Hello!
I search over google for times and times, and cannot find useful info about that cards performance ( GT630 and GT640 rev 2 (gk208) and GTX650 ) in seti@home.
good to see what GFLOPS they show in astropulse and multibeam, to can be compare this.
the prices on that cards is 50, 70, and 90 eur respectively, it all works without additional power, and consumes (TDP) about 25,49, and 65 watts.
there be a very good to see, how it is performance so i can decide.
other BOINC projects performance too is welcome.
thanks

---

oh yes, and all three of this have a very similar GFLOPs rating:
for 630.640 and 650 respectively it is 692 , 803 and 812 GFLOPS (FMA).

as so, if their not have significant differences on SETI performance, i think, is the best to buy a cheapest and less energy consumped one. if not - there can be various decisions. also be a very sadly, there, in setiathome site i can not see possibility to search for coprocessor type for see, what in average results people have with them. possibility to get that type of statistics, i think, be a good thing for all of us, because "top gpu" is not very handful there...

---

Memory bandwidth of the GT640(GK208) vs GT630(GK208) is almost 3 times more. It may be significant for SETI@home.
It would be good we had a SETI@home benchmark list resource for quick comparison. In the past I tried to get details to make a list of best # of task to run on GPU at a time, but did not receive much input.

---

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

we can try again. i can ( after some time, when i get 3 - 5 wu calcilated)provide ap results of hd4350, 9400gt, asus 630gt (rev 2.0 with slightly underclocked memory ( 1600 mhz), and gf 210....

---

I can tell you that one of my hosts has a 630 in it, running 2 tasks at a time, no CPU processing. It crunches both Seti (resource share 110) and Einstein (30) and Boincstats says its average credit is 5,181. It does not do Astropulse.

---

David
Sitting on my butt while others boldly go,
Waiting for a message from a small furry creature from Alpha Centauri.

you do a investigation? two tasks at once on 630 is optimal? three is worse?
one too is worse performance than two?
why not astropulse?

---

http://setiathome.berkeley.edu/show_host_detail.php?hostid=7210225

there are core 2 duo e4500 computer with nvidia gt630 rev 2 ( gk208) videocard.
note, that is ASUS 1 gb version with underclocked memory ( 1600 mhz vs 1800 mhz in standart).
GT630-SL-1GD3-L
http://www.asus.com/Graphics_Cards/GT630SL1GD3L/

Asus GT630 with 2 Gb ram model have standart memory clock.

---

there is host with GTX650 (asus, 1 gb ram):
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7225057

on gt640 prices goes up. thinking - buy a asus gt630 2 gb asus ( 49 eur) for home desktop, or asus gt640 1 gtb ( 74 eur). more to gt630 - i think, difference between this two not very big, and on another side - it is sufficient for my needs. i think, not very wise buy a videocard in keeping in mind generally it speed for Boinc...

---

Here is another host with the GTX 650: http://setiathome.berkeley.edu/show_host_detail.php?hostid=7144422
Comparing different video cards seems to me not to be a trivial GFLOPs rating comparison nor a whole memory bandwidth matter at all.

---

IZ3ATV

microsoft sucks, as always? :D
yes, sadly, but boinc itself not contain any usable benchmark for correct comparision of setups ( cpu / gpu) for real crunching...
now i get ordered asus gt630 2 gb version. can be comparing with asus gt630 with 1 gb and underclocked memory ( 1600 vs 1800 mhz). both is rev 2.0 cards on gk208 and with best watt / gflops rating based on
http://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_600_Series

---

Boinc don´t have but the Lunatics team have, see at: http://lunatics.kwsn.net/index.php?module=Downloads;catd=5 there are some builds (AP/MB) that allow you to test the real performance of your CPU/GPU on the SETI world.

---

Comparing different video cards seems to me not to be a trivial GFLOPs rating comparison nor a whole memory bandwidth matter at all.

If you are talking about the GFLOPS ratings on his Application Details, the problem is it does not show total GFLOPS but GFLOPS per task. He may be running 3 tasks at a time but we have no way of knowing...

If he is running 3, his total would be 3x43.69 GFLOPS but if 2 then 2x43.69 GFLOPS [for SETI@home v7 (anonymous platform, NVIDIA GPU)].

Also the reason Memory Bandwidth is important is when more than one task is crunched per GPU (more Bandwidth helps for more tasks).

