RC5-72 and OGR

**Carld** · February 15th, 2012, 04:30 PM

I noticed there's still a TechIMO team.. so count on some points coming this way.

**Carld** · February 24th, 2012, 05:34 AM

I have been running both clients. One is using Stream with my ATI HD 4850, while I have OGR running on the i7 920. One core though needs to be dedicated on the Stream client.

I also run the OGR client on my i7 3930K machine while Folding@home CUDA using my GTX 580.

**Carld** · April 26th, 2012, 06:28 PM

I have been running the 64 bit OGR clients. for about 80k ppd. Even with folding on the GPU the i7 3939K doesn't lose too much output.

I also just got another HD 4850 for $50 bucks to CrossfireX and get some more RC5-72 ppd. The HD 4850 is in the Windows 8 beta box in the screen shot. The DFI supports CrossfireX, the old Abit doesn't. So both HD 4850s with their 800 stream cores each with have to go in the DFI.

**ShyguyXPC** · April 26th, 2012, 06:35 PM

Does it actually require SLI or Crossfire to run, reason I ask is F@H doesn't use either, but just runs the cards independent of each other... does OGR/RC5 actually use that, or does it run it independent of each other?

**Carld** · April 26th, 2012, 06:39 PM

oh I don't know. Probably doesn't. Then there's that 16x thing that I'm not sure makes a difference, like if one PCIe is 8x and the other is 16x.

I'm sure the 400w psu the Abit is using will die any minute, and if I add the other 4850 in it I'm not even sure it will see both cards, let alone not have the psu blow up.

**ShyguyXPC** · April 26th, 2012, 07:27 PM

usually with Folding and some of these other programs, the x1, x4, x8, x16 thing doesn't matter much, as its not Gaming data thats streaming 10's of GB's in data throughput as fast as possible to output to a screen.

I know with F@H, you can run an x16 card on a x1 or x4 slot with little to no impact on PPD rating (using an adapter), but not sure on Dnet's programs.

one of the reasons I've been hunting for a GOOD nvidia based (GT/GTS/GTX 400/500 or better), or newer Radeon based (HD 5500/5600/5700/6500/6600/6700/7700 series) half Height 6 or 7 inch long single slot card for my Mini ITX system, For cheap, since it only has an x1 slot, and using a low profile x1 to x16 adapter riser card. Have also been on the look out for a Socket 775 Mini ITX board with an x16 slot in it, which I'd rather do, since I can find more cards this way, full height single slot card.

**Carld** · May 6th, 2012, 08:44 AM

I just tried out the second HD 4850. It didn't seem supported in 3dmark06, even though it still scored 3000 points higher than if there is just one HD 4850 in the system.

With RC5-72 both HD 4850s combined as 2 of 2 crunchers pushing 700,000,000 keys/per sec. The only drawback is both cards need one dedicated cpu core each, and that means one less core to send OGR-27 output. I may just buy a nice PSU for the AMD 4400+ box, and just have both ATI cards in that instead of the DFI.

**Carld** · May 9th, 2012, 05:19 PM

ok here's two HD 4850's running it in crossfirex or just on the board. I don't see a keyrate difference with crossfirex on or off.

2 CPU cores have to be open for use, like from the OGR client (setting affinity in the task manager).

**ShyguyXPC** · May 9th, 2012, 05:50 PM

Originally Posted by Carld

I just tried out the second HD 4850. It didn't seem supported in 3dmark06, even though it still scored 3000 points higher than if there is just one HD 4850 in the system.

With RC5-72 both HD 4850s combined as 2 of 2 crunchers pushing 700,000,000 keys/per sec. The only drawback is both cards need one dedicated cpu core each, and that means one less core to send OGR-27 output. I may just buy a nice PSU for the AMD 4400+ box, and just have both ATI cards in that instead of the DFI.

Which CPU is in the 4850's rig now? wondering if you have an Intel HT capable CPU, if running a single core with HT would fulfill the dedicated core for both GPU's? This way you might only need a single core, and can reallocate the 2nd core for OGR.

**ShyguyXPC** · May 9th, 2012, 05:52 PM

Nevermind I see in the pic its an i7 920, so you might be able to do as I mentioned above, though no idea, but worth a try if you haven't already.

**Carld** · May 9th, 2012, 06:55 PM

I set the affinity on the OGR client to core 1 and 2 unchecked with the other 6 still checked.

This seems ok. I'll see how the electric bill is the coming month.

**Carld** · May 23rd, 2012, 12:14 AM

hmm, I also found that using slightly older drivers, like the 10.4's over the new 12.4 to increase the output. The heat is also increased though

**Carld** · May 23rd, 2012, 02:03 AM

seems the drivers up to 11.10 for the HD 4850's anyway use two dedicated cores.. which is why the output is higher.

The 11.12 and up only do a output (2x HD4850's) of 700,000,000 keys/sec, when the 11.10 and older do almost 1,000,000,000 keys/sec.

The temp of the cards are 15C hotter though with the older drivers.. It's seeming the 800 cores aren't all being used with the newest drivers for some odd reason.

**ShyguyXPC** · May 23rd, 2012, 02:06 AM

that is odd, using all cores with newest drivers, but output total is less.

unless they messed up on their GPGPU API they use, I forget which it is, I know its not CUDA, but, ATI/AMD has used several over the years, not sure if they use OpenCL now or not?

**Carld** · May 23rd, 2012, 05:38 PM

It might even be the client also, since it is older than the cpu clients, well the beta clients.

otherwise... fans need to be 70% for them to stay at 70C. The coolers are the OEM, flat single slot copper heatsinks. I maybe buy some better ones.

**Carld** · July 3rd, 2012, 04:40 PM

The GTX 680 with the CUDA 3 client isn't all that great.. It probably just needs to support the card.

500 million is ok, just not what is hoped for with the 1536 cores. It's like the 1000 cores are "off limits".

**Carld** · July 3rd, 2012, 04:43 PM

comparing to the 2 HD 4850's; If all I was doing was crunching.. the 2 or maybe a 3rd HD 4850 would be well worth the cost (especially if they are going for $50 a pc). 4 in Xfire... would still be a total of 200 bucks.. and that would be about a 2G key/sec output.

**ShyguyXPC** · July 4th, 2012, 12:10 AM

Sell you my 4870 for $50? 1GB, fan can get a bit loud, but it still works fine, not sure on the temps, I know it did run warm, and ran hot in some games, I want to say low to mid 80's at least, I think I did see it hit 90C in a couple games.

can crossfire it with 4850's.

Only difference between the 4850 and 4870 was the Clocks and I think the latter used GDDR5.

Video Card Comparison - GPUReview.com

but with the GTX 600's, I have read a few times with F@H that they seem to not be working as well as they should and that the Radeon HD 7000's are actually better right now. Maybe like F@H, maybe Dnet's client needs to be tweaked and optimized better for the new GTX 600's.

**Carld** · July 4th, 2012, 12:35 AM

I'm thinking the same.. as for no support on the new architecture.

The 680 is quiet. Even with the fan at 70% it's still unnoticeable.

Folding is working. The beta.. work unit seems to be a 5 hour long one though. I had to have the gpu on "0" not the default "-1" to get it running.

PPD is just 21K assuming I run the card all day. The cpu alone, even on smp 10 is doing 40K. (these are the client estimates).

**ShyguyXPC** · July 4th, 2012, 01:10 AM

only 21K for the 680?

Dang, thats not that far away from the 14K that my 470 can do. When you consider the fact the 680 has more than 3 times as many cores, and higher clock speeds, and a more powerful/efficient Core architecture, it should be capable of much higher numbers IMO.