Sounded to me like just a normal upgrade of the BOINC server code. GPU has been around for a while so I would think that the upgrade is needed to support that. Now it could be that it's needed to work with current levels of other software WCG may be using in their GPU implementation.

It's one of a slew of prerequisites... the v7 Server software requires more power [to support the oodles of additional memory required as well], so the hamsters had to be trained on endurance:

--//--

Read "so the hamsters had to..." as "and the hamsters have to...". And to get another fact straight, the big upgrade is on March 6.

--//--

The upgrade on Tuesday is being done so that we can run GPU applications. The new server software that we will use is up to date with recent code from Berkeley. The older code that we have currently running doesn't have what is needed to send out application versions for GPUs. Therefore, we are updating and will be using updated server software.

However, this update will have some disruptions when we deploy it.

One of the biggest disruptions will be the new way that the server code handles identifying 'trusted' computers (ones that can process non-redundant computations without a wingman) and 'reliable' computers (ones that process resent workunits quickly). The determination of 'trusted' and 'reliable' will be changed from being determined for a specific device to being determined for a specific pairing of a device and version of a research application. This means that your computer could be trusted on FightAIDS@Home but perhaps not on Computing for Clean Water (for example). Since initially we will not have data at the device-application version level, no-one will be trusted or reliable until the data is accumulated to make that determination. This new feature is a great enhancement to the system but it will cause a short term disruption.

Another disruption is that we will start using the new credit system. The aim of the new system is to better measure the efficiency of different computers and different application versions so that computers are assigned credit in a more consistent way - even across different operating systems or cpu vs gpu. I'm hopeful it will work well. However, this is one area where our development resources just doesn't reflect the diversity and scale of the thousands of volunteer systems that you all provide. As a result, we expect some surprises following the upgrade. Our plan is to monitor what happens and work through issues as they are found.

The good news is that we have confidence that the research will continue to be processed correctly during the change.

As a preview of the performance of the GPU version, one machine that we have been testing on is a Win7 machine with 6GB RAM, a AMD Phenom II processor oc'd to 3.9GHz and an ATI/AMD 6870 graphics card. On this computer the cpu version runs in about 70 minutes while it runs in about 6 minutes on the gpu version.

The first step in moving towards the beta is to get the server code updated in production and ensure that it is stable. Then we will be ready to look at starting the beta testing for gpu.