I have had this card running 24/7 for a few days now. it as of right now is up to 229RAC, im hoping the GPU can out do my Atom 330 box i had a while ago. The Atom box topped out little over 300RAC, would be nice if this GPU at 25watts could match it at least.

figured some might be interested to watch the number grow and/or even out. not sure how much it would be worth investing if for seti. priced from egg is $59 for the 1gb and 2gb cards, is a single slot but the heatsink is passive so it takes up two slots.
http://www.newegg.com/Product/Product.aspx?Item=N82E16814121778

other thoughts...
does pcie lane speed effect crunching?
Reason being on the lane speed, GPUZ reports the card running at x1, when its suppose to be at x8. the slot is a 16x slot and the i3 370, has pcie 2.0 spec at x16 for the slot. cant seem to figure out why/how the card is stuck at x1, it has now power adapter, nor should it for 25watts.
does ram speed effect crunching?
even though the ram is clocked at a lowly 1800mhz with gDDR3, would the faster gDDR5 really be worth it? i mean the same GK208 used on the GT640 using gDDR5 uses 49watts. the gpu speed is the same as the GT630 GK208, i cant see it making a huge difference to be worth the near $100 and 49watts.

hope this works
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7289909

If you going to spend $100 dollars for the 640 why not spend an extra 20 and get a gtx 750. Gigabyte and MSI have one for 119.99, they use PCI-e for their power and run much faster and are relative cool. They are double slot. I put one in my optiplex 4 core and it jump up to about 15 thousand RAC. Just a thought.

Zalster

the last portion of my post was just a thought, i never really though of getting one at that price tbh. i only got the GT630 gk208 for the updated VP engine for my pc it is in now. that at some point once i get other things finished up will be my HTPC.

i always did wonder, why is it that seti doesnt have some kind of benchmarking program. say like a self contained .exe for windows, that uses the most cpu/gpu demanding WU. then tells you how long it took to complete the WU, while not ideal for RW numbers. it would be nice to have some kind of set benchmark for compare other cpu/cards too.

From the data so far, I expect that GT 630 to attain a RAC of over 1500 once evrything settles out. That's based on CUDA42 times of about 1.5 hours for tasks which give about 95 credits. When BOINC tries some CUDA50, that is likely to be slightly better.

CUDA32 seems to be taking about 45 minutes more on average, the BOINC server code will gradually adapt and not send many of those. But I noticed a couple tasks took around ten hours. If that was just because you were actively using the system at the same time, it doesn't indicate any problem, If not, it may mean the passive cooling isn't getting enough airflow.

If the cooling is good enough, running 2 tasks at a time on the card would almost certainly make it more productive, and it has enough memory to try even more than that. OTOH, if its basic function is HTPC use then pushing for crunching productivity probably doesn't make sense.

                                                                   Joe

i havent touched the computer since i got Seti and R@H running at the same time. checking GPUZ with GPU load shows anywhere from 75-88% load, using .5sec intervals. if letting gpuz run in the background might be a causing a problem then i can close it. otherwise right now for testing the run of both R@H and seti on that pc, i dont use it.

so is the PCIE link speed a non issue?

should note, right now its open case for the pc. im still trying to figure out which case to go with, im looking for something to fit in. right now its some cheap apex case that newegg had, was originally housing a Atom 330 setup till i sold the board.

cant edit my post, odd to me. i checked out my mini box, dont know where to look for txt's for errors or anything. i saw gpuz showing no load on the gpu so i reloaded bonic and things started working again. i did add a fan near the gpu hs but looking over temps during the day from the past, highest i saw was 48c. i didnt log temps so i dunno if it got higher then that, guess its possible.

after letting one of the WUs run for a bit, then starting GPUZ backup. now it shows near 96-98% gpu load and memory controller pegged at 100%. starting to think a bit of extra memory speed will help.

on another note, the other day seti was freaking out or something. as gpu load was showing 0% in gpuz, restarted the computer for just in case. link speed upped to x8 instead of x1. i also just on the safe side added a fan on the hs of the gpu, temps now dont go above 35c depending on amb temp in the house. lowest full load temp i saw was 30c at night.

rac right now is 335, which is nice for the price and power used imo vs the old Atom 330 setup.

so is the PCIE link speed a non issue?

