I'm #1!

Glad to have people helping out on all these other projects.

Congrats! Now that you have made this post however..... lets see if you can hold on to it ;)
 
Nice Work InvisiBill! What are you using to get 27K PPD? If I ever drop some $$ on a 5000 series ATI card i would like to run some dnet. :)
 
metallicafan said:
Nice Work InvisiBill! What are you using to get 27K PPD? If I ever drop some $$ on a 5000 series ATI card i would like to run some dnet. :)
Click to expand...

My main PC (i7 920 @ 167x20 with Turbo and HT on, 5870 main GPU, and 9800GT PhysX card) does the vast majority of it. I've got dnetc on the CPU paused while FahCore is running, so it's mostly doing F@H, but will kick in if there's a Folding dry spell (which happened to be the case when I first started). I also have an old Dell running just the STREAM client (no CPU client on that Prescott) on a passive 4350 that I got really cheap in a combo to have around as a spare. No farms here, just my PC and a second one to hold an extra GPU I had around. I started running the dnet proxy on the Dell too, so the main rig would spend less time updating and more time crunching.

My numbers bounce around a bit depending on how much gaming I do and if I'm having issues getting the ATI and Nvidia cards to get along. =) Looking back at my history, you can see where I upgraded to the 5870, where there was the client bug that caused only half of the block to get processed (resulting in double the blocks getting "completed" in the same time), where the GPU client betas expired for a few days at a time, and where I first started using the GPU client. It's crazy to look back at when it took several days just to complete one block.

ATI murders everything else in dnetc. The 4350 does about the same as the i7 (~40mkeys/sec). The 9800GT does ~115mkeys/sec. My GTX285 was ~315mkeys/sec. The 5870 does ~1800mkeys/sec. Yes, almost 6x what the 285 did. I asked the Fermi owners to run the benchmark, and the highest I've seen an overclocked 480 do is 736mkeys/sec. That's over twice what the 285 does, but the 5870 is still about 2.5x that. Even a stock 5770 will do over 900mkeys/sec. The biggest numbers (1,000,000+ blocks per day) were from a guy running the BOINC wrapper on 14 systems with multiple 5800's.

Hopefully a good ATI or OpenCL F@H client would reflect this too. I have no idea what the issue is - poor client programming, difference in the calculations being done by the two projects, etc. I really think it's possible that ATI could slaughter Nvidia in F@H though (or it might not, if the architecture just doesn't match up well with the calculations). While OpenCL wouldn't guarantee that the code is completely equal for both camps, it should at least avoid major disparities in the quality of the clients if they can both run the same app.
 
Nice numbers - you run OGR at all, or just RC5? The Commandos are thinking about doing and OGR run in the next few months... Of course, Kendrak's right - we may have to push RC5 for a month to give you some competition! (Not that my 8th place 60K pernt are much to write home about... :D)

Keep holding down the fort - guys like you are the reason the team can make group pushes without losing ground in other projects! One man vs. the world!
 
The GPU clients are RC5 only. I switched my CPU from RC5 to OGR a while back because it seemed to be more beneficial for the real world, but I switched it back to RC5 when I started using the GPU clients, to put all my points on one thing.
 
You must log in or register to reply here.
Back
Top