600 Point Gromacs!

wooglin · Post by **wooglin** » Thu Jan 20, 2005 3:46 pm

herosformula wrote:I built a dual Opteron system after the holidays, all it appears to get are these 600 point units.

After 6 or 7 straight 600 pointers, I logged on to my desktop at home tonight to find it whittling away at a crap 63 pointer.

Now I feel dirty.

unregistered · Post by **unregistered** » Fri Jan 21, 2005 1:04 am

These 600 pointers are THE test for your system. More so than Prime95 and memtest.

My 2600 mobile barton is experiencing some early WU endings. I am not running hot, 49-50 tops, is a little warm but well within AMD specs. My 2800 barton is 100% successful completion rate.

I seem to break past 100 of 250 and then it ends. I have tried different clock and bus speed combos and even tried slight underclocking. It takes days to find out results because of the size of these.

Anybody with a mobile barton completing these? If so what are your clock/bus speeds.

I am now thinking a "faulty" stick of ram is the culprit. (1)512 (2)256 2700 Kingston sticks. I am going to start pulling/swapping them to see if I can narrow it down.

JanW · Post by **JanW** » Mon Mar 21, 2005 7:55 am

Looks like all 600 point projects are gone (for now). First the projects 1134 and 1135 stopped, and now 1140 and 1141 disappeard from the Stanford project summary page. I finished two of them yesterday... probably the last 1000+-point-day for me in a while

. Hope they'll be back soon.

EDIT: spelling

tay · Post by **tay** » Mon Mar 21, 2005 8:11 am

My P3-750 system got three of them while the Athlon box never did. Any ideas why that could be? I am simply launching the client in a dos box, becase I gave up on running as a service after getting f@h service stubs littered all over in my services list.

JanW · Post by **JanW** » Mon Mar 21, 2005 8:22 am

Do you use the same options on all your clients? I think you'll need both the "-advmethods" flag and explicitely allow big WUs when configuring your client (run the client with -config or -configonly flag). The client.cfg (that's what it's called under Linux, but I assume Windows is the same) should contain a line saying "bigpackets=yes" in the "[settings]" section. Note that, IIRC, at least on one platform (don't recall if it's Linux or Windows) it's not recommended to edit that file manually, as there is an issue with (Windows- or Linux-style ) line endings.

tay · Post by **tay** » Mon Mar 21, 2005 8:25 am

Blarg! The athlon is a windows box while the p3 is linux. I'll double check everything. Is a 343 point WU considered big, cuz the athlon got that? Thanks!

JanW · Post by **JanW** » Mon Mar 21, 2005 9:05 am

I don't know. It's not necessarily related to points, but rather to memory useage. Projects 1475 and 1476 (364 points) are "big", and the client I have currently running a 1476 uses 93MB. I forgot to mention a third condition to get big WUs: lots of memory. Your client reports the amount of memory to the assignment server, and big WUs have minimum memory requirements (I have a vague recollection that it's 384MB for the 600 pointers).

The easiest way to find out if a WU is considered "big" is to look at the PPD you get from it. "Big" WUs should give you roughly twice wrt an "ordinary" WU, as there is a 100% bonus for them.

Tibors · Post by **Tibors** » Mon Mar 21, 2005 10:11 am

You shouldn't need the "-advmethods" flag for BigWu's anymore, as they are no longer "late beta" but "mainstream" units. You currently only need the "-advmethods" flag for the QMD WU's I mentioned in another thread.