Not a significant one. The way SETI works is it sends a small block up data to the CUDA card, then lets it process it internally. It's constantly feeding in new calcs, and receiving the results, but this doesn't saturate the PCI bus. The actual data transfer might only be 1% of the total work unit time? Even if that takes 2X as long to complete, that only increases your total time by 1%, which is a non-issue really.

rac right now is 335, which is nice for the price and power used imo vs the old Atom 330 setup.

Doing the maths like Joseph did before I get ~2hours for ~100 credit WU. Do 12 of those a day and that's a RAC of ~1200. Maybe a bit more. Give it a few weeks of running and you should be closer to that.

Also you should be getting something useful from the I3 CPU and get another 1,000 there?

Ian

well the cpu is running R@H which is cpu only, as i would have been R@H on the gpu. not sure why they havent gone the gpu route like seti or F@H. one thing i should look into is running 2 of the wu but not sure how. though after that one restart the gpu is now around 90% pegged most of the time. im not sure how much more useful or efficient it would be to run two on that card.

now my GTX 660 i started the GPU wu's on, thats another matter. has a load of cuda 50's and chew them up with 80% usage on the gpu.

also my google foo is failing me, is there a current list of points per gpu? looking to see if moving to a maxwell would be worth it due to the way they changed the arch layout vs kepler.

i find my self having to baby sit the GPU client for seti. i dont know how where to explain whats going on, just leaving GPUZ open in the back ground to check temps and gpu load. i thought i would check it today, which showed a rather odd 0% load then jump up to 80% or so then show 0% for a while. this might explain some of the 10hr wu's, i though it might be gpuz issue. i restarted that, same thing going, i snoozed the GPU client for a minute, woke it back up. bing things are humming along agian, this isnt something i expected. if i have to tell bionic to work for 23hrs and 55mins or so then allow to work agian. thats a bit much but a workable work-around to the issue.

while i am off/on running the GPU client on my GTX660 on my other rig. it hasnt run long periods of time like the GT630 card. i dont know if im the only one with the issue or if there are others.

also my google foo is failing me, is there a current list of points per gpu?

I am not familiar with "points per gpu", but if you are looking for relative power, try this site: http://www.videocardbenchmark.net.

[EDIT]

Also, the ASUS site (http://www.asus.com/Graphics_Cards/GT630SL2GD3L/specifications/) specs your card at "up to 75 watts", not 25.

---

then i guess nvidia is wrong?
http://www.geforce.com/hardware/desktop-gpus/geforce-gt-630/specifications

Remember that these cards were aimed at gamers not for crunching SETI@Home. Why do you think new cards come online like popcorn? It aint for crunching Seti@ home. Its for bitcoiners, And gamers.

---

[/quote]

Old James

Three different chips called the GT 630.

If yours is the new Kepler based one, then 25w seems correct. The earlier models used more power, and had a heap less CUDA cores as well.

Ian

Nice, just the info I was looking for after installing a GT630 in my new cruncher.

However, this is what often confused me with regards to gpu's.

The Nvidia link that Evilsizer provided is the one I use to reference their cards. It does say that add-on card manufacturers can vary to Nvidia's official specs.

The gpu I installed was Gigabyte Geoforce GT630-2GI

Comparing the specs on both sites, it seems that I have the Kepler card. However, from the info on the box which stated 1600mhz & 128 bit, that would make it the middle card of the three.

However, GPU-Z says different...


What can this card crunch effectively?

Edit: -

On going back to the Gigabyte link, I noticed that they had related products at the bottom of the page.

The GT630-2GI(Rev 2.0) is 1600mhz so assume that Rev 1.0 is the 1400mhz, so I can safely assume that Rev 3.0 will be the 1800mhz Kepler.

The sticker on the packaging of mine states: -

GV-N630-2GI (Rev 3.0)

Hmmn, can't trust anybody these days... :-(

---

then i guess nvidia is wrong?
http://www.geforce.com/hardware/desktop-gpus/geforce-gt-630/specifications

I wouldn't say that. I would say it is the ASUS site that is wrong (possibly misread the '2' for a '7' when posting the power). You didn't say which brand of GT 630 you had, and I just looked at ASUS for the specs.