But yes, the problem remains. There is no single GPU specification (or even "outside" benchmark for that matter) that hints to S@H performance unfortunately :(

Boinc don´t have but the Lunatics team have, see at: http://lunatics.kwsn.net/index.php?module=Downloads;catd=5 there are some builds (AP/MB) that allow you to test the real performance of you CPU/GPU on the SETI world.

yes, but both of them for linux is 32bit app versions, who may be relevant, if you run it on 64bit system...
anyway i try to look on it, and maybe run on all my machines...

---

IIRC MB uses only 32bits apps, no 64 bits apps are avaiable, no matther if is in Linux or Windows. Seems like the tests made by the devs team shows: 64 bit version of MB crunchig is slower than the 32Bits (exactly why i don´t know), so they decide not to release them.

AP has some new 64 bits builds for CPU crunching (windows at least, not sure about linux) but they are in Alpha/Beta testing and not actualy avaiable to the general public yet. AFAIK they are faster than the 32 bit version.

---

Boinc don´t have but the Lunatics team have, see at: http://lunatics.kwsn.net/index.php?module=Downloads;catd=5 there are some builds (AP/MB) that allow you to test the real performance of you CPU/GPU on the SETI world.

yes, but both of them for linux is 32bit app versions, who may be relevant, if you run it on 64bit system...
anyway i try to look on it, and maybe run on all my machines...

The banchmark files are scripts that run apps.
Get the 64 bit linux apps from Lunatics linux downloads

petri33

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

IIRC MB uses only 32bits apps, no 64 bits apps are avaiable, no matther if is in Linux or Windows. Seems like the tests made by the devs team shows: 64 bit version of MB crunchig is slower than the 32Bits (exactly why i don´t know), so they decide not to release them.

Eh, hum, for Linux that is not true:

http://lunatics.kwsn.net/index.php?module=Downloads;catd=1

01 - Linux 64bit Cuda Multibeam (x41g), Dec 2011

02 - Linux 64bit OpenCL Multibeam v7 HD5 (r1844), May 2013

03 - Linux 64bit OpenCL Multibeam v7 (r1844), May 2013

04 - Linux 64bit Multibeam v7 for SSE2 CPUs (r1848), June 2013

05 - Linux 64bit Multibeam v7 for SSE3 CPUs (r1848), June 2013

Claggy

That´s why i use: IIRC ... i only try to point him where to find the builds, there he could find the one he needs.

Anyway is interesting to ask why exist a 64 bit MB version for Linux and the x41zc are not avaiable in 64 bits for Windows.

---

That´s why i use: IIRC ... i only try to point him where to find the builds, there he could find the one he needs.

Anyway is interesting to ask why exist a 64 bit MB version for Linux and the x41zc are not avaiable in 64 bits for Windows.

Thats because of different compilers.
They dont give any benefit for windows code (GPU) in 64 bit yet.
Might change in the future.

---

With each crime and every kindness we birth our future.

That´s why i use: IIRC ... i only try to point him where to find the builds, there he could find the one he needs.

Anyway is interesting to ask why exist a 64 bit MB version for Linux and the x41zc are not avaiable in 64 bits for Windows.

Thats because of different compilers.
They dont give any benefit for windows code (GPU) in 64 bit yet.
Might change in the future.

Very different story than with AMD/Intel 64 Bit CPUs.

As 64 bit addresses use twice as much register space as 32 bit ones, and GPUs use thousands of threads (each with many registers), 64 bit GPU code tends to be slower as well, chewing up precious gpu registers quickly. That can shift with driver latency being dominant at the moment (at least on Windows), and improved compiler technologies...

Unless future data like GBT AP or such require huge amounts of video memory (iteslf a possible sign of bad programming/optimisation) , 64 Bit Cuda on Windows, Current to earliest Cuda architecture, All Cuda revisions that support it, will always be slower with the same GPU code.

Linux gets x64 Cuda builds, solely because Cuda 32 bit building seems to be broken on x64 Linux (cross-compiling).

---

"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.

That´s why i use: IIRC ... i only try to point him where to find the builds, there he could find the one he needs.

Anyway is interesting to ask why exist a 64 bit MB version for Linux and the x41zc are not avaiable in 64 bits for Windows.

I think on Linux the majority of hosts are x86_64, with i686 in minority, So far Urs has supplied x86_64 apps, be it CPU or ATI,
For the Linux x41g port, the 32bit version didn't produce good results, so a 64bit x41g app is the only possibility for Linux Cuda crunching,
For windows, 64bit addressing slows the x41 Cuda app down, since we don't need 64bit addressing a 32bit Cuda app is what we have.

Edit: Jason beat me to it.

Claggy