I believe he is referring to the use of GPU's.

I really do hope that there will be more GPU projects in the pipeline.

I got some new WUs (not resends) created 05/04/2013. Huh?

Well, did notice the enforced back-off after a work request had been upped from 11 to 31 seconds a week or more ago, but after installing a brand new Linux box and first work request>

Sat 04 May 2013 04:28:47 PM UTC | World Community Grid | Requesting new tasks for CPU

got 1

exercized a little patience, became impatient and hit update to be answered with

Sat 04 May 2013 04:30:48 PM UTC | World Community Grid | Requesting new tasks for CPU
Sat 04 May 2013 04:30:52 PM UTC | World Community Grid | Scheduler request completed: got 0 new tasks
Sat 04 May 2013 04:30:52 PM UTC | World Community Grid | Not sending work - last request too recent: 119 sec

On my watch that is 119 seconds plus X which appeared to be, when really really being patient, about 10 minutes.

Sat 04 May 2013 04:37:57 PM UTC | World Community Grid | Sending scheduler request: To fetch work.
Sat 04 May 2013 04:37:57 PM UTC | World Community Grid | Requesting new tasks for CPU
Sat 04 May 2013 04:38:00 PM UTC | World Community Grid | Scheduler request completed: got 3 new tasks

Lots of other projects can be visited in nearly 10 minutes to de-idle CPU cores or back-fill a buffer to the set brim.

(Do I understand why that long an enfiorced back-off at this phase of GPU project conclusion... take a guess, which is what we WCG members seem to perpetuate in having to do, absent volunteered information)

Now going to take a walk, before straying on the wild side

As long as enough GPU-work was available, even someone with a fast single-GPU managing on average 1 result/minute would fill-up to his low-cache-setting, and get some more to stop work-request for a time. The largest pause for someone making 1 result/minute would be 64 minutes, since client will start reporting if more results than this (assuming a resent v7-client). So, except for users with 2 or more GPU's, and anyone with so large cache-setting was hitting the 7500-task-limit, even the fastest computers would do only 1 or 2 scheduler-request/hour.

With lack of enough GPU-work on the other hand, as long as these computers isn't out of GPU-work, they'll ask for more work but don't get any. While this will trigger a GPU-backoff, all these resource-type backoffs is reset every time a task is finished. Meaning, a computer finishing 1 result/minute, is also likely to do one scheduler-request per minute.

This is upto 60x higher load on the scheduling-server from these GPU-crunchers, so increasing the backoff to 2 minutes immediately caps the GPU-crunchers to only 30 scheduler-request/hour. For a multi-GPU-rig doing 120 scheduler-request each hour, even 30 is a marked decrease.

As more and more GPU-crunchers runs out of work, they'll make scheduler-requests less frequently.