Completed P6701 - no bonus?

[Nathan_P]

Well, these *are* getting bonus points now. Good thing, since they seem to be all that I am drawing.

I have an X6 at 3840, getting about 9:25 TPF on the 6701's (running -smp 5).

Have you got any GPU's running in there? my x6 @3.4 is showing 10minutes a frame and i'm wondering if the gpu's are getting in the way

 

[zero2dash]

So I finally realized what this thread was about and why my PPD are in the gutter...these 6701's absolutely suck. I've got 3 machines (all my SMP2 clients) chugging these things -
i7-2 @ 3.6 :: 07:42 TPF :: 11,609.5k PPD :: 6,228 credit :: no gpu clients
q9450 @ 3.6 :: 11:14 TPF :: 6,639.5k PPD :: 5,177 credit :: 1 gpu client - 8800GT stock clocks
e8400 @ 3.0 :: 31:06 TPF :: 1,447k PPD :: 3,125 credit :: no gpu clients

PG loves bogging down our machines lately... between these things and 2684's

My lone dual usually nets around 2-3k PPD depending on WU's, man this 6701 is just eating the C2D alive.
i7-2 usually nets around 16-17k PPD, q9450 is around 9k IIRC.

 

[APOLLO]

For those people running GPU clients alongside your SMP, you may want to reduce the number of cores your SMP clients are using by one and isolate a core for the GPU client(s). I consistently see higher PPD doing this. In the case of quads, I use the -smp 3 flag, and with my dual quads I use the -smp 7 flag. This might not work on i7 machines because of the HT architecture, though.

 

[Nathan_P]

For those people running GPU clients alongside your SMP, you may want to reduce the number of cores your SMP clients are using by one and isolate a core for the GPU client(s). I consistently see higher PPD doing this. In the case of quads, I use the -smp 3 flag, and with my dual quads I use the -smp 7 flag. This might not work on i7 machines because of the HT architecture, though.

Thanks for that, i will experiment at the weekend and see how I get on

 

[Mr. Pedantic]

For those people running GPU clients alongside your SMP, you may want to reduce the number of cores your SMP clients are using by one and isolate a core for the GPU client(s). I consistently see higher PPD doing this. In the case of quads, I use the -smp 3 flag, and with my dual quads I use the -smp 7 flag. This might not work on i7 machines because of the HT architecture, though.

It does. I don't know what Stanford did with threading the client, but I don't think they did it very well at all. How on earth could an -smp 7 flag run faster than a -smp 8 flag with half a core taken to something else?

 

[APOLLO]

It does. I don't know what Stanford did with threading the client, but I don't think they did it very well at all. How on earth could an -smp 7 flag run faster than a -smp 8 flag with half a core taken to something else?

OK, thanks for the confirmation, I wasn't too sure about i7. I'm surprised this isn't more widely known and implemented then. I have reduced the core allocation by one core for my SMP2 clients since they were introduced under EA's recommendation. It still irks me that I cannot utilize all the cores in a given system, but I digress.

 

[ccityinstaller]

Since I'm running an X6 (-smp 6) and an 8800GT, should I switch to (-smp 5)? I was under the impression the CUDA cards didn't require any (I use 1-2%) core time?

On another note, I have finished roughly 4 of these now..just checked and I have a P 6017 sitting in queue so it's back to 15K PPD soon

 

[APOLLO]

Since I'm running an X6 (-smp 6) and an 8800GT, should I switch to (-smp 5)?

To be honest Thuban is a very different architecture from Intel processors, you can try it and let us know the results. It doesn't matter what WU you're folding, it has essentially the same effect on all SMP WUs from my observations. But, you need to use the spare core only for the GPU clients and the GPU clients cannot touch the other cores for this to make any difference, so a priority/affinity utility app will be necessary because windows won't accomplish this. Without an appropriate app, the results are worse than having all cores enabled by the flags.

I was under the impression the CUDA cards didn't require any (I use 1-2%) core time?

On paper that's true, but in practice I have never seen this (regardless of OS). Some GPU WUs like the 353s really consume a lot of CPU power whatever theory dictates or Task Manager reports. As soon as I get one of these, my SMP TPF skyrockets. Some GPU WUs are more efficient in CPU tapping while others are in between. It varies a lot.

 

[tmossman01]

Reply to Nathan_P....

Have you got any GPU's running in there? my x6 @3.4 is showing 10minutes a frame and i'm wondering if the gpu's are getting in the way

Yes, I have a GPU client running (an ATI HD5770). I know with it that if you do not tweak the environment variables, that -smp 6 does not make any progress...(I was doing this at first, and had zero frames after 8 hours (overnight).) Needless to say, I was not very pleased.

I have the 5770 tweaked well for the 511 point WU's, but have not found a sweet spot for the 384's.
(BROOK_YIELD=2, CAL_NO_FLUSH=1, CAL_PRE_FLUSH=1, FLUSH_INTERVAL=320)

I have debated going smp only, but work from home a couple days a week, so going all in on SMP does not seem to be profitable, especially with the server issues this past weekend.

 

[tjmagneto]

I just swapped motherboards so I decided to try a few Win A3s for kicks and got a 6701 right off the bat. With the 980x running at the stock speed of 3.33 ghz I'm getting a TPF of 6 min 45 sec. This is with 12 cores, 2 GPUs and no smp -11 flag. I'm going to get this OC'd back to normal pretty soon so we''ll see how low I can go.