Today I have a strange mail problem - maybe it's some campus upgrade gone wrong. Anyway, what I was going to mention is that we're in our final 1-1.5 weeks of work and that means: switch to panic mode. I expect progress and problem reports on this blog from all of you almost every day! I'd much prefer the former :-)
The St George campus will be closed from Thu so we'll meet on Fri at UTSC,
and on Sat. at my place. We'll of course spend some time with some of you who haven't been to Buttonville, within 50 km radius from there >->--
The meeting is 29 June, 2pm in AA bldg rm 401 or something like that (Vce Principal Research room on 4th floor). Those who can attend, say so!
There are many things which don;t work right now.
cudak4/5 don't have:
1. fixed IP addresses,
2. a working set of driver + SDK which sees 3 fermi cards properly (I installed driver 195.36.31 which was released days ago, but that doesn't work with the more experimental 256.25 (or some such)
that we had before. I didn;t have time to solve that thing and see if all cards are usable with the new recommended driver.
3. device 2 on cudak5 overheated, was 78C at idle, instead of normal 56C.
The reason was that it was in the high=gear state all the time (max clock speed on gpu and mem).
After driver installation, toolkit failed to install so I don;t yet know if that problem went away.
nvidia-settings showed that 1 of of 3 devices (0,1,2) has a different version of
VBIOS. however, it was dev 1 not 2, so it's false hint.
4. even though most SDK demos ran correctly on dev 0 and 1, there was a disturbingly long startup time, as opposed to instantaneous load on old cuda setups. dev 2 wass also hanging for a looong time before generating the Cuda Mem Alloc Error or Unspecified Launch Error.
this is some kind of software/driver problem. Robert swapped 2 cards and
dev 2 was always having problems (confusingly, dev 1 has problems on cudak4).
Any help in solving these fascinating problems will be appreciated.
Maybe I'll go to UTSC today..
keep up the good work