If I twisted your tail, I sincerely apologize.

---

Remember that these cards were aimed at gamers not for crunching SETI@Home. Why do you think new cards come online like popcorn? It aint for crunching Seti@ home. Its for bitcoiners, And gamers.

well the card im using im more inclined to think it is more of a HTPC card then anything with a 25watt power usage and DDR3 ram.

Three different chips called the GT 630.

If yours is the new Kepler based one, then 25w seems correct. The earlier models used more power, and had a heap less CUDA cores as well.

Ian

right as the title states it is the GK208 GPU being the kepler GT630. i referenced the wiki, where it is easier to sort though different releases. i looked for the lowest retail TDP with the newest VP engine from NV. since my main goal was for HTPC but then wondered what it might do RAC wise for seti. it has already passed the highest RAC that my Atom 330 box from long time ago would put out.

Nice, just the info I was looking for after installing a GT630 in my new cruncher.

However, this is what often confused me with regards to gpu's.

The Nvidia link that Evilsizer provided is the one I use to reference their cards. It does say that add-on card manufacturers can vary to Nvidia's official specs.

The gpu I installed was Gigabyte Geoforce GT630-2GI

Comparing the specs on both sites, it seems that I have the Kepler card. However, from the info on the box which stated 1600mhz & 128 bit, that would make it the middle card of the three.

However, GPU-Z says different...


What can this card crunch effectively?

Edit: -

On going back to the Gigabyte link, I noticed that they had related products at the bottom of the page.

The GT630-2GI(Rev 2.0) is 1600mhz so assume that Rev 1.0 is the 1400mhz, so I can safely assume that Rev 3.0 will be the 1800mhz Kepler.

The sticker on the packaging of mine states: -

GV-N630-2GI (Rev 3.0)

Hmmn, can't trust anybody these days... :-(

the kepler based cards only sport 64bit wide memeory width so your would be the older gpu. also the easier way to tell which arch it is ie kepler or fermi is the number of cuda cores. yours sports 96cores vw the updated GT630 being 384 cores.

then i guess nvidia is wrong?
http://www.geforce.com/hardware/desktop-gpus/geforce-gt-630/specifications

I wouldn't say that. I would say it is the ASUS site that is wrong (possibly misread the '2' for a '7' when posting the power). You didn't say which brand of GT 630 you had, and I just looked at ASUS for the specs.

If I twisted your tail, I sincerely apologize.

no you didnt, but i feel like some info i first posted to cover which card may have been over read. i think asus put up to 75watts to cover them self, since the spec match up to nv specs for the 25watts.


***
i have always been interested in finding cards/cpus that can provide the most work per power used. if for the 25watts of the new kepler GT630 does more work per watt then other cards would it not be worth it?

i have been out of the computer game loop for a bit other things in life happening. im a bit behind and so the performance of the kepler GT630/640 maybe known to some. i how ever did not find any info as far as max rac for them. i wanted to provide some info for others out there thinking of/or looking at the kepler GT630.

Sirius B wrote:

...
The gpu I installed was Gigabyte Geoforce GT630-2GI

Comparing the specs on both sites, it seems that I have the Kepler card. However, from the info on the box which stated 1600mhz & 128 bit, that would make it the middle card of the three.

However, GPU-Z says different...
...

That shows a GF108 chip so you have a Fermi 630.

                                                                   Joe

This is the new Kepler series GT 630

GV-N630D3-1GI
http://www.gigabyte.co.nz/products/product-page.aspx?pid=4831#ov

I've just installed one, running standard speeds with optimised apps.
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7282947

Seems to be taking about an hour per WU, which isn't too bad for a 25watt card. I see they are selling for about $50 US. What's not to like really ? It's 384, cores, although you sacrifice some performance with only 64bit memory and 8X PCI. But whadda you want for $50.

Ian

It's 384, cores, although you sacrifice some performance with only 64bit memory and 8X PCI.

Several others and myself proved a long time ago that using a 16x, 8x or 4x PCI-e slot makes no difference at all to crunching times has each has more than enough bandwidth not to be a bottleneck. ;-)

Cheers.