in its highest throttle.. then Yes..

Pair a q9450 with atleast a gtx260 core 216 beast and you would have really high rac later when other computers start to turn in your AP units too..

But if they fix so you can pair cpu + gpu to do s@h wu's then the RAC would fly if it wouldn't be buggy.

Regards Vyper

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

The interesting part is that you ll get cheaper way to accelerate with the CPU than with the GPU ... For example, going to Dual Nehalem will give you better scalling than adding GPUs.

I'm not so sure. If we look at "stock clocks", "street" prices & the products w/ best $/performance ratio then the best examples to compare are i7-920 ($284.00) & new 55nm gtx260 w/ 216 shaders ($254.00).

This i7-920 host has been running @ 5 weeks w/ stock clock 2.67 Also is runnng 3GBs ram, so appears to be in tri-channel mode.
http://setiathome.berkeley.edu/show_host_detail.php?hostid=4683881
RAC = 6600

Projected RAC (interpolated from speed & WU credits granted) of GTX260 should vary in 6K-8K range depending on 192 or 216 versions. The new 55n 260 mentioned above is actually running @ same as GTX280 in most recent benchmarks.

So...I'm compelled to think scaling is at the very least, similar if not favoring the GPU in this example at similar price-points. However, when one factors the option of a 2nd CPU, the price/performance ratio isn't very good at all. 2-slot boards are 50% to 100% more expensive than single-slot. Also, a person will need to populate 6 memory slots for tri-channel instead of 3? Add another $284.00 for 2nd i7 920.

Compare that to just adding a new 55nm GTX260 w/ 216 shaders for $254.00 (EVGA on Newegg today....not a rebate price).

The NV super mega high end card from NV will have to drop to 120$ to be competitive in term of RAC/$ ... i let you do the calculation.
So, mister NV, can we get those NV G92 for 120$?
remember , the way to calculate "how good the GPU is" force you to get a Core i7 in the price, otherwise, your GPU is not competitive... hehehehhe

I think my above examples are a more level playing field to compare but if we are to look at mega high-end as you mentioned, then we'd be comparing i7-965 Extreme @ 999.00 EACH & GTX295 @ 499.00 each. You'd have $2K alone, just into the 2 CPUs. I'd have @ $2500.00 (One 965 + 3 GTX295s) but would save on the single slot MB & needing only 1/2 of the DDR3 Ram. However, I'd probably have to spend another $150 on a high-output power supply. Still, the build might be within $200-300.00 of your 2 slot. BUT....I'd have 6 GTX-series GPU cores + the 4-core 965. Really can't see how a dual-slot i7 can compete against the RAC that a 1 CPU +3 card(6 GPU)could produce...if looking at "high end".

Lastly, from a software standpoint, several of the RAC discussions compare Core i7 running OPTIMIZED MB AK_v8 vs. the RAC of a GPU-add on running Berkeley-stock CUDA app. This is apples & oranges. Obviously, once the Cuda application is optimized closer to current MB AK_v8 level.....there will be considerable performance improvement as well. Even a 20-30% performance increase in the Cuda app would appear to solidify the performance/value proposition vs. an added CPU slot.

BTW...FWIW, I'm not a Cuda fanboy....I actually favor a different approach & API solution for future GPU apps ; > )

... Shameless marketing, speaking about the good, and only the good. ...

[...]

Bottom line: One more lie.

And Who is throwing the most Marketing?

And as for "Farting in the wind", that must be a very French term!

All very entertaining :-)


Please underwhelm us with your top-line RAC and then show the code that did it. (Nota Bene: We know the trick for gaining short term explosive RACs...)


Let the fun run!

Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

My Job is my hobby, i never have the feeling of working ... the down side, I never stop working :)

Amen to that! I enjoy and suffer exactly the same thing!

... Shameless marketing, speaking about the good, and only the good. ...

[...]

Bottom line: One more lie.

And Who is throwing the most Marketing?

And as for "Farting in the wind", that must be a very French term!

All very entertaining :-)


Please underwhelm us with your top-line RAC and then show the code that did it. (Nota Bene: We know the trick for gaining short term explosive RACs...)


Let the fun run!

Happy fast crunchin',
Martin

if i take the top 1 position, do you agree to stop posting for a year?

lol

i just got the new client for boinc and updated the drivers to use cuda.

what i'm seeing is that units run through considerably faster than before. my problem lies in that it seems to corrupt about 2/3 of the work units instead of doing them properly. it also seems to cause my video drivers to fail and recover every time it pooches a work unit which does nothing good for my temper.

Please look another threads (and maybe even better - threads on beta site - here too much noise) about current CUDA errors. There is a scripts (2 of them already) and modified app build that could help you diminish effects of bugs in current app version (we all wait more stable one soon).
Summary of current bugs:
driver crashing/freezing/overflows on VLARs (tasks with AR <0.1).
Crashing/overflows on VLARS with AR~0.13.
Overflows on some of VHARs (AR>~2,7).

I recommend to abort all VLARs with AR <0.1 and keep eye or abort too all other tasks from "group of risk". That way you will get much more stable and productive work with current CUDA MB versions. Hope new version will be more easy in use :)

i've discovered that cuda in it's current incarnation is much better behaved if i shut off boinc when actually using the machine myself and just let it run with the monitor shut off etc. this has dramatically dropped the error rate.

i appreciate the advice but given my rather marginal computer skills i think i'd better not be tempting fate.

It depend how much money you want to pay..

A complete new rig with Core i7.
Or an Intel Quad you have at home and with one new GTX 285 or 295.


OTOH.. the power consumption..
A new i7-rig.. low GPU.. maybe ~ 200 W ?

My old (but loved ;-) QX6700 @ stock 2.66 GHz ~ 200 W and one GTX 285.. ~ 250 W ?
With GTX 295.. ~ 300 W ?

Only crunching wattage!
Not high performance 3D gaming! ;-)

---

Hey guys just thought i would tell you something i am going to use seti's example of strength in numbers here so i started a cuda team calledCUDA Crunchers so we could see the result of Cuda devices and there credit. Check it here

http://setiathome.berkeley.edu/team_search.php?keywords=CUDA&country=&type=0&submit=Search

---

if i take the top 1 position, do you agree to stop posting for a year?

lol

I don't see how having access to prereleased HW and getting a large RAC overclocking said HW is special. It's not like your actually improving the optimized app like you said you were going too. msattler or someone else has pretty much the same setup that you have about 3 or 4 weeks later because it has launched.

It really would be about time for Intel to make a graphics processor that didn't completely suck. Once again not special considering VIA of all companies has had better graphics then Intel for YEARS.

Didn't you throw a fit and leave? I know your trying to hype new HW before it's out but couldn't we get a discount for having to read all of these ads or at least get some amount of benefit.

---

If you ask me now then it would be a no. But if you ask me later on when we doesn't need to babysit our machine ta notice that too many errors occured so that you need to reboot to get the gpu to go in its highest throttle.. then Yes..

Pair a q9450 with atleast a gtx260 core 216 beast and you would have really high rac later when other computers start to turn in your AP units too..
...
Regards Vyper

Here's, at a quick glance at the on the top list currently fastest Q9450, it runs at 3.5GHZ and has a GTX280 and is able to run w/o CUDA errors causing rampant following errors. The GPU gives it at least 2 (mabye 3) extra cores worth of performance.
Note the "EXTRA" cores. The way I see it's better with CPU+GPU than just CPU. And I can't for the life of me grasp the resoning behind saying that is slower than (intel) CPU power.

/Anton

... Shameless marketing, speaking about the good, and only the good. ...

[...]

Bottom line: One more lie.

And Who is throwing the most Marketing?

And as for "Farting in the wind", that must be a very French term!

All very entertaining :-)


Please underwhelm us with your top-line RAC and then show the code that did it. (Nota Bene: We know the trick for gaining short term explosive RACs...)


Let the fun run!

Happy fast crunchin',
Martin

if i take the top 1 position, do you agree to stop posting for a year?

lol

OMG... today is monday, your RAC 333. So what ? What about your promises?....Just bragging?

The problem here, as I see it, is that who? either does not (or does not want to know), how BOINC and the Seti CUDA app works at the moment.

BOINC only allows the project to use one application with the same name on a host computer. This means a host with the appropriate hardware and drivers can either use the CUDA app or the normal app but not both. Therefore as detailed in the CUDA FAQ

Q) Does SETI@home run GPU and CPU versions simultaneously?
No. If BOINC determines your CPU is capable of running the CUDA version, only the CUDA version of SETI@home will run. One copy will run on each GPU you have installed. If you want to keep your CPUs occupied at the same time, you can join another BOINC project.

The BOINC developers expect the capability to run CPU and GPU jobs simultaneously for the same application will be added in the next couple months.

This also means that you cannot run the MB CUDA app and use the remaining cpu cores to run the AP app without an app_info.xml file.

Therefore who? is probably correct that a i7 system using the opt MB app on all cores is going to have a larger Seti RAC than a host using the CUDA app. BUT not an overall BOINC RAC. And therefore at the moment who? is completely missing the point. And also has difficulty in knowing which orifice to use, the phase is p*ssing into the wind.

Hopefully in the near future the CUDA app will be sorted and the BOINC client, which we are told is work in progress, will allow CUDA plus the normal apps on all the other cores.

The problem here, as I see it, is that who? either does not (or does not want to know), how BOINC and the Seti CUDA app works at the moment.

BOINC only allows the project to use one application with the same name on a host computer. This means a host with the appropriate hardware and drivers can either use the CUDA app or the normal app but not both. Therefore as detailed in the CUDA FAQ

Q) Does SETI@home run GPU and CPU versions simultaneously?
No. If BOINC determines your CPU is capable of running the CUDA version, only the CUDA version of SETI@home will run. One copy will run on each GPU you have installed. If you want to keep your CPUs occupied at the same time, you can join another BOINC project.

The BOINC developers expect the capability to run CPU and GPU jobs simultaneously for the same application will be added in the next couple months.

This also means that you cannot run the MB CUDA app and use the remaining cpu cores to run the AP app without an app_info.xml file.

Therefore who? is probably correct that a i7 system using the opt MB app on all cores is going to have a larger Seti RAC than a host using the CUDA app. BUT not an overall BOINC RAC. And therefore at the moment who? is completely missing the point. And also has difficulty in knowing which orifice to use, the phase is p*ssing into the wind.

Hopefully in the near future the CUDA app will be sorted and the BOINC client, which we are told is work in progress, will allow CUDA plus the normal apps on all the other cores.

Actually, you can do MB CUDA and AP CPU at the same time, with or without an app_info.xml - it's MB CUDA and MB CPU which don't play together.

Otherwise, I agree with everything you've said. And I would add:

Another thing that Francois doesn't understand is how to choose an appropriate performance metric for benchmarking. RAC, as everyone except Francois knows, is a decaying average which takes at least two months to stabilise, and which is affected by every server glitch/AWOL wingman/AR variation in the task supply chain. The CUDA app has been running on Main for less than three weeks (though it feels much longer, reading all these posts): there is absolutely no way that RAC can be regarded as an appropriate tool for this sort of p*ssing contest. For once (and it's the only good thing I will say about the CUDA release, ever), the NVidia benchmarking footnote in their 17 December press release at least attempted to measure the right thing.

[...]

Let the fun run!

Happy fast crunchin',
Martin

if i take the top 1 position, do you agree to stop posting for a year?

lol

What on earth for? You offer nothing here but hype and a few giggles.

Just three of my posts and the Great Who? is wallowing?

Please demonstrate something more than merely your windward bluster.

(Cue Monty Python music for The Search for the Holy Grail! :-) )


Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

... Shameless marketing, speaking about the good, and only the good. ...

[...]

Bottom line: One more lie.

And Who is throwing the most Marketing?

And as for "Farting in the wind", that must be a very French term!

All very entertaining :-)


Please underwhelm us with your top-line RAC and then show the code that did it. (Nota Bene: We know the trick for gaining short term explosive RACs...)


Let the fun run!

Happy fast crunchin',
Martin

if i take the top 1 position, do you agree to stop posting for a year?

lol

OMG... today is monday, your RAC 333. So what ? What about your promises?....Just bragging?

it sound like you did not figure out the way the RAC is calculated yet. Based on a small increase, you can guess what the final RAC is going to be over 2 or 3 days ...
"Heureux, prospères sont les pauvres en esprit, car le royaume des cieux est à eux"

Vanitas vanitatum et omnia vanitas
Tullio

---

... Shameless marketing, speaking about the good, and only the good. ...

[...]

Bottom line: One more lie.

And Who is throwing the most Marketing?

And as for "Farting in the wind", that must be a very French term!

All very entertaining :-)


Please underwhelm us with your top-line RAC and then show the code that did it. (Nota Bene: We know the trick for gaining short term explosive RACs...)


Let the fun run!

Happy fast crunchin',
Martin

if i take the top 1 position, do you agree to stop posting for a year?

lol

OMG... today is monday, your RAC 333. So what ? What about your promises?....Just bragging?

it sound like you did not figure out the way the RAC is calculated yet. Based on a small increase, you can guess what the final RAC is going to be over 2 or 3 days ...
"Heureux, prospères sont les pauvres en esprit, car le royaume des cieux est à eux"

Yes, it will be not even closer to something about what worth to speak :)
I collected enough info about your claims and their realization now, enough for me.
Better will spend some time to analyze noise from outer space than noise from you ;)
good luck

Now Now, this is getting out of hand.

Please Cuda, OpenMP, Close to Metal, StreamProcessing, and Threading Building blocks aficionados please get real, and back to your jobs, providing me with an OpenCL SDK/API uniting these disparate platforms and ending the warfare. separate they are 'close to useless'.

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.

Now Now, this is getting out of hand.

Please Cuda, OpenMP, Close to Metal, StreamProcessing, and Threading Building blocks aficionados please get real, and back to your jobs, providing me with an OpenCL SDK/API uniting these disparate platforms

I very much agree.

Can we have all that in a common standard rather than descending into the usual proprietary quagmire of deliberate Marketing foisted brokenness?

and ending the warfare. separate they are 'close to useless'.

Vanity of vanities; all is vanity - Ecclesiastes 1:2

... for Who? ;-p


Enough of the Marketing hype and giggles.

Instead: "Quod erat faciendum".

Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Let me explain: Vanitas vanitatum is a (bad) Latin translation of the Hebrew words Hebel habalim. whose translation might be Vaporware.
Tullio